perm filename OUTGO.MSG[1,JMC]13 blob
sn#758248 filedate 1984-06-22 generic text, type C, neo UTF8
COMMENT ⊗ VALID 00118 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00011 00002 ∂01-Apr-84 1222 JMC
C00012 00003 ∂01-Apr-84 1331 JMC dictionary or word list
C00013 00004 ∂01-Apr-84 2032 JMC alleged corollary and also generic models
C00017 00005 ∂01-Apr-84 2047 JMC afore-mentioned proof
C00025 00006 ∂01-Apr-84 2112 JMC
C00026 00007 ∂02-Apr-84 0841 JMC corollary
C00027 00008 ∂02-Apr-84 0847 JMC your paper
C00028 00009 ∂02-Apr-84 0849 JMC
C00029 00010 ∂02-Apr-84 1543 JMC
C00030 00011 ∂02-Apr-84 1746 JMC
C00031 00012 ∂02-Apr-84 1747 JMC [John McCarthy <JMC@SU-AI.ARPA>:]
C00033 00013 ∂02-Apr-84 2222 JMC re: military funding of computer research (from SAIL's BBOARD)
C00034 00014 ∂03-Apr-84 0926 JMC
C00035 00015 ∂03-Apr-84 1014 JMC
C00036 00016 ∂03-Apr-84 1443 JMC
C00037 00017 ∂03-Apr-84 1608 JMC
C00038 00018 ∂03-Apr-84 1652 JMC
C00039 00019 ∂03-Apr-84 1724 JMC
C00040 00020 ∂03-Apr-84 2218 JMC
C00041 00021 ∂03-Apr-84 2223 JMC
C00042 00022 ∂03-Apr-84 2228 JMC SAIL accounts
C00043 00023 ∂03-Apr-84 2244 JMC re: UCLA, Ronald Austin (from SAIL's BBOARD)
C00045 00024 ∂03-Apr-84 2251 JMC
C00046 00025 ∂03-Apr-84 2254 JMC
C00047 00026 ∂03-Apr-84 2308 JMC
C00048 00027 ∂03-Apr-84 2321 JMC
C00049 00028 ∂04-Apr-84 0730 JMC secretary
C00050 00029 ∂04-Apr-84 0747 JMC
C00051 00030 ∂07-Apr-84 2121 JMC for David Etherington
C00052 00031 ∂07-Apr-84 2126 JMC seminar
C00053 00032 ∂08-Apr-84 1146 JMC S.F. OPERA SEASON HALF-SUBSCRIPTION FOR SALE
C00055 00033 ∂09-Apr-84 0011 JMC
C00056 00034 ∂09-Apr-84 1006 JMC message for John Cocke
C00057 00035 ∂09-Apr-84 1313 JMC paper
C00058 00036 ∂09-Apr-84 1709 JMC Is this a correct address?
C00059 00037 ∂09-Apr-84 1747 JMC
C00060 00038 ∂09-Apr-84 2024 JMC re: Undergraduates, and CS (from SAIL's BBOARD)
C00061 00039 ∂10-Apr-84 0133 JMC
C00062 00040 ∂10-Apr-84 0717 JMC
C00063 00041 ∂10-Apr-84 1631 JMC
C00064 00042 ∂10-Apr-84 1634 Mailer failed mail returned
C00065 00043 ∂10-Apr-84 1635 JMC
C00066 00044 ∂10-Apr-84 1657 JMC
C00067 00045 ∂10-Apr-84 1658 JMC Selker and McSun
C00068 00046 ∂10-Apr-84 1701 JMC
C00069 00047 ∂10-Apr-84 1901 JMC more non-monotonic
C00070 00048 ∂10-Apr-84 1927 JMC
C00071 00049 ∂10-Apr-84 2333 JMC name
C00072 00050 ∂11-Apr-84 0919 Mailer failed mail returned
C00073 00051 ∂11-Apr-84 0922 Mailer failed mail returned
C00074 00052 ∂11-Apr-84 0922 JMC
C00075 00053 ∂11-Apr-84 1054 JMC
C00076 00054 ∂11-Apr-84 1113 JMC
C00077 00055 ∂11-Apr-84 1512 JMC
C00078 00056 ∂11-Apr-84 2006 JMC
C00079 00057 ∂12-Apr-84 1011 JMC
C00080 00058 ∂12-Apr-84 1237 JMC
C00081 00059 ∂12-Apr-84 1259 JMC
C00082 00060 ∂12-Apr-84 1332 JMC
C00083 00061 ∂12-Apr-84 1548 JMC
C00084 00062 ∂12-Apr-84 1557 JMC
C00085 00063 ∂12-Apr-84 1857 JMC
C00086 00064 ∂12-Apr-84 1902 JMC
C00087 00065 ∂12-Apr-84 2051 JMC
C00088 00066 ∂13-Apr-84 0802 JMC
C00089 00067 ∂13-Apr-84 1835 JMC
C00090 00068 ∂13-Apr-84 1837 JMC
C00091 00069 ∂13-Apr-84 2259 JMC
C00092 00070 ∂14-Apr-84 1655 JMC Circumscription and default logic and abstract
C00096 00071 ∂14-Apr-84 1814 JMC
C00097 00072 ∂15-Apr-84 1907 JMC
C00098 00073 ∂16-Apr-84 0010 JMC
C00099 00074 ∂16-Apr-84 0746 JMC Reiter title and abstract
C00100 00075 ∂16-Apr-84 1021 JMC
C00101 00076 ∂16-Apr-84 1038 JMC
C00102 00077 ∂16-Apr-84 1044 JMC
C00103 00078 ∂16-Apr-84 1806 JMC
C00104 00079 ∂16-Apr-84 1816 JMC meeting
C00105 00080 ∂16-Apr-84 2321 JMC Re: NONMONOTONIC REASONING SEMINAR MEETS 4/18 1PM IN REDWOOD: MCCARTHY
C00107 00081 ∂17-Apr-84 1814 JMC
C00109 00082 ∂17-Apr-84 1816 JMC nonmonotonic paper list
C00110 00083 ∂17-Apr-84 1820 JMC
C00111 00084 ∂17-Apr-84 2027 JMC
C00112 00085 ∂18-Apr-84 1611 JMC
C00113 00086 ∂19-Apr-84 1423 JMC
C00114 00087 ∂19-Apr-84 2331 JMC re: video displays (from SAIL's BBOARD)
C00119 00088 ∂20-Apr-84 1004 JMC re: Relativity (from SAIL's BBOARD)
C00120 00089 ∂20-Apr-84 1112 JMC
C00121 00090 ∂20-Apr-84 1647 JMC Datamedia
C00122 00091 ∂21-Apr-84 1047 JMC death rate
C00124 00092 ∂21-Apr-84 1051 JMC vdt
C00125 00093 ∂21-Apr-84 1102 JMC re: "Washington Times" -- a warning (from SAIL's BBOARD)
C00127 00094 ∂21-Apr-84 1343 JMC Free will and determinism
C00137 00095 ∂21-Apr-84 2216 JMC 1980 Proceedings
C00138 00096 ∂22-Apr-84 1021 JMC
C00139 00097 ∂22-Apr-84 2301 JMC re: Supercomputers in the News (from SAIL's BBOARD)
C00140 00098 ∂22-Apr-84 2310 JMC
C00141 00099 ∂23-Apr-84 1351 JMC
C00142 00100 ∂23-Apr-84 1352 JMC
C00143 00101 ∂23-Apr-84 2302 JMC
C00144 00102 ∂23-Apr-84 2312 JMC
C00145 00103 ∂23-Apr-84 2316 JMC Talk by ray reiter at ibm
C00150 00104 ∂24-Apr-84 0026 JMC
C00151 00105 ∂24-Apr-84 1309 JMC change of program
C00152 00106 ∂24-Apr-84 1330 JMC
C00153 00107 ∂24-Apr-84 1350 JMC
C00154 00108 ∂24-Apr-84 1506 JMC
C00155 00109 ∂24-Apr-84 1513 JMC reference request
C00156 00110 ∂24-Apr-84 2309 JMC Yu-lin Feng
C00159 00111 ∂24-Apr-84 2325 JMC re: video displays (from SAIL's BBOARD)
C00161 00112 ∂24-Apr-84 2328 JMC
C00163 00113 ∂25-Apr-84 0049 JMC Where do Katz and Chomsky leave AI?
C00171 00114 ∂25-Apr-84 0106 JMC Where do Katz and Chomsky leave AI?
C00179 00115 ∂25-Apr-84 0130 JMC re: The Conspiracy in Military Funding (from SAIL's BBOARD)
C00181 00116 ∂25-Apr-84 0136 JMC The situation in logic
C00182 00117 ∂25-Apr-84 1041 JMC
C00183 00118 ∂25-Apr-84 1753 JMC message from David Etherington
C00185 ENDMK
C⊗;
∂01-Apr-84 1222 JMC
To: kjb@SRI-AI.ARPA
What's area F?
∂01-Apr-84 1331 JMC dictionary or word list
To: reid@SU-GLACIER.ARPA
Is the file a dictionary or just a list of words? If a dictionary, do you
know where I could steal it?
∂01-Apr-84 2032 JMC alleged corollary and also generic models
To: reiter@RUTGERS.ARPA
There has to be something wrong or a misunderstanding. I have just finished
an EKL proof, which I shall send you, that starts from the axioms I
presented in New York and proves that the only objects that fly are
birds that aren't ostriches. If we add the one more axiom, not included
in my proof that Tweety is a bird and not an ostrich, then we will
conclude that Tweety flies, which was not deducible before the
circumscription. While I circumscribe the abnormality of aspects,
and that only, and aspects are given by functions, this is not
essential. I could, as Vladimir Lifschitz and others have advocated,
have used a collection of abnormality predicates.
As for equality, I am still puzzling. I have generalized the problem
so that now we are looking for what seems to be the generic models
of a collection of axioms. This has led me to look at forcing and
other exotica. The logicians study generic models in a much more
complicated setting than what is wanted for database theory or
common sense reasoning and certainly don't come up with second
order formulas characterizing the generic models, but I still
have hopes. I am not restricting myself to circumscription.
The idea of generic model is related to the simpler idea of
a generic point of an algebraic variety or the even simpler
idea of a geometric figure in general position, i.e. satisfying
no unnecessary relations.
A related but still apparently different approach involves
trying to characterize an initial model by a suitable second
order formula. It has no maps into it from other models other
than isomorphisms. Therefore, it has as many inequalities as
possible.
Both approaches are still rather vague, and I don't even have
any formulas. Moreover, I'm having considerable difficulty
trying to understand the logical literature that I imagine
to be relevant. I'd ask for help if I knew what I wanted.
I understand that this is the best address for you.
∂01-Apr-84 2047 JMC afore-mentioned proof
To: reiter@RUTGERS.ARPA
I learned out how to get out of the fancy SAIL character set, so I
can MAIL you the proof. I suppose your corollary doesn't apply here,
but exactly why isn't clear to me. If EKL is obscure, I can send you
a manual if you like.
1. (DEFINE A
|ALL AB FLIES.A(AB,FLIES) IFF
(ALL X.NOT AB(ASPECT1(X)) IMP NOT FLIES(X))&
(ALL X.BIRD(X) IMP AB(ASPECT1(X)))&
(ALL X.BIRD(X)&NOT AB(ASPECT2(X)) IMP FLIES(X))&
(ALL X.OSTRICH(X) IMP AB(ASPECT2(X)))&
(ALL X.OSTRICH(X)&NOT AB(ASPECT3(X)) IMP NOT FLIES(X))|
NIL)
;labels: UNEQUAL SIMPINFO
2. (AXIOM
|(ALL X Y.NOT ASPECT1(X)=ASPECT2(Y))&(ALL X Y.NOT ASPECT1(X)=ASPECT3(Y))&
(ALL X Y.NOT ASPECT2(X)=ASPECT3(Y))&
(ALL X Y.ASPECT1(X)=ASPECT1(Y) IFF X=Y)&
(ALL X Y.ASPECT2(X)=ASPECT2(Y) IFF X=Y)&
(ALL X Y.ASPECT3(X)=ASPECT3(Y) IFF X=Y)|)
3. (DEFINE A1
|ALL AB FLIES.A1(AB,FLIES) IFF
A(AB,FLIES)&
(ALL AB1 FLIES1.A(AB1,FLIES1)&(ALL Z.AB1(Z) IMP AB(Z))
IMP
(ALL Z.AB(Z) IFF AB1(Z)))| NIL)
4. (ASSUME |A1(AB,FLIES)|)
deps: (4)
5. (DEFINE FLIES2 |ALL X.FLIES2(X) IFF BIRD(X)&NOT OSTRICH(X)| NIL)
6. (DEFINE AB2
|ALL Z.AB2(Z) IFF
(EX X.BIRD(X)&Z=ASPECT1(X)) OR (EX X.OSTRICH(X)&Z=ASPECT2(X))|
NIL)
7. (RW 4 (USE 3 MODE: EXACT))
A(AB,FLIES)&
(ALL AB1 FLIES1.A(AB1,FLIES1)&(ALL Z.AB1(Z) IMP AB(Z)) IMP
(ALL Z.AB(Z) IFF AB1(Z)))
deps: (4)
8. (TRW |A(AB,FLIES)| (USE 7))
A(AB,FLIES)
deps: (4)
9. (TRW
|ALL AB1 FLIES1.A(AB1,FLIES1)&(ALL Z.AB1(Z) IMP AB(Z)) IMP
(ALL Z.AB(Z) IFF AB1(Z))| (USE 7))
ALL AB1 FLIES1.A(AB1,FLIES1)&(ALL Z.AB1(Z) IMP AB(Z)) IMP
(ALL Z.AB(Z) IFF AB1(Z))
deps: (4)
10. (RW 8 (USE 1 MODE: EXACT))
(ALL X.NOT AB(ASPECT1(X)) IMP NOT FLIES(X))&(ALL X.BIRD(X) IMP AB(ASPECT1(X)))&
(ALL X.BIRD(X)&NOT AB(ASPECT2(X)) IMP FLIES(X))&
(ALL X.OSTRICH(X) IMP AB(ASPECT2(X)))&
(ALL X.OSTRICH(X)&NOT AB(ASPECT3(X)) IMP NOT FLIES(X))
deps: (4)
11. (ASSUME |AB2(Z)|)
deps: (11)
12. (RW 11 (OPEN AB2))
(EX X.BIRD(X)&Z=ASPECT1(X)) OR (EX X.OSTRICH(X)&Z=ASPECT2(X))
deps: (11)
13. (DERIVE |AB(Z)| (12 10) NIL)
AB(Z)
deps: (4 11)
14. (CI (11) 13 NIL)
AB2(Z) IMP AB(Z)
deps: (4)
15. (DERIVE |ALL Z.AB2(Z) IMP AB(Z)| (14) NIL)
ALL Z.AB2(Z) IMP AB(Z)
deps: (4)
16. (DERIVE |ALL X.NOT AB2(ASPECT1(X)) IMP NOT FLIES2(X)| (5 6) NIL)
ALL X.NOT AB2(ASPECT1(X)) IMP NOT FLIES2(X)
17. (DERIVE |ALL X.BIRD(X) IMP AB2(ASPECT1(X))| (5 6) NIL)
ALL X.BIRD(X) IMP AB2(ASPECT1(X))
18. (DERIVE |ALL X.BIRD(X)&NOT AB2(ASPECT2(X)) IMP FLIES2(X)| (5 6) NIL)
ALL X.BIRD(X)&NOT AB2(ASPECT2(X)) IMP FLIES2(X)
19. (DERIVE |ALL X.OSTRICH(X) IMP AB2(ASPECT2(X))| (5 6) NIL)
ALL X.OSTRICH(X) IMP AB2(ASPECT2(X))
20. (DERIVE |ALL X.OSTRICH(X) IMP NOT FLIES2(X)| (5 6) NIL)
ALL X.OSTRICH(X) IMP NOT FLIES2(X)
21. (DERIVE |ALL X.OSTRICH(X)&NOT AB2(ASPECT3(X)) IMP NOT FLIES2(X)| (20)
NIL)
ALL X.OSTRICH(X)&NOT AB2(ASPECT3(X)) IMP NOT FLIES2(X)
22. (DERIVE
|(ALL X.NOT AB2(ASPECT1(X)) IMP NOT FLIES2(X))&
(ALL X.BIRD(X) IMP AB2(ASPECT1(X)))&
(ALL X.BIRD(X)&NOT AB2(ASPECT2(X)) IMP FLIES2(X))&
(ALL X.OSTRICH(X) IMP AB2(ASPECT2(X)))&
(ALL X.OSTRICH(X)&NOT AB2(ASPECT3(X)) IMP NOT FLIES2(X))|
(16 17 18 19 20 21) NIL)
(ALL X.NOT AB2(ASPECT1(X)) IMP NOT FLIES2(X))&
(ALL X.BIRD(X) IMP AB2(ASPECT1(X)))&
(ALL X.BIRD(X)&NOT AB2(ASPECT2(X)) IMP FLIES2(X))&
(ALL X.OSTRICH(X) IMP AB2(ASPECT2(X)))&
(ALL X.OSTRICH(X)&NOT AB2(ASPECT3(X)) IMP NOT FLIES2(X))
23. (UE ((AB.|AB2|) (FLIES.|FLIES2|)) 1 NIL)
A(AB2,FLIES2) IFF
(ALL X.NOT AB2(ASPECT1(X)) IMP NOT FLIES2(X))&
(ALL X.BIRD(X) IMP AB2(ASPECT1(X)))&
(ALL X.BIRD(X)&NOT AB2(ASPECT2(X)) IMP FLIES2(X))&
(ALL X.OSTRICH(X) IMP AB2(ASPECT2(X)))&
(ALL X.OSTRICH(X)&NOT AB2(ASPECT3(X)) IMP NOT FLIES2(X))
24. (RW 23 (USE 22))
A(AB2,FLIES2)
25. (DERIVE |ALL Z.AB(Z) IFF AB2(Z)| (9 15 24) NIL)
ALL Z.AB(Z) IFF AB2(Z)
deps: (4)
26. (RW 8 ((USE 1 MODE: EXACT) ((USE 25 MODE: EXACT) (OPEN AB2))))
(ALL X.NOT (EX X3.BIRD(X3)&X=X3) IMP NOT FLIES(X))&
(ALL X.BIRD(X)&NOT (EX X4.OSTRICH(X4)&X=X4) IMP FLIES(X))&
(ALL X.OSTRICH(X) IMP NOT FLIES(X))
deps: (4)
27. (DERIVE |ALL X.FLIES(X) IFF BIRD(X)&NOT OSTRICH(X)| (26) NIL)
ALL X.FLIES(X) IFF BIRD(X)&NOT OSTRICH(X)
deps: (4)
28.
∂01-Apr-84 2112 JMC
To: JK
The bird files have moved to s84,jmc.
∂02-Apr-84 0841 JMC corollary
To: reiter@RUTGERS.ARPA
It occurs to me that you may have identified the predicates which are
taken as variable with those circumscribed. In my proof, you will
see that ab is the only predicate circumscribed, but both ab and
flies are taken as variable. Assuming the additional axiom
bird Tweety & not ostrich Tweety will they lead to the additional
ground instance flies Tweety.
∂02-Apr-84 0847 JMC your paper
To: SJG
I suggest you submit it to the October workshop on non-monotonic
reasoning. The deadline for that is May 1. I'll have a few other
comments. One is to remove the title Dr. from your name. It's
not customary here.
∂02-Apr-84 0849 JMC
To: DFH
Please check that I have reservations at the Doral Hotel Wed. thru Fri.
∂02-Apr-84 1543 JMC
To: gotelli@SU-SCORE.ARPA
CC: muller@SU-SCORE.ARPA, DFH@SU-AI.ARPA
Please keep the SAIL account for Eric Muller (EXM) alive for this
quarter. It may be charged to my DARPA account.
∂02-Apr-84 1746 JMC
To: roode@SRI-NIC.ARPA
We haven't made any decisions, but I agree with your comment.
∂02-Apr-84 1747 JMC [John McCarthy <JMC@SU-AI.ARPA>:]
To: DFH
∂02-Apr-84 1632 GOTELLI@SU-SCORE.ARPA [John McCarthy <JMC@SU-AI.ARPA>:]
Received: from SU-SCORE.ARPA by SU-AI.ARPA with TCP; 2 Apr 84 16:32:09 PST
Date: Mon 2 Apr 84 16:25:16-PST
From: Lynn Gotelli <GOTELLI@SU-SCORE.ARPA>
Subject: [John McCarthy <JMC@SU-AI.ARPA>:]
To: JMC@SU-AI.ARPA
Would that be budget account number 2-FCZ780?
---------------
Return-Path: <JMC@SU-AI.ARPA>
Received: from SU-AI.ARPA by SU-SCORE.ARPA with TCP; Mon 2 Apr 84 15:42:25-PST
Date: 02 Apr 84 1543 PST
From: John McCarthy <JMC@SU-AI.ARPA>
To: gotelli@SU-SCORE.ARPA
CC: muller@SU-SCORE.ARPA, DFH@SU-AI.ARPA
Please keep the SAIL account for Eric Muller (EXM) alive for this
quarter. It may be charged to my DARPA account.
-------
∂02-Apr-84 2222 JMC re: military funding of computer research (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - Some of us argue that research that helps defense is worthwhile
for that reason.
∂03-Apr-84 0926 JMC
To: DFH
Yes, all but Muller's.
∂03-Apr-84 1014 JMC
To: buchanan@SUMEX-AIM.ARPA
ok
∂03-Apr-84 1443 JMC
To: DFH
Please put off Svendsen one hour.
2pm, Werner Svendsen, Danish TV, 202 376-7210 USIA, Don McDonough
The Search Committee will meet with Nils Nilsson Monday, April, 9
from 2:00 to 3:00 in room 220 (Chairman's conference room).
Knuth dinner for Nilsson
∂03-Apr-84 1608 JMC
To: YOM
Can you make me a copy Stalnaker's note and one for the non-monotonic file?
∂03-Apr-84 1652 JMC
To: YOM
You don't list your reference to Mc2 among references.
∂03-Apr-84 1724 JMC
To: YOM
Well, I don't like to correct you on the references of your own paper,
but since it occurs in reference to the Wise men puzzle, Joe probably
intended to refer to my work on that puzzle. The main work is
unpublished, but there is probably something about it in my
AI memo with the Japanese. Unless you want to mention unpublished
work or track down the other reference, you had probably best
delete the reference.
∂03-Apr-84 2218 JMC
To: grosof@SUMEX-AIM.ARPA
Ask Carolyn Talcott for her mailing list. Also make sure that VAL (Vladimir
Lifschitz) gets the notices.
∂03-Apr-84 2223 JMC
To: hull@SU-SCORE.ARPA
The one-liner was because I have spoken my piece on why I think defense
research is important for itself and I only wanted to point out that
this point of view, which ought to be mentioned if only to be refuted,
was being ignored. Because I have said all I want to say for the present,
I feel it is the other side's move on that issue.
∂03-Apr-84 2228 JMC SAIL accounts
To: faculty@SU-SCORE.ARPA, reges@SU-SCORE.ARPA
jmc - I haven't got around to arguing about it, but I don't see why students
shouldn't have equal access to SAIL as to SCORE, especially as SCORE is
now said to be fully loaded.
∂03-Apr-84 2244 JMC re: UCLA, Ronald Austin (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - If I remember what Brian Reid said, this Austin really did a
substantial amount of sabotage. Moreover, the stolen airplane tickets
do not substantiate his claim to be an innocent who didn't know what
he was doing was against the law. Of course, they may have the charges
screwed up due to inexperience in computer matters. In this case, the
computer science community should help the DA get the charges straight.
However, I agree with BH that the media is exaggerating the
harm that can be done. In part this seems to be deliberate; it gets
a journalist's articles published and it makes money for movie makers.
∂03-Apr-84 2251 JMC
To: wiederhold@SUMEX-AIM.ARPA
I see no reason why it should be "either or".
∂03-Apr-84 2254 JMC
To: hull@SU-SCORE.ARPA
While I see nothing actually incorrect about my reply to your message,
I must confess that I read it before reading your BBOARD message.
Many thanks for the encouragement.
∂03-Apr-84 2308 JMC
To: wiederhold@SUMEX-AIM.ARPA
I would be very surprised if account management expense were significant,
since each of the systems have automated their accounting and their
transmission of the bill to SCORE for allocation to accounts. Of course,
some of the machines may have minimum charges per account, but at least
on SAIL, these are trivial.
∂03-Apr-84 2321 JMC
To: DFH
Please print toyota.ns[s84,jmc] and send it to Prof. Paul Ehrlich, Biology.
∂04-Apr-84 0730 JMC secretary
To: bscott@SU-SCORE.ARPA
I have two thoughts. The first is that the girl who was second choice
when we hired Diana would be suitable although apparently not for financial
work. However, it isn't clear that an interest in financial work is
an advantage for Zohar and me. The second is that Fran Larson should
be asked if she is interested in working temporarily. Finally, I'd
really like to know if Diana found some aspects of the job unsatisfactory.
∂04-Apr-84 0747 JMC
To: DFH
It would be good if you could prepare some notes for your succesor.
∂07-Apr-84 2121 JMC for David Etherington
To: reiter@RUTGERS.ARPA
The promised paper with a change of name became my 1980 AI Journal paper.
∂07-Apr-84 2126 JMC seminar
To: reiter@RUTGERS.ARPA
You will be talking to a seminar on non-monotonic reasoning. They will
have heard about circumscription and heard of the other papers in
the 1980 AI Journal including yours. Some will have read them.
Please transmit an invitation to him. Perhaps it would be convenient
for the two of you to come at the same time, but I want to wait on that
till the first meeting of the seminar to see if the others find it
reasonable to have two seminars in one week. Otherwise, we'll make
it later.
∂08-Apr-84 1146 JMC S.F. OPERA SEASON HALF-SUBSCRIPTION FOR SALE
To: CLT
∂08-Apr-84 1121 SIEGMAN@SU-SIERRA.ARPA S.F. OPERA SEASON HALF-SUBSCRIPTION FOR SALE
Received: from SU-SIERRA.ARPA by SU-AI.ARPA with TCP; 8 Apr 84 11:21:14 PST
Date: Sun 8 Apr 84 11:18:27-PST
From: Tony Siegman <SIEGMAN@SU-SIERRA.ARPA>
Subject: S.F. OPERA SEASON HALF-SUBSCRIPTION FOR SALE
To: su-bboards@SU-SIERRA.ARPA
cc: SIEGMAN@SU-SIERRA.ARPA
SAN FRANCISCO OPERA SUBSCRIPTION FOR SALE
Two Grand Tier seats (B-26,28, second row, on the right-center aisle)
for the 1984 Autmn Half-Series "O" (5 Sunday afternoon performances):
"La Sonnambula" (Bellini), with Frederica von Stade, Oct. 7
"Madama Butterfly" (Puccini), with Leona Mitchell, Oct. 14
"Anna Bolena" (Donizetti), with Joan Sutherland, Oct. 28
"Khovanshchina" (Mussorgsky), with Matti Salminen, Nov. 11
"Rigoletto" (Verdi), with Hendricks, Raffanti, Wixell, Dec. 2
For sale only in toto; no individual performances: $550 (of which $100
is tax-deductible contribution). Msgs to Siegman@Sierra, 7-0222, 7-1683
or 326-6669 (evenings).
-------
∂09-Apr-84 0011 JMC
To: PHY
I was out of town. I'll attend the dinner for Nilsson.
∂09-Apr-84 1006 JMC message for John Cocke
To: capek%ibm-yorktown@CSNET-RELAY.ARPA
Please give him the following and remind him to set himself up to
receive csnet mail directly. As agreed I will visit Yorktown for
the week of May 7. Please get the business matters arranged. I'm
now getting $750 per day plus expenses for consulting. I hope that's
feasible.
∂09-Apr-84 1313 JMC paper
To: YOM@SU-AI.ARPA, halpern.ibm-sj@CSNET-RELAY.ARPA
You many be interested in "First order reasoning about knowledge and
belief" by Wolfgang Bibel.
∂09-Apr-84 1709 JMC Is this a correct address?
To: Capek.yktvmx.ibm-sj@CSNET-RELAY.ARPA
Originally, I had a message for John Cocke that I was asking
you to relay, since he isn't on csnet yet but supposed you
would be. However, my first attempt failed, so I transmitted
the message by telephone. So all I want to know now is whether
this is a correct network path.
∂09-Apr-84 1747 JMC
To: LEP
Thanks, I'll look at it.
∂09-Apr-84 2024 JMC re: Undergraduates, and CS (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - The Math Sciences program lets a student go pretty far in the
direction of a CS major - perhaps far enough. A student would need
fairly special circumstances to convince me that an individually
designed program - lots of trouble for all concerned - was warranted
in such a case.
∂10-Apr-84 0133 JMC
To: jiml@SU-SCORE.ARPA
It was a mistake I've made before - thinking I was reading bboard
while reading my junk mail file. There was a message sent to
faculty suggesting a discussion of undergraduates who want
a special undergraduate major that amounts to a major in computer
science.
∂10-Apr-84 0717 JMC
To: schreiber@SU-SCORE.ARPA
I have to be in L.A. todαay.
∂10-Apr-84 1631 JMC
To: CLT
Rod is having dinner with Goguen, so we'll call him there.
Goguen, Joe 326-6832
∂10-Apr-84 1634 Mailer failed mail returned
To: JMC
In processing the following command:
MAIL bgb
The following message was aborted because of a command error,
namely, nonexistent recipient(s):
BGB
------- Begin undelivered message: -------
∂10-Apr-84 1634 JMC
I would like to get reactions to that talk, but I'll be in L.A. this
Friday. Name another.
------- End undelivered message -------
∂10-Apr-84 1635 JMC
To: buchanan@SUMEX-AIM.ARPA
I would like to get reactions to that talk, but I'll be in L.A. this
Friday. Name another.
∂10-Apr-84 1657 JMC
To: EJS
Are you interested in completing McSun?
∂10-Apr-84 1658 JMC Selker and McSun
To: bosack@SU-SCORE.ARPA
There is a bboard message from Ted Selker saying that he is looking for
a job. We should get him to do the McSun project if it isn't already
well started. I MAILed him an inquiry.
∂10-Apr-84 1701 JMC
To: DFH
Please try to arrange a Computer Facilities Committee meeting.
∂10-Apr-84 1901 JMC more non-monotonic
To: DFH
The file foo[1,jmc], page 2, contains the names of more papers to be
copied from my library for the non-monotonic file. Page 3 contains names
of papers which I believe have already been copied or are in the group to
be copied in your office. Please complete the list of papers and make
15 copies (just of the list) and put them in a file folder in front of the
one (or several) holding the papers themselves. It would be good to have
all the copying done and the file set up by the time of my seminar at 2:00
tomorrow, since I want to refer to the collection.
∂10-Apr-84 1927 JMC
To: CLT
Rod wanted to put it off till his next trip.
∂10-Apr-84 2333 JMC name
To: grosof@SUMEX-AIM.ARPA
Vladimir Lifschitz, VAL@SU-AI, should be on the non-monotonic list.
∂11-Apr-84 0919 Mailer failed mail returned
To: JMC
In processing the following command:
REMIND/date=monday/time=9 senior
The following message was aborted because of a command error,
namely, nonexistent recipient(s):
senior
------- Begin undelivered message: -------
∂11-Apr-84 0919 JMC
------- End undelivered message -------
∂11-Apr-84 0922 Mailer failed mail returned
To: JMC
In processing the following command:
MAIL bgb
The following message was aborted because of a command error,
namely, nonexistent recipient(s):
BGB
------- Begin undelivered message: -------
∂11-Apr-84 0922 JMC
On Friday April 27, I can talk with title "Some expert systems need common sense"
------- End undelivered message -------
∂11-Apr-84 0922 JMC
To: buchanan@SU-SCORE.ARPA
On Friday April 27, I can talk with title "Some expert systems need common sense"
∂11-Apr-84 1054 JMC
To: DFH
Here's another for the non-monotonic file.
Doyle, Jon\A truth maintenance system\MIT, AI Memo 521\1979.
∂11-Apr-84 1113 JMC
To: buchanan@SUMEX-AIM.ARPA
Abstract: Several aspects of common sense knowledge and reasoning ability will
be discussed. We will emphasize the ability
to predict the consequences of contemplated actions. Sometimes human
expertise involves such prediction, but many expert systems attempt to
fit all expertise into production rules that recommend an action based
on the situation fitting a pattern. When does this suffice and when is
prediction required for good performance?
∂11-Apr-84 1512 JMC
To: buchanan@SUMEX-AIM.ARPA
I assume that SIGLUNCH is at 12:05 not 1:05.
∂11-Apr-84 2006 JMC
To: PHY
Dinner chez Knuth, 6pm
∂12-Apr-84 1011 JMC
To: RPG
I suggest you mention the packing list problems to Lynn Gotelli's boss if you
think they are serious enough to warrant it. As for Symbolics, try your
influence, and if that doesn't work, I'll try mine.
∂12-Apr-84 1237 JMC
To: ashok@SU-SCORE.ARPA
Please come and see me about Advanced Reading and Research.
∂12-Apr-84 1259 JMC
To: DFH
bynum.1
∂12-Apr-84 1332 JMC
To: DFH
griffi.1
∂12-Apr-84 1548 JMC
To: pack@SU-SCORE.ARPA
CC: DFH@SU-AI.ARPA
∂12-Apr-84 1511 DFH Leslie Pack
We were paying for her SAIL charges. Since she is
now your SRA, we will also be getting charged for
her SCORE and ALTOS charges, unless you don't want
to do this, in which case the accounts would have to
be closed.
Leslie:
We plan to pick up the charges mentioned, but do you need the
Alto account? Also will you still be doing work for Luckham for the
computer use of which he should pay?
∂12-Apr-84 1557 JMC
To: pack@SU-SCORE.ARPA
If the charges for the Alto account are trivial, by all means keep it.
However, my impression is that charges for the Altos are not usage
based but are account based, so we are charged whether you use it
or not. I believe we had one Alto account in the group.
∂12-Apr-84 1857 JMC
To: goguen@SRI-AI.ARPA
Could you send me your "Initiality Primer"?
∂12-Apr-84 1902 JMC
To: DFH
Sato will be a visiting scholar without salary or expenses.
∂12-Apr-84 2051 JMC
To: faculty@SU-SCORE.ARPA
I agree with making concessions to those we want most.
∂13-Apr-84 0802 JMC
To: DFH
Hope you like your new job. We're all sorry you're leaving.
∂13-Apr-84 1835 JMC
To: goguen@SRI-AI.ARPA
I'll wait for the revised and improved version.
∂13-Apr-84 1837 JMC
To: SJG
They are Ira Goldstein and Bert Raphael at HP and Peter Hart at Fairchild.
∂13-Apr-84 2259 JMC
To: suppes@IMSSS
Carolyn and I are happy to accept your April 29 party invitation.
∂14-Apr-84 1655 JMC Circumscription and default logic and abstract
To: grosof@SUMEX-AIM.ARPA
Here are some remarks on their relation. See if you agree.
1. Circumscription handles disjunction and default logic doesn't.
2. Because default logic uses consistency, it is sometimes more powerful
than circumscription, but the user has to find the models required to
prove consistency. Therefore, it is in general highly non-constructive.
3. Circumscription produces a second order formula in an entirely
effective way. However, to make good use of the formula one has to
find out what to substitute for the predicate or function variables.
In general, this is undecidable. However, I believe there will be cases
in which default logic (non-constructively) yields a conclusion that the
corresponding circumscription cannot yield. This is because there may
exist no formulas that constitute suitable substitutions. I don't have
a proof of this.
4. In the simple cases that have been used as examples, there is a
substantial equivalence between the models used to establish consistency
in default logic and the formulas substituted for the predicate variables
in circumscription. However, an ordinary first order theorem prover
or proof checker would have to be supplemented by a metamathematical
component to find the models for default logic, while a second order
theorem prover or proof checker would work directly in circumscription
case. In fact I have made several circumscription proofs using EKL.
You may print copies if you want.
Now here's the abstract.
Abstract: A somewhat generalized form of circumscription applies to a wff
rather than a predicate. There are also some new applications - to
the unique names hypothesis, to is-a hieararchies and to the frame problem.
If time permits, the Bossu and Siegel form of non-monotonic inference will
be described and related to circumscription.
∂14-Apr-84 1814 JMC
To: grosof@SU-SCORE.ARPA
I'm a duplicate, since I get the CSLI announcements.
∂15-Apr-84 1907 JMC
To: pack@SU-SCORE.ARPA
If your FINGER message had fewer blank lines, the top wouldn't vanish ...
∂16-Apr-84 0010 JMC
To: pack@SU-SCORE.ARPA
I don't have anything for you yet. Probably tomorrow am.
∂16-Apr-84 0746 JMC Reiter title and abstract
To: grosof@SUMEX-AIM.ARPA
The title of my talk is "The Closed World Assumption". It will contain much
of the material of my Brooklyn talk, but will also include recent work on
circumscription and equality that I have been doing with Etherington and
Mercer, and work on negation as failure in PROLOG that I have been doing with
Besnard.
∂16-Apr-84 1021 JMC
To: DFH
I don't recall making any such reservation, and I have nothing
scheduled for that time. They should check whether it might be
Perry McCarty. While you have them on the line, please ask for
a reservation for Raymond Reiter to stay April 24 and 24. If you
can't get that, make a reservation at the Holiday Inn. Also please
make a reservation for two for me today at 12:15.
∂16-Apr-84 1038 JMC
To: CLT
Lipp Hydronics 327-1943 will tentativel come Thursday.
∂16-Apr-84 1044 JMC
To: reiter@RUTGERS.ARPA
CC: DFH@SU-AI.ARPA
∂16-Apr-84 1038 DFH
To: JMC, DFH
There was nothing available for Reiter at Faculty Club on 24th and 25th.
I have made reservation for him at Holiday Inn. Confirmation # 81AC31F6.
The Holiday Inn is on El Camino Real just South of the entrance to the
Stanford campus. Do you need more directions. It will be guaranteed
for late arrival.
∂16-Apr-84 1806 JMC
To: LEP
αXDEFINE REcon⊗↔⊗; reconnecting a maclisp job after re-entering e
α-αxsubjob⊗↔αxrec⊗↔αβ⊗↓
∂16-Apr-84 1816 JMC meeting
To: ashok@SU-SCORE.ARPA
How about tomorrow at 2 or 4:30 or 5:15?
∂16-Apr-84 2321 JMC Re: NONMONOTONIC REASONING SEMINAR MEETS 4/18 1PM IN REDWOOD: MCCARTHY
To: nilsson@SRI-AI.ARPA
This message posted on BBOARD@SU-AI.
∂16-Apr-84 0915 @SU-SCORE.ARPA,@SUMEX-AIM.ARPA:NILSSON@SRI-AI.ARPA Re: NONMONOTONIC REASONING SEMINAR MEETS 4/18 1PM IN REDWOOD: MCCARTHY
Received: from SU-SCORE.ARPA by SU-AI.ARPA with TCP; 16 Apr 84 09:15:11 PST
Received: from SUMEX-AIM.ARPA by SU-SCORE.ARPA with TCP; Mon 16 Apr 84 08:56:40-PST
Received: from SRI-AI.ARPA by SUMEX-AIM.ARPA with TCP; Mon 16 Apr 84 08:58:24-PST
Date: Mon 16 Apr 84 09:01:10-PST
From: NILSSON@SRI-AI.ARPA
Subject: Re: NONMONOTONIC REASONING SEMINAR MEETS 4/18 1PM IN REDWOOD: MCCARTHY
To: GROSOF@SUMEX-AIM.ARPA
cc: su-bboard@SUMEX-AIM.ARPA, NILSSON@SRI-AI.ARPA
In-Reply-To: Message from "Benjamin Grosof <GROSOF@SUMEX-AIM.ARPA>" of Sun 15 Apr 84 14:10:33-PST
Benjamin, I received two of your nonmonotonic logic seminar notices. -Nils
-------
∂17-Apr-84 1814 JMC
To: RPG@SU-AI.ARPA
CC: golub@SU-SCORE.ARPA
∂17-Apr-84 1738 RPG
Here's an example of my not being able to do things I'd like to do
and which are appropriate for me to do (from Andy Freeman):
Marilyn Walker says that you can't be the official supervisor of
293 so I signed up with McCarthy instead. I'll get in touch with
Ullman if you think he would be more appropriate. I'd also like
to get a comp. programming project out of this, but it doesn't
have to be finished this quarter.
Dick:
I think we can fix this one. I think the Department should certify Senior
Research Associates automatically eligible for this.
∂17-Apr-84 1816 JMC nonmonotonic paper list
To: DFH
In the nonmonotonic library file drawer is a list of the papers in
the collection, presumably with the file name attached. Please update
that file to include the papers that have been added. I'll need 40 copies
to take to my lecture at 1pm.
∂17-Apr-84 1820 JMC
To: DFH
Please make a reservation for two for noon Friday at Faculty Club.
∂17-Apr-84 2027 JMC
To: CLT
A La Carte 494-7478, 2310 El Camino (just N of California)
(former operators of Alouette)
Sophie or Marie-Jose
∂18-Apr-84 1611 JMC
To: library@SU-SCORE.ARPA
CC: feigenbaum@SUMEX-AIM.ARPA, buchanan@SUMEX-AIM.ARPA
My own opinion is that the Library should regard the AI trade press as
contrasted with the AI scientific press as something of lower priority.
I haven't read any of it, but my impression is that it is of lower
interest than the engineering trade press. At least no-one has ever
referred in writing or orally to anything in any of these newsletters.
∂19-Apr-84 1423 JMC
To: PHY
That will be fine. Any specific subject?
∂19-Apr-84 2331 JMC re: video displays (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - The video display danger is almost certainly just another of the
panics started by the politically paranoid who suspect that any innovation
is likely to be a plot of the bosses against the workers or at least
another example of how their greedy search for profits endangers the
workers. As long as video displays were used in labs, no-one imagined any
danger.
Well, you say, after all there may be a danger and isn't it better
to be safe than sorry. Better spend any amount of money rather than risk
life. Wrong, because spending money uselessly costs lives. We can
quantify this. Namely, take your trusty HP-15 calculator or similar and
correlate income with death rate by states. The data are conveniently
available in the Statistical Abstract of the United States published each
year. The result of this computation is, as any good liberal would
expect, is that the rich live longer than the poor - the death rate in a
state declines as income rises. The co-efficient is $2.5 million per
life. By throwing out some states as exceptional or other fiddling, you
can make the co-efficient anything between $1.0 million and $4.0 million.
I don't care which number you prefer.
This indicates that if you pump $2.5 million into the economy of a
state, people will value their lives a bit more, go to the doctor more
often or pay taxes for safety measures enough to save one life. Of
course, one can save a life for a lot less than $2.5 million if you target
the money for life saving, but just throwing money into the economy saves
lives at that rate.
Conversely, wasting money costs lives at $2.5 million per life.
Therefore, any safety measure that doesn't save at least one life for
every $2.5 million spent costs lives. As it happens many federally and
state mandated safety measures don't meet this criterion. Some nuclear
power plant safety measures cost $150 million per life saved and many of
the earthquake precautions required in building construction cost 5 to 6
million per expected life saved. To put it clearly, if utility rates go
up to pay for a safety measure, lives are lost unless the safety measure
saves a life for every $2.5 million spent.
Research money spent chasing will-o-the-wisp dangers is subject to
the same criterion. I'll bet those suspicious of VDTs have forced the
spending of more than $10 million and thus have killed four or more
people. So far there is no evidence they have saved anyone.
∂20-Apr-84 1004 JMC re: Relativity (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - The physicists have haggled over this at length, and many popular
books on physics discuss it. As for the theory of relativity, it would
seem that the theory of relativity isn't that relative relative to rotation.
∂20-Apr-84 1112 JMC
To: bmoore@SRI-AI.ARPA
Gabbay's talk is at 1pm in Redwood.
∂20-Apr-84 1647 JMC Datamedia
To: pack@SU-SCORE.ARPA
CC: ME@SU-AI.ARPA
Please MAIL to ME and to DFH the serial numbers of the Datamedia and
keyboard that you now have.
∂21-Apr-84 1047 JMC death rate
To: tyson@SRI-AI.ARPA
The death rate is the fraction of the population dying each year. If the
population were staying constant, the death rate would be the reciprocal
of the average length of life. Actually, if I remember correctly, the
death rate ranges between 4.5 per thousand in Alaska to about 11 per thousand
I forget where. If these were steady state figures, then one would imagine
that Alaskans live 220 years and that people in the worst state live about
90 years. The regrettable discrepancy with reality is caused by the fact
that the population was increasing, and the elderly who are contributing the
most to the death rate were born when the population was about half what it
is today.
The number you want, how many dollars in income extends the average life
a year, could be computed from the age specific death rates and the age
distribution. All these are available in the Statistical Abstract of
the United States (hint).
∂21-Apr-84 1051 JMC vdt
To: amsler@SRI-AI.ARPA
Thanks for the news story. I didn't post it, because information along
the same lines is already posted. However, I don't think you should be
shy about MAILing to SU-BBOARDS. Everyone does it, and there seems to
be no more complaint about mail from outsiders than from Stanford people.
∂21-Apr-84 1102 JMC re: "Washington Times" -- a warning (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - I am an atheist and think that cults are particularly bad for
you. While I've never met a Moonie, I can't imagine that they are
an exception in their cult aspect. However, I am also inclined to
believe that much of the animus against them in particular is political.
Some of it comes from people who praised Jim Jones. As for the Washington
Times and the New York Tribune, I've read both, and I find them to be
good newspapers. They are as good as, and in time may achieve the
respectability of, the Christian Science Monitor, which is also put
out by a crackpot religion. The Moonie papers don't even have a religious
page as I recall, and in the few samples I have seen, don't seem to
give great prominence to Moon and his activities. The people who
actually put out the papers are presumably professional journalists,
and I doubt that many are members of the Unification Church.
Admittedly we conservatives feel rather lonely as far as newspapers
are concerned and have to be grateful for what we can get.
∂21-Apr-84 1343 JMC Free will and determinism
To: perlis%umcp-cs.csnet@CSNET-RELAY.ARPA, dam%oz@MIT-MC.ARPA
Dear Don:
This letter should be considered as submitted to DAM's refereed
Logic in AI mailing list, should it be implemented. This is why a put the
references in a proper style.
Thanks for recommending (Berofsky 1966) which took me a while to
read.
As you said, Hobart was more or less on our side. However, it
seems to me that AI has a lot to say about the question of what free will
is that was missed by the philosophers. There are several key areas where
AI can contribute and needs to contribute for its own benefit.
1. AI is inclined to take the "design stance" as Daniel Dennett
calls it. Therefore, the issue for us is what attitude shall we build
into a robot about its own free will and that of other robots and people.
Clearly a robot needs to apply the word "can" to itself. If it is to
choose satisfactorily among alternatives it must reason about what it can
and cannot do.
Notice that it is much more important for it to reason about the
future than about the past. It may sometimes be worthwhile for it to
reason about what it could have done, i.e. in order to learn from its
experience to do better next time, but this is less important for its
useful function than listing its present alternative feasible objectives.
2. The philosophers' emphasis on "could have" rather than "can" is
a consequence of their interest in whether praise or blame or reward or
punishment can appropriately be attached to actions. While these are
interesting problems, it is important to notice that the problem of "can"
can be attacked without touching them.
3. I am still of the opinion expounded in my paper with Pat Hayes
(1969), that what a robot (or automaton) can is describable in terms of an
automaton system in which the robot's outputs are replaced by external
inputs to the system. However, that paper left itself open to carping by
not making clear enough that a wide variety of different automaton systems
are needed to explicate the variety of useful human usages of "can".
4. Notice also that the question of moral judgment can be
discussed independently of reward or punishment. A child might say, "It's
a bad robot, it kicked my cat" without having any opinion on whether
robots are more appropriately punished or debugged.
5. While we're on children, I remember that my younger daughter at
a quite early age replied to a question of whether she could do something
with the statement, "I can, but I won't". We have a lot of work to do
before robots can behave in this sophisticated a way. Even a useful use
of "but" won't be trivial to achieve.
ADDITIONAL QUESTIONS
Here are some questions suggested by the papers.
1. A person often wishes that his wishes were different from what
they are. For example, a person may wish he wasn't hungry or didn't want
a cigarette. Is it ever worthwhile to program robots this way? It would
seem that, unlike a human, a robot should be able to change its wishes
arbitrarily, so that its higher level goal would automatically take
precedence.
2. What about unrealizable wishes? Is it ever appropriate for the
robot to wish that an airplane flight were shorter? It seems to me that
the answer is yes. In the first place, it may not be obvious whether a
wish is realizable. Unexpected opportunities may arise. Second, the
remembered experience of wishes, even unrealizable ones, affects future
behavior. Our robot may now put some value on avoiding long airplane
flights.
3. The simplest way to look at a robot or even a human is to
regard its motivation as unitary. It weighs various aspects of the
possible outcomes of its actions and decides on an overall preference.
For humans this models seems unrealistic. We avoid weighing different
motivations against one another as long as possible, because it is
difficult and pursue different goals in parallel and suffer when they turn
out to be incompatible. It may be worthwhile to program robots in a
similar way to pursue goals in parallel and decide among them only when
they turn out to be incompatible in the immediate future.
Indeed it may be too difficult to design an overall evaluation
function. How to weigh providing its master with an ice cream soda
against avoiding the risk of exciting another person's allergy to cats may
never arise.
4. There are some problems which may turn out to be difficult in
formalizing how reasoning causes action in robots. However, Lewis
Carroll's infinite regress (so much admired by Hofstadter) in modus ponens
should have been answered as follows by Achilles. "There are various
justifications of modus ponens that might be given, but the reason I use
it is that that's the way I'm built".
5. Of all the papers, Davidson's was the most concrete from the AI
point of view. It might just be possible to program a Davidsonian robot.
The action scanner looks for reasoned conclusions of the form "I should do
X", where X is a "primary action".
6. The key to avoiding philosophical problems is to refrain from
excessive generality. All common sense psychology consists of
"approximate theories" in the sense of my 1980 paper. This means that the
concepts fall apart if examined too closely.
References:
Berofsky, Bernard (ed.) (1966): "Free will and Determinism", Harper and
Row.
McCarthy, John and P.J. Hayes (1969): "Some Philosophical Problems from
the Standpoint of Artificial Intelligence", in D. Michie (ed), Machine
Intelligence 4, American Elsevier, New York, NY.
McCarthy, John (1979): "Ascribing Mental Qualities to Machines" in
Philosophical Perspectives in Artificial Intelligence, Ringle, Martin
(ed.), Harvester Press, July 1979. .<<aim 326, MENTAL[F76,JMC]>>
Best Regards,
John McCarthy
∂21-Apr-84 2216 JMC 1980 Proceedings
To: dswise%oregon-state.csnet@CSNET-RELAY.ARPA,
boyer@UTEXAS-20.ARPA
I'd prefer a solution that keeps paper copies in print. Also it
would be nice if all the Lisp conference proceedings became and
remained available from the same source.
∂22-Apr-84 1021 JMC
To: dam%oz@MIT-MC.ARPA
Yes, it is all right to forward my message. Good luck with the thesis.
∂22-Apr-84 2301 JMC re: Supercomputers in the News (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - Or, most likely of all, has overstaffed.
∂22-Apr-84 2310 JMC
To: ARK
jmc - Arthur you are over-simplifying. A calander is a cross between
a colander and a calendar. It is a device for draining spaghetti in
the holes of which are in the shape of dates. Besides, it isn't just
that Scott Meyers wants one; he's politely asking if someone at SAIL
took it. We need to re-assure him that no-one has seen it
in [1,3] or anywhere else at SAIL. [Forwarded to you know whom.]
∂23-Apr-84 1351 JMC
To: MDD
I have five suggestions:
1. Check whether there is a discount program for Berkeley faculty.
2. Check with NYU to see if IBM has offered a discount there.
3. Call Ralph Phillips, who bought a PC and recently retired
from the Stanford Math. Dept..
4. Call IBM and ask if there is any discount program for
university faculty generally.
5. Check the ads in the magazines oriented towards PC owners.
If you get yourself a PC, you should get a configuration
appropriate for LISP and Prolog. You may find them interesting after
all these years.
∂23-Apr-84 1352 JMC
To: MDD
Here are the first two lines of the previous message.
I don't know. I'm not presently motivated to get a micro, because
my home terminal on this machine seems to dominate it. However,
∂23-Apr-84 2302 JMC
To: ARK
His talk will be Wednesday at 1pm in the non-monotonic reasoning
seminar. I believe it will be in Redwood Hall G-19. You can get
te abstract from Ben Grosof.
∂23-Apr-84 2312 JMC
To: ARK
You'll miss him almost certainly. He's arriving at 9pm tomorrow,
and I don't think he intends to stay till Friday.
∂23-Apr-84 2316 JMC Talk by ray reiter at ibm
To: ARK
∂23-Apr-84 1801 HALPERN.SJRLVM1%ibm-sj.csnet@csnet-relay.arpa Talk by ray reiter at ibm
Received: from SU-HNV.ARPA by SU-AI.ARPA with TCP; 23 Apr 84 18:00:17 PST
Received: from csnet-relay by Diablo with TCP; Mon, 23 Apr 84 17:56:47 pst
Received: From ibm-sj.csnet by csnet-relay; 23 Apr 84 20:23 EST
Date: 23 Apr 1984 15:07:55-PST (Monday)
From: Joe Halpern <HALPERN%ibm-sj.csnet@csnet-relay.arpa>
To: KNOWLEDGE@su-hnv.ARPA
Subject: Talk by ray reiter at ibm
Besides the regular knowledge seminar which will be held this Friday,
April 27 at 10, with Ruth Manor speaking, there will also be a
knowledge-related talk given by Ray Reiter on Thursday, April 26,
at 1 PM, in Cafeteria A of Building 28 at IBM. This is not the same
talk that Reiter is giving at Stanford on Wednesday. I've appended
the abstract below. If you're interested
in coming to Reiter's talk, please come a few minutes early to Building
28 (the same place where the regular knowledge seminars are held)
and have the receptionist call me. I will have to escort you
to Cafeteria A. -- Joe
TOWARDS A LOGICAL RECONSTRUCTION OF RELATIONAL DATABASE THEORY
Raymond Reiter
Canadian Institute for Advanced Research
and
Department of Computer Science
University of British Columbia
Insofar as relational database theory owes a debt to logic, the currency
on loan is model theoretic in the sense that a database is viewed as a parti-
cular kind of first order interpretation, and query evaluation is a process of
truth functional evaluation of first order formulae with respect to this inter-
pretation. I shall argue that a proof theoretic view of databases is possible,
and indeed is much more fruitful.
To illustrate the utility of this perspective I shall show how to formal-
ize, within first order logic, null values denoting existing but unknown indi-
viduals, thereby providing such nulls with an intuitively correct semantics
without appealing to many valued logics. I shall then sketch an extended rel-
ational algebra which provides a provably sound and sometimes complete query
evaluation algorithm for null values. These soundness and completeness results
are proved relative to the logical semantics of such nulls.
En route I shall touch on various logical issues concerning integrity con-
straints, conceptual modelling, and why it is that database theoreticians
should worry about AI exotica like non-monotonic reasoning.
∂24-Apr-84 0026 JMC
To: ARK
Leave it with Fran Larson.
∂24-Apr-84 1309 JMC change of program
To: feng@SU-SCORE.ARPA
I think you would be better off with Boyer and Moore if they have
time to help you. I have as many people as I can pay attention to.
If you wish I will telephone them and inquire for them or you can
do it yourself.
∂24-Apr-84 1330 JMC
To: *
It took AP 15 minutes to get the story on the wire.
∂24-Apr-84 1350 JMC
To: genesereth@SUMEX-AIM.ARPA
What's missing is the Lisp and Prolog intro. Also Do Prevention and ab
along with common sense can be compressed into two rather than 3.
∂24-Apr-84 1506 JMC
To: elyse@SU-SCORE.ARPA
I have no addition to propose, except that the conventional wording
would be either "the motion was passed unanimously" or "the vote
in favor of the motion was unanimous". "vote is unanimously passed"
is linguistically anomalous.
∂24-Apr-84 1513 JMC reference request
To: minsky%oz@MIT-MC.ARPA
Is there a published reference to your expression of doubt that logic
could properly handle the exceptions to birds flying and the exceptions
to the exceptions. I thought it was in your AAAI Presidential address,
but I can't find it in the published form in the AI Magazine.
I have written a new paper on circumscription, and this is one of
the examples.
∂24-Apr-84 2309 JMC Yu-lin Feng
To: cl.boyer@UTEXAS-20.ARPA, cl.moore@UTEXAS-20.ARPA
Are you interested in receiving a Chinese interested in working with you
or in discussing it with him. He worked with Zohar Manna this year but
is interested in changing, and I have too many people to interact with
now as well as no desk space. His expenses are being paid by the
"People's Republic". He seems reasonable, but Zohar would know more.
His message to me follows. He can be reached as FENG@SU-SCORE, and
Zohar is ZM@SU-AI.
∂24-Apr-84 1441 FENG@SU-SCORE.ARPA Re: change of program
Received: from SU-SCORE.ARPA by SU-AI.ARPA with TCP; 24 Apr 84 14:40:55 PST
Date: Tue 24 Apr 84 14:36:58-PST
From: Yu-lin Feng <FENG@SU-SCORE.ARPA>
Subject: Re: change of program
To: JMC@SU-AI.ARPA
In-Reply-To: Message from "John McCarthy <JMC@SU-AI.ARPA>" of Tue 24 Apr 84 13:09:00-PST
I am very thankful if you could telephone Prof. Boyer and Moore and recommend
me involving a new research topic. During the past year I listened your
courses and lectures and I am interested in the LISP proving very much.
I have read Boyer & Moore's book "A computational logic". I very apreciate
their "Theorem prover". Also, when I was in China, I did much work on
program verification and theorem proving.
If Prof. Boyer and Moore agree with the arrangement, please let me know.
Thank you very much for your kind help.
--Yulin
-------
∂24-Apr-84 2325 JMC re: video displays (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - With regard to Stolfi's remarks, I didn't mention any trickle down
theory, nor was I comparing lives and money in value, although I think
this is sometimes appropriate. My remark was that when excessive money
is spent on safety, net lives are lost, because money randomly going
into the economy saves lives at an estimated cost of $2.5 million per
life. I agree with Stolfi that building pyramids is unlikely to save
lives at all, and have heard that improving paramedic services is the
most cost-effective way to save lives for money in the U.S. Reducing
utility rates or industrial prices or raising wages by not requiring
non cost-effective safety measures ought to be at least as good as
average, but I don't claim that estimates formed by simple regression
are accurate.
∂24-Apr-84 2328 JMC
To: stolfi.pa@XEROX.ARPA
jmc - With regard to Stolfi's remarks, I didn't mention any trickle down
theory, nor was I comparing lives and money in value, although I think
this is sometimes appropriate. My remark was that when excessive money
is spent on safety, net lives are lost, because money randomly going
into the economy saves lives at an estimated cost of $2.5 million per
life. I agree with Stolfi that building pyramids is unlikely to save
lives at all, and have heard that improving paramedic services is the
most cost-effective way to save lives for money in the U.S. Reducing
utility rates or industrial prices or raising wages by not requiring
non cost-effective safety measures ought to be at least as good as
average, but I don't claim that estimates formed by simple regression
are accurate.
∂25-Apr-84 0049 JMC Where do Katz and Chomsky leave AI?
To: csli-friends@SU-CSLI.ARPA
I missed the April 19 TINLUNCH, but the reading raised
some questions I have been thinking about. Also I apologize
for violating the custom of mailing only adminstrative
communications to the CSLI mailing lists. If this communication
seems inappropriate, perhaps another mechanism can be devised
to make our electronic facilities available for substantive
discussion.
Reading "An Outline of Platonist Grammar" by Katz leaves
me out in the cold. Namely, theories of language suggested by
AI seem to be neither Platonist in his sense nor conceptualist
in the sense he ascribes to Chomsky. The views I have seen and
heard expressed by Chomskyans similarly leave me puzzled.
Suppose we look at language from the point of view of
design. We intend to build some robots, and to do their jobs
they will have to communicate with one another. We suppose
that two robots that have learned from their experience for
twenty years are to be able to communicate when they meet.
What kind of a language shall we give them.
It seems that it isn't easy to design a useful language
for these robots, and that such a language will have to satisfy
a number of constraints if it is to work correctly. Our idea
is that the characteristics of human language are also determined
by such constraints, and linguists should attempt to discover them.
They aren't psychological in any simple sense, because they will
apply regardless of whether the communicators are made of meat or silicon.
Where do these constraints come from?
Each communicator is in its own epistemological situation.
For example, it has perceived certain objects. Their images
and the internal descriptions of the objects inferred from these
images occupy certain locations in its memory. It refers to them
internally by pointers to these locations. However, these locations
will be meaningless to another robot even of identical design, because
the robots view the scene from different angles. Therefore, a robot
communicating with another robot, just like a human communicating
with another human, must generate and transmit descriptions in
some language that is public in the robot community. The language
of these descriptions must be flexible enough so that a robot can
make them just detailed enough to avoid ambiguity in the given
situation. If the robot is making descriptions that are intended
to be read by robots not present in the situations, the descriptions
are subject to different constraints.
Consider the division of certain words into adjectives
and nouns in natural languages. From a certain logical point
of view this division is superfluous, because both kinds of
words can be regarded as predicates. However, this logical
point of fails to take into account the actual epistemological
situation. This situation may be that usually an object is
appropriately distinguished by a noun and only later qualified
by an adjective. Thus we say "brown dog" rather than "canine brownity".
Perhaps we do this, because it is convenient to associate
many facts with such concepts as "dog" and the expected behavior
is associated with such concepts, whereas few useful facts would
be associated with "brownity" which is useful mainly to distinguish
one object of a given primary kind from another.
This minitheory may be true or not, but if the world has
the suggested characteristics, it would be applicable to both
humans and robots. It wouldn't be Platonic, because it depends
on empirical characteristics of our world. It wouldn't be
psychological, at least in the sense that I get from Katz's
examples and those I have seen cited by the Chomskyans, because
it has nothing to do with the biological properties of humans.
It is rather independent of whether it is built-in or learned.
If it is necessary for effective communication to divide
predicates into classes, approximately corresponding to nouns
and adjectives, then either nature has to evolve it or experience
has to teach it, but it will be in natural language either way,
and we'll have to build it in to artificial languages if the
robots are to work well.
From the AI point of view, the functional constraints
on language are obviously crucial. To build robots that
communicate with each other, we must decide what linguistic
characteristics are required by what has to be communicated
and what knowledge the robots can be expected to have. It
seems unfortunate that the issue seems not to have been of
recent interest to linguists.
Is it perhaps some kind of long since abandoned
nineteenth century unscientific approach?
∂25-Apr-84 0106 JMC Where do Katz and Chomsky leave AI?
To: friends@SU-CSLI.ARPA
I missed the April 19 TINLUNCH, but the reading raised
some questions I have been thinking about. Also I apologize
for violating the custom of mailing only adminstrative
communications to the CSLI mailing lists. If this communication
seems inappropriate, perhaps another mechanism can be devised
to make our electronic facilities available for substantive
discussion, e.g. a CSLI scientific BBOARD.
Reading "An Outline of Platonist Grammar" by Katz leaves
me out in the cold. Namely, theories of language suggested by
AI seem to be neither Platonist in his sense nor conceptualist
in the sense he ascribes to Chomsky. The views I have seen and
heard expressed by Chomskyans similarly leave me puzzled.
Suppose we look at language from the point of view of
design. We intend to build some robots, and to do their jobs
they will have to communicate with one another. We suppose
that two robots that have learned from their experience for
twenty years are to be able to communicate when they meet.
What kind of a language shall we give them.
It seems that it isn't easy to design a useful language
for these robots, and that such a language will have to satisfy
a number of constraints if it is to work correctly. Our idea
is that the characteristics of human language are also determined
by such constraints, and linguists should attempt to discover them.
They aren't psychological in any simple sense, because they will
apply regardless of whether the communicators are made of meat or silicon.
Where do these constraints come from?
Each communicator is in its own epistemological situation.
For example, it has perceived certain objects. Their images
and the internal descriptions of the objects inferred from these
images occupy certain locations in its memory. It refers to them
internally by pointers to these locations. However, these locations
will be meaningless to another robot even of identical design, because
the robots view the scene from different angles. Therefore, a robot
communicating with another robot, just like a human communicating
with another human, must generate and transmit descriptions in
some language that is public in the robot community. The language
of these descriptions must be flexible enough so that a robot can
make them just detailed enough to avoid ambiguity in the given
situation. If the robot is making descriptions that are intended
to be read by robots not present in the situations, the descriptions
are subject to different constraints.
Consider the division of certain words into adjectives
and nouns in natural languages. From a certain logical point
of view this division is superfluous, because both kinds of
words can be regarded as predicates. However, this logical
point of fails to take into account the actual epistemological
situation. This situation may be that usually an object is
appropriately distinguished by a noun and only later qualified
by an adjective. Thus we say "brown dog" rather than "canine brownity".
Perhaps we do this, because it is convenient to associate
many facts with such concepts as "dog" and the expected behavior
is associated with such concepts, whereas few useful facts would
be associated with "brownity" which is useful mainly to distinguish
one object of a given primary kind from another.
This minitheory may be true or not, but if the world has
the suggested characteristics, it would be applicable to both
humans and robots. It wouldn't be Platonic, because it depends
on empirical characteristics of our world. It wouldn't be
psychological, at least in the sense that I get from Katz's
examples and those I have seen cited by the Chomskyans, because
it has nothing to do with the biological properties of humans.
It is rather independent of whether it is built-in or learned.
If it is necessary for effective communication to divide
predicates into classes, approximately corresponding to nouns
and adjectives, then either nature has to evolve it or experience
has to teach it, but it will be in natural language either way,
and we'll have to build it in to artificial languages if the
robots are to work well.
From the AI point of view, the functional constraints
on language are obviously crucial. To build robots that
communicate with each other, we must decide what linguistic
characteristics are required by what has to be communicated
and what knowledge the robots can be expected to have. It
seems unfortunate that the issue seems not to have been of
recent interest to linguists.
Is it perhaps some kind of long since abandoned
nineteenth century unscientific approach?
∂25-Apr-84 0130 JMC re: The Conspiracy in Military Funding (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - I think my views are being distorted. For example, I believe the
Defense Department understands that support of basic research in AI
(a phrase I have always included in my proposals to them) indeed helps
defend the country by providing a basis for future applied projects.
and increasing the probability that if a technological surprise occurs
from AI, we will be the surprisers rather than the surprisees.
However, the elliptical style of this contribution makes it difficult
to determine (for purposes of discussion) what views its authors are
actually advancing.
∂25-Apr-84 0136 JMC The situation in logic
To: barwise@SU-CSLI.ARPA
I have just read that CSLI report and find myself in disagreement
with the views expressed in the first section, but I note that
they are perhaps elaborated in the BBS article referred to. Could
you send me a copy? The basis of my disagreement (though not an
argument for it) is the AI or design stance. Have you considered
what form of language a robot should be designed to use for
internal representation or communication with other robots?
To me logic seems fine for both. Also has there been other
written disagreement with your views?
∂25-Apr-84 1041 JMC
To: barwise@SU-CSLI.ARPA
Unfortunately, I can't come at 1, because Raymond Reiter is speaking at
1 to our non-monotonic reasoning seminar. I did not wish to presuppose
that humans have a mentalese, although I must admit that all my thinking
about a robot's internal representation involved a mentalese for them.
∂25-Apr-84 1753 JMC message from David Etherington
To: grosof@SUMEX-AIM.ARPA
∂19-Apr-84 1950 reiter%ubc.csnet@csnet-relay.arpa message from David Etherington
Received: from CSNET-RELAY.ARPA by SU-AI.ARPA with TCP; 19 Apr 84 19:50:28 PST
Received: From ubc.csnet by csnet-relay; 19 Apr 84 21:33 EST
Received: from ubc-vision.UUCP (ubc-vision.ARPA) by ubc-ean.UUCP (4.12/3.14)
id AA09158; Thu, 19 Apr 84 16:00:19 pst
Date: Thu, 19 Apr 84 16:00:17 pst
From: Ray Reiter <reiter%ubc.csnet@csnet-relay.arpa>
Message-Id: <8404200000.AA18209@ubc-vision.UUCP>
Received: by ubc-vision.UUCP (4.12/3.14)
id AA18209; Thu, 19 Apr 84 16:00:17 pst
To: JMC@su-ai.arpa
Subject: message from David Etherington
I forgot about a commitment during May:
May 17 P.M. to May 21 A.M., I have
in-laws visiting. If no other time
is suitable, I can come then, but
I'd like to be here for as much
of that time as possible. I hope
that this doesn't create problems.
Thanx,
David Etherington.
∂25-Apr-84 1822 JMC thanks
To: rwg%spa-nimbus@MIT-MC.ARPA
thanks
∂26-Apr-84 1134 JMC
To: DFH
Please tell her I'll take Mt. Kisco and ask her how to get there
from the Taconic Parkway.
∂26-Apr-84 1135 JMC
To: DFH
P.S. That's John Cocke.
∂26-Apr-84 1136 JMC
To: mullen@SUMEX-AIM.ARPA
An overhead projector will suffice.
∂26-Apr-84 1314 JMC 1984
To: ALS
04-25 Wed. Reiter in non-monotonic seminar
04-26 Thurs. 5pm, Gene Lawler 57 642-4019, (277 Corey Hall, 5:10pm), 591 Evans
parking permit at E. Gate, NE corner of Campus.
Social consequences of the design of interactive programs.
04-27 Friday, Siglunch, Some expert systems need common sense
7-9pm, AAAI prog. com. reception, Holiday Inn
04-28 Sat. dinner 7pm, Gatehouse
04-29 Leave for UCLA - see April 30 for schedule
04-30 thru 05-04, Fundamentals of Artificial Intelligence
UCLA. Lv./ret from San Jose on Western
Lv Sun.,Apr 29 6:55 p.m.; Ret Fri., May 4 6:20 p.m., arriv SJ 7:25
05-01 DEADLINE for abstract for non-monotonic conference
2:30 Senior Faculty
05-06 NYC, Pan Am. Lv SFO 9 am, Ar NYC 5:15 pm. (Returning 5/11)
talk on AI at Senior Center
Many thanks, Art, for replacing me. They said it would be simpler if I
endorsed their check to you, and I will. Here are the details, and Mr.
Baker will phone you.
05-07 Senior Center, Palo Alto, 450 Bryant St. 327-2811, Derwood Baker
2 to 3pm
05-07 week thereof, Yorktown, John Cocke
05-09 Sigma Xi initiation, Ken Arrow speaker
05-11 Lv. NYC 4 pm, Ar SFO 6:50 pm.
05-14 Monday, lunch with Knuth, Faculty Club
05-16 Wed. III Board
05-18 Fri. MTC qual for Martin Abadi
05-19 and 20, Comprehensive, AI on Sunday, grades due Tuesday
05-21 Monday, grading comp.
05-23 Wed. Inference Board
06-05 Tues., 2:30 Senior Faculty
06-11 thru 15, W.G. 2.2, Endicott House, M.I.T., Jack Dennis 617 253-6856
Nonimperative Languages and implementation concepts
Modifying programs without reading them. Perhaps also CBCL.
06-25 thru 29, Paris and Marseilles, tentative
06-27 to 29 Logic conference in France
07-02 to 6 July, Logic Programming in Uppsala, panel invite from Sten-Ake
08-06 Mon to 10 fri, AAAI, Austin, technical program, 7-10
08-15 Wed. III Board
08-20 Inference show for NASA Houston
10-17 10-19, Wed.-Fri., Conference on Non-monotonic reasoning, New Paltz, N.Y.
11-21 Wed. III Board
01-12 and 13, sat and sun, 1985, ASL in Anaheim, one hour lecture
∂26-Apr-84 1314 JMC talk on AI at Senior Center
To: ALS
Many thanks, Art, for replacing me. They said it would be simpler if I
endorsed their check to you, and I will. Here are the details, and Mr.
Baker will phone you.
05-07 Senior Center, Palo Alto, 450 Bryant St. 327-2811, Derwood Baker
2 to 3pm
∂26-Apr-84 1539 JMC letter to sato
To: DFH
sato.3[let,jmc] is such a letter. Please put in the amount of money,
show it to Betty, and have it ready for me tomorrow.
∂26-Apr-84 2251 JMC
To: ZM, MA
05-17 Thu. MTC qual for Martin Abadi
∂26-Apr-84 2304 JMC re: The Conspiracy in Military Funding (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - The problem with an argument put in an ironical form is that it
cannot be answered except by putting words in the mouths of the
arguers, i.e. you have restate the point you suppose your ironical
opponents are making. They can the always say you got it wrong.
That's why some of us found the Hochschild-Fine opus rather frustrating,
especially when it referred to our own views and activities by name.
How about a non-ironic version.
∂27-Apr-84 0948 JMC concurrent planning workshop
To: georgeff@SRI-AI.ARPA
Thanks for inviting me to speak at your concurrent action workshop.
Assuming no problem with the date, I would very much like to do so
and hear what others have to say. You correctly remember that I
said I had some ideas about how to formalize rules giving the
consequences of action when other events are taking place concurrently.
We'll see if I really do.
Incidentally, if the workshop needs money, AAAI has a program of sponsoring
workshops with up to $5,000 apiece. Such money is most commonly used
for the transport of the impecunious.
∂27-Apr-84 1538 JMC
To: reiter@RUTGERS.ARPA
What is the original reference for "unique names hypothesis".
∂27-Apr-84 1541 JMC AI memos
To: JK
What is the status of "publishing" your report on types, your paper
with Joe and the EKL manual itself as CSD reports (AI memos)?
∂27-Apr-84 1556 JMC reprints
To: DFH
Please send copies of "Some expert systems need common sense" and
"Coloring maps and the Kowalski doctrine" to David and Gregory
Chudnovsky at an address given in phon[1,jmc].
If it turns out we have sent these papers already, then don't.
∂27-Apr-84 1800 JMC Steiner
To: stan@SRI-AI.ARPA
I could meet him Friday only in the evening since I will be in L.A.
till then. If feasible, 8pm in my office would be ok. MAIL me
a message if this works. If he comes, he should phone my office
7-4430 from the Stanford phone 50 feet S of the door to be let in.
Otherwise, he can tailgate.
∂27-Apr-84 2248 JMC
To: ostrom@SU-CSLI.ARPA
Eric:
A few days ago I sent a message to FRIENDS@CSLI. I have been
receiving copies of it back since as undeliverable, the largest
number from UCB.KIM. I have deleted more than 20. However, just now
I received the following which I don't understand at all. I sent
no message today.
∂27-Apr-84 2241 Mailer@SU-CSLI.ARPA Message of 27-Apr-84 19:02:20
Received: from SU-CSLI.ARPA by SU-AI.ARPA with TCP; 27 Apr 84 22:41:20 PST
Date: Fri 27 Apr 84 22:40:00-PST
From: The Mailer Daemon <Mailer@SU-CSLI.ARPA>
To: JMC@SU-AI.ARPA
Subject: Message of 27-Apr-84 19:02:20
Message undelivered after 0 days -- will try for another 0 days:
*<Michele>folk.txt@SU-CSLI.ARPA: File not found
Appelt@SRI-AI.ARPA: Cannot connect to host.
BMoore@SRI-AI.ARPA: Cannot connect to host.
Bratman@SRI-AI.ARPA: Cannot connect to host.
Chappell@SRI-AI.ARPA: Cannot connect to host.
Georgeff@SRI-AI.ARPA: Cannot connect to host.
Goguen@SRI-AI.ARPA: Cannot connect to host.
Hans@SRI-AI.ARPA: Cannot connect to host.
Hobbs@SRI-AI.ARPA: Cannot connect to host.
Kells@SRI-AI.ARPA: Cannot connect to host.
Konolige@SRI-AI.ARPA: Cannot connect to host.
Lauri@SRI-AI.ARPA: Cannot connect to host.
Meseguer@SRI-AI.ARPA: Cannot connect to host.
PCohen@SRI-AI.ARPA: Cannot connect to host.
Pereira@SRI-AI.ARPA: Cannot connect to host.
Rperrault@SRI-AI.ARPA: Cannot connect to host.
Shieber@SRI-AI.ARPA: Cannot connect to host.
Stan@SRI-AI.ARPA: Cannot connect to host.
Stickel@SRI-AI.ARPA: Cannot connect to host.
Stucky@SRI-AI.ARPA: Cannot connect to host.
Tyson@SRI-AI.ARPA: Cannot connect to host.
*<bboard>csli.txt@SRI-AI.ARPA: 500 Syntax error or field too long: RCPT TO:<*<bboard>csli.txt@SRI-AI.ARPA>
Archbold@SRI-AI.ARPA: Cannot connect to host.
Bear@SRI-AI.ARPA: Cannot connect to host.
Croft@SRI-AI.ARPA: Cannot connect to host.
Derek@SRI-AI.ARPA: Cannot connect to host.
Dymetman@SRI-AI.ARPA: Cannot connect to host.
Fischler@SRI-AI.ARPA: Cannot connect to host.
Georgeff@SRI-AI.ARPA: Cannot connect to host.
JRobinson@SRI-AI.ARPA: Cannot connect to host.
Kuntz@SRI-AI.ARPA: Cannot connect to host.
Macken@SRI-AI.ARPA: Cannot connect to host.
Nilsson@SRI-AI.ARPA: Cannot connect to host.
Pednault@SRI-AI.ARPA: Cannot connect to host.
Pentland@SRI-AI.ARPA: Cannot connect to host.
PMartin@SRI-AI.ARPA: Cannot connect to host.
Rcooper@SRI-AI.ARPA: Cannot connect to host.
Rich@SRI-AI.ARPA: Cannot connect to host.
Waldinger@SRI-AI.ARPA: Cannot connect to host.
Walker@SRI-AI.ARPA: Cannot connect to host.
Wilkins@SRI-AI.ARPA: Cannot connect to host.
Bresnan@PARC-GW.ARPA: Timed out waiting for SMTP reply
Briansmith@PARC-GW.ARPA: Timed out waiting for SMTP reply
Christman@PARC-GW.ARPA: Timed out waiting for SMTP reply
DesRivieres@PARC-GW.ARPA: Timed out waiting for SMTP reply
Ford@PARC-GW.ARPA: Timed out waiting for SMTP reply
Halvorsen@PARC-GW.ARPA: Timed out waiting for SMTP reply
Kay@PARC-GW.ARPA: Timed out waiting for SMTP reply
Kiparsky@PARC-GW.ARPA: Timed out waiting for SMTP reply
Wallace@PARC-GW.ARPA: Timed out waiting for SMTP reply
Withgott@PARC-GW.ARPA: Timed out waiting for SMTP reply
ABell@PARC-GW.ARPA: Timed out waiting for SMTP reply
AHenderson.pa@PARC-GW.ARPA: Timed out waiting for SMTP reply
Blomberg.pa@PARC-GW.ARPA: Timed out waiting for SMTP reply
Bobrow@PARC-GW.ARPA: Timed out waiting for SMTP reply
Brown@PARC-GW.ARPA: Timed out waiting for SMTP reply
card@PARC-GW.ARPA: Timed out waiting for SMTP reply
Ckiparsky@PARC-GW.ARPA: Timed out waiting for SMTP reply
Deutsch@PARC-GW.ARPA: Timed out waiting for SMTP reply
Ellis@PARC-GW.ARPA: Timed out waiting for SMTP reply
Guibert@PARC-GW.ARPA: Timed out waiting for SMTP reply
JLarson@PARC-GW.ARPA: Timed out waiting for SMTP reply
Jordan@PARC-GW.ARPA: Timed out waiting for SMTP reply
JRoberts@PARC-GW.ARPA: Timed out waiting for SMTP reply
Mittal@PARC-GW.ARPA: Timed out waiting for SMTP reply
Murage.PA@PARC-GW.ARPA: Timed out waiting for SMTP reply
Nuyens@PARC-GW.ARPA: Timed out waiting for SMTP reply
Ritchie@PARC-GW.ARPA: Timed out waiting for SMTP reply
Roach@PARC-GW.ARPA: Timed out waiting for SMTP reply
Rosenberg@PARC-GW.ARPA: Timed out waiting for SMTP reply
Rypa.pa@PARC-GW.ARPA: Timed out waiting for SMTP reply
Spinrad@PARC-GW.ARPA: Timed out waiting for SMTP reply
Stefik@PARC-GW.ARPA: Timed out waiting for SMTP reply
Suchman.pa@PARC-GW.ARPA: Timed out waiting for SMTP reply
Pollard%hp-hulk@CSNET-RELAY.ARPA: Cannot connect to host.
Pullum%hp-hulk@CSNET-RELAY.ARPA: Cannot connect to host.
csli-hp.hp-hulk@CSNET-RELAY.ARPA: Cannot connect to host.
Adrian.ibm-sj@CSNET-RELAY.ARPA: Cannot connect to host.
Halpern.IBM-SJ@CSNET-RELAY.ARPA: Cannot connect to host.
HY%sri-tsc@CSNET-RELAY.ARPA: Cannot connect to host.
Jakob.ibm-sj@CSNET-RELAY.ARPA: Cannot connect to host.
Loebner%hp-hulk@CSNET-RELAY.ARPA: Cannot connect to host.
Paulo.ct@CSNET-RELAY.ARPA: Cannot connect to host.
PHS.YKTVMV.IBM@CSNET-RELAY.ARPA: Cannot connect to host.
Tekchips!brianp%tektronix@CSNET-RELAY.ARPA: Cannot connect to host.
Trow.pa%parc-maxc@CSNET-RELAY.ARPA: Cannot connect to host.
Vanlehn%PARC@CSNET-RELAY.ARPA: Cannot connect to host.
wimmers.ibm-sj@CSNET-RELAY.ARPA: Cannot connect to host.
csli-friends@UCBKIM.ARPA: Cannot connect to host.
Brown@KESTREL.ARPA: Cannot connect to host.
Pressburger@KESTREL.ARPA: Cannot connect to host.
crow@SRI-CSL.ARPA: Cannot connect to host.
Joan@SRI-CSL.ARPA: Cannot connect to host.
Linde@SRI-CSL.ARPA: Cannot connect to host.
DIsrael@BBNG.ARPA: Connection closed unexpectedly
Engelbart@OFFICE-2.ARPA: Cannot connect to host.
Kirk.tym@OFFICE-2.ARPA: Cannot connect to host.
WBD.TYM@OFFICE-2.ARPA: Cannot connect to host.
greep@SU-DSN.ARPA: Cannot connect to host.
JAR@SRI-UNIX.ARPA: Cannot connect to host.
Moran@SRI-TSC.ARPA: Cannot connect to host.
------------
Received: from SU-AI.ARPA by SU-CSLI.ARPA with TCP; Wed 25 Apr 84 01:02:14-PST
Date: 25 Apr 84 0106 PST
From: John McCarthy <JMC@SU-AI.ARPA>
Subject: Where do Katz and Chomsky leave AI?
To: friends@SU-CSLI.ARPA
-------
∂27-Apr-84 2301 JMC re: Logo installed on Score (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - "Logorrhea" starts with "logo".
∂27-Apr-84 2338 JMC
To: emma@SU-CSLI.ARPA
Emma, your TGIF announcement mailed at 1649 arrived at SAIL at 2159,
coming via SCORE. I suggest you mail earlier and/or telnet to SAIL,
for example, and mail * and send *. That will arrive right away.
∂28-Apr-84 0150 JMC paper to Reiter
To: DFH
Please mail Reiter at UBC a copy of the "Applications of circumscription ... "
paper with a note saying that it is submitted to the non-monotonic logic
conference.
∂28-Apr-84 0840 JMC Returned mail: User unknown
To: ostrom@SU-CSLI.ARPA
I just got 12 messages like this.
∂28-Apr-84 0834 MAILER-DAEMON@csnet-relay.csnet Returned mail: User unknown
Received: from CSNET-RELAY.ARPA by SU-AI.ARPA with TCP; 28 Apr 84 08:34:24 PST
Received: From hp-labs.csnet by csnet-relay; 28 Apr 84 11:03 EST
Date: 25 Apr 84 0106 PST
From: Mail Delivery Subsystem <MAILER-DAEMON%hp-labs.csnet@csnet-relay.arpa>
Received: by HP-VENUS id AA04313; Sat, 28 Apr 84 07:41:59 pst
Subject: Returned mail: User unknown
Message-Id: <8404281541.AA04313@HP-VENUS>
To: John McCarthy <JMC%su-ai.arpa@csnet-relay.arpa>
Source-Info: From (or Sender) name not authenticated.
----- Transcript of session follows -----
550 filman... User unknown
AA04288: to=csli-thor@thor, delay=00:00:17, stat=Sent
AA04288: to=csli-hulk@hulk, delay=00:00:21, stat=Sent
AA04288: to=pollard@hulk, delay=00:00:26, stat=Sent
AA04288: to=pullum@hulk, delay=00:00:29, stat=Sent
AA04288: to=loebner@hulk, delay=00:00:31, stat=Sent
AA04313: message-id=<8404281541.AA04313@HP-VENUS>
AA04313: from=MAILER-DAEMON, size=0, class=0
----- Unsent message follows -----
Received: by HP-VENUS id AA04288; Sat, 28 Apr 84 07:41:59 pst
Mail-From: HP-VENUS received 28-Apr-84 07:40:34
Date: 25 Apr 84 0106 PST
From: John McCarthy <JMC%su-ai.arpa@csnet-relay.csnet>
Received: by HP-VENUS via CSNET; 28 Apr 1984 07:41:44-PST (Sat)
Received: From 36.18.0.93.arpa by csnet-relay; 28 Apr 84 1:04 EST
Received: from SU-AI.ARPA by SU-CSLI.ARPA with TCP; Wed 25 Apr 84 01:02:14-PST
To: friends%su-csli.arpa@csnet-relay.csnet
Via: UDel; 28 Apr 84 7:41-PDT
Subject: Where do Katz and Chomsky leave AI?
Message-Id: <452014905.4267.hplabs@HP-VENUS>
I missed the April 19 TINLUNCH, but the reading raised
some questions I have been thinking about. Also I apologize
for violating the custom of mailing only adminstrative
communications to the CSLI mailing lists. If this communication
seems inappropriate, perhaps another mechanism can be devised
to make our electronic facilities available for substantive
discussion, e.g. a CSLI scientific BBOARD.
Reading "An Outline of Platonist Grammar" by Katz leaves
me out in the cold. Namely, theories of language suggested by
AI seem to be neither Platonist in his sense nor conceptualist
in the sense he ascribes to Chomsky. The views I have seen and
heard expressed by Chomskyans similarly leave me puzzled.
Suppose we look at language from the point of view of
design. We intend to build some robots, and to do their jobs
they will have to communicate with one another. We suppose
that two robots that have learned from their experience for
twenty years are to be able to communicate when they meet.
What kind of a language shall we give them.
It seems that it isn't easy to design a useful language
for these robots, and that such a language will have to satisfy
a number of constraints if it is to work correctly. Our idea
is that the characteristics of human language are also determined
by such constraints, and linguists should attempt to discover them.
They aren't psychological in any simple sense, because they will
apply regardless of whether the communicators are made of meat or silicon.
Where do these constraints come fro≠?
Each communicator is in its own epistemological situation.
For example, it has perceived certain objects. Their images
and the internal descriptions of the objects inferred from these
images occupy certain locations in its memory. It refers to them
internally by pointers to these locations. However, these locations
will be meaningless to another robot even of identical design, because
the robots view the scene from different angles. Therefore, a robot
communicating with another robot, just like a human communicating
***Error on net connection***
=== csnet-relay netread error from 36.18.0.93 at Sat Apr 28 01:08:48 ===
∂28-Apr-84 1156 JMC
To: barwise@SU-CSLI.ARPA
The address I would try from SAIL is "minker%umcp-cs"@csnet-relay, but it
seems to me that there is a misprint in the address minker@umct-cs you
mention in your message, so therefore you should try minker@umcp-cs.
umcp-cs stands for University of Maryland, College Park, Computer Science.
∂29-Apr-84 0015 JMC
To: genesereth@SUMEX-AIM.ARPA
That's western at 6:55pm.
∂29-Apr-84 0030 JMC thanks for comments
To: grosof@SUMEX-AIM.ARPA
Thanks for your comments. I'll get to them tomorrow. I have finished
my proposal, so I'll probably be in the office and a mood for discussion
in early afternoon.
∂29-Apr-84 1340 JMC
To: su-bboards@SU-AI.ARPA
a010 0057 29 Apr 84
BC-Diablo Canyon,530
California's Newest Nuclear Plant Begins Operation
By LEE SIEGEL
AP Science Writer
SAN LUIS OBISPO, Calif. (AP) - Billions over budget and years late,
the Diablo Canyon nuclear power plant started operation for the first
time early Sunday, about 12 hours before protesters planned another
march and rally against the controversial facility.
A chain reaction began within the Unit 1 reactor at California's
newest nuclear plant at 4:07 a.m. EDT, just after midnight local time,
utility and Nuclear Regulatory Commission officials said.
''At 12:07 a.m. today, Diablo Canyon nuclear power plant Unit 1
achieved its first self-sustaining fission chain reaction - called
criticality - marking the start of low-power testing,'' George
Sarkisian, spokesman for Pacific Gas & Electric Co., said.
''It was a very smooth, uneventful startup,'' Sarkisian added.
The process of starting up the reactor came more than two weeks
after the NRC approved a low-power testing license for the first of
two reactors at the $4.9 billion seaside plant, the last to be built
in California.
PG&E engineers proceeded very slowly before starting the fission
reaction.
''They're under the spotlight - that's why they're not rushing. ...
They're doing a thorough job,'' said Al Johnson, NRC regional
enforcement officer.
Johnson said he was in the control room prior to startup ''to see if
anybody was pressuring anybody to hurry. I saw none of that.''
Under the low-power test license granted April 13 by the NRC, the
utility will operate Unit 1 at up to 5 percent of its rated capacity.
PG&E must complete low-power trsting and receive approval from the
NRC before it can begin operating Unit 1 at full power. When Unit 1
achieves full power commercial operation, it will generate 1,084,000
kilowatts of electricity - enough to serve a city with more than 1
million residents.''
Johnson said PG&E officials told him they expect low-power testing
to take up to two months. During that period the reactor cooling
system will be checked to make sure it will respond properly during
simulated emergencies.
Construction on the seaside power plant started 15 years ago, with
Unit 1 initially set to start operation in 1976, Sarkisian said. The
facility's initial price tag was about $400 million, said Johnson and
Jess Crews, chief regional engineer for the NRC.
But costs skyrocketed and the start-up was repeatedly stalled
because of the discovery of an earthquake fault three miles offshore,
numerous design errors and other problems.
With a protest march and rally planned for Sunday by the
anti-nuclear Abalone Alliance, only a handful of demonstrators was
outside the plant's main gate Saturday, maintaining a vigil that since
Jan. 13 has resulted in 532 arrests, mainly for trespassing.
The scene contrasted sharply with previous protests, such as one in
1981 when thousands of people rallied at the plant gate.
While Abalone Alliance officials vowed earlier to continue their
battle against the nuclear plant, a tape-recorded message left at the
group's office Saturday sounded forlorn.
''All legal efforts to prevent low-power testing have failed,'' the
recording said. ''If you are not here to stop this from happening,
you are making a big mistake.''
ap-ny-04-29 0455EDT
***************
∂29-Apr-84 1349 JMC It's about time.
To: su-bboards@SU-AI.ARPA
a010 0057 29 Apr 84
BC-Diablo Canyon,530
California's Newest Nuclear Plant Begins Operation
By LEE SIEGEL
AP Science Writer
SAN LUIS OBISPO, Calif. (AP) - Billions over budget and years late,
the Diablo Canyon nuclear power plant started operation for the first
time early Sunday, about 12 hours before protesters planned another
march and rally against the controversial facility.
A chain reaction began within the Unit 1 reactor at California's
newest nuclear plant at 4:07 a.m. EDT, just after midnight local time,
utility and Nuclear Regulatory Commission officials said.
''At 12:07 a.m. today, Diablo Canyon nuclear power plant Unit 1
achieved its first self-sustaining fission chain reaction - called
criticality - marking the start of low-power testing,'' George
Sarkisian, spokesman for Pacific Gas & Electric Co., said.
''It was a very smooth, uneventful startup,'' Sarkisian added.
The process of starting up the reactor came more than two weeks
after the NRC approved a low-power testing license for the first of
two reactors at the $4.9 billion seaside plant, the last to be built
in California.
PG&E engineers proceeded very slowly before starting the fission
reaction.
''They're under the spotlight - that's why they're not rushing. ...
They're doing a thorough job,'' said Al Johnson, NRC regional
enforcement officer.
Johnson said he was in the control room prior to startup ''to see if
anybody was pressuring anybody to hurry. I saw none of that.''
Under the low-power test license granted April 13 by the NRC, the
utility will operate Unit 1 at up to 5 percent of its rated capacity.
PG&E must complete low-power trsting and receive approval from the
NRC before it can begin operating Unit 1 at full power. When Unit 1
achieves full power commercial operation, it will generate 1,084,000
kilowatts of electricity - enough to serve a city with more than 1
million residents.''
Johnson said PG&E officials told him they expect low-power testing
to take up to two months. During that period the reactor cooling
system will be checked to make sure it will respond properly during
simulated emergencies.
Construction on the seaside power plant started 15 years ago, with
Unit 1 initially set to start operation in 1976, Sarkisian said. The
facility's initial price tag was about $400 million, said Johnson and
Jess Crews, chief regional engineer for the NRC.
But costs skyrocketed and the start-up was repeatedly stalled
because of the discovery of an earthquake fault three miles offshore,
numerous design errors and other problems.
With a protest march and rally planned for Sunday by the
anti-nuclear Abalone Alliance, only a handful of demonstrators was
outside the plant's main gate Saturday, maintaining a vigil that since
Jan. 13 has resulted in 532 arrests, mainly for trespassing.
The scene contrasted sharply with previous protests, such as one in
1981 when thousands of people rallied at the plant gate.
While Abalone Alliance officials vowed earlier to continue their
battle against the nuclear plant, a tape-recorded message left at the
group's office Saturday sounded forlorn.
''All legal efforts to prevent low-power testing have failed,'' the
recording said. ''If you are not here to stop this from happening,
you are making a big mistake.''
ap-ny-04-29 0455EDT
***************
∂29-Apr-84 1400 JMC holding bboard
To: FY
Why do you find it necessary to hold bboard while editing it?
It is really only necessary to ask for it for a few seconds in
when doing αβ. or αβe or αm or αp.
∂29-Apr-84 1723 JMC
To: DFH
nsf.xgp[s84,jmc] is an improved version.
∂04-May-84 1918 JMC
To: georgeff@SRI-AI.ARPA
Apart from AAAI, I will be available except for Aug 15 and 20 (one day each).
∂04-May-84 1920 JMC
To: PHY
May 16 isn't possible for me to meet with Don. Any other day that week
is ok.
∂04-May-84 1923 JMC
To: pack@SU-SCORE.ARPA
As for the unique names proof, we can wait for the new DERIVE. Otherwise,
how about meeting me at 4pm in my office Saturday.
∂04-May-84 1929 JMC
To: BH
∂02-May-84 0613 BH LISP history question
What does the F stand for in FEXPR and FSUBR?
I didn't introduce it, and I don't know. If it's in the LISP 1.5
manual, you might try Dave Luckham, but if it didn't appear until
Maclisp try Greenblatt.
∂04-May-84 1931 JMC handwriting interpretation
To: tajnai@SU-SCORE.ARPA
Your interpretation is correct.
∂04-May-84 1939 JMC Letter
To: aaai-office@SUMEX-AIM.ARPA
Claudia, here is a corrected version. I toned down a few expressions.
My additions are in caps, but note also the deletions.
∂02-May-84 1534 AAAI-OFFICE@SUMEX-AIM.ARPA Letter
Received: from SUMEX-AIM.ARPA by SU-AI.ARPA with TCP; 2 May 84 15:34:02 PDT
Date: Wed 2 May 84 11:13:31-PDT
From: AAAI <AAAI-OFFICE@SUMEX-AIM.ARPA>
Subject: Letter
To: jmc@SU-AI.ARPA
cc: aaai-office@SUMEX-AIM.ARPA
Telephone: (415) 328-3123
Postal-Address: 445 Burgess Drive, Menlo Park, CA 94025
John,
If we want UT's Pres. Flawn to give a brief welcoming address to
the first assembly of all the conference attendees, I've been advised
we will need to write a letter to Flawn from you. I've taken
the liberty to compose such a letter (pls excuse the hipe in the
last paragraph).
Claudia
Dr. Peter T. Flawn, President
University of Texas
Austin, TX 78712
Dear Dr. Flawn:
The American Association for Artificial Intelligence is planning to
hold its fourth National Conference on Artificial Intelligence from
August 6 to 10, 1984 on the University of Texas at Austin campus.
The Departments of Computer Science and Mathematics are our gracious
co-sponsors of this year's conference.
This year we anticipated to have approximately 2,500 computer
scientists, mathematicians, computational linguists, engineers, and
government scientists attending this conference.
We cordially invite you or [SOMEONE YOU DESIGNATE] to welcome the
first assembly of all the conference attendees on Thursday, August 9,
11:00 a.m., in the Concert Hall of the Performing Arts Center.
We appreciate the hospitality of the University in sharing its
facilities for our conference. This annual conferenCe is the most
significant activity of our organization of any year. With
the GROWING interest IN artificial intelligence,
and with the HELP of the University of Texas at
Austin, we feel that the August 1984
conference in Austin will be our best effort ever.
Sincerely,
John McCarthy
President
JMC/CCM/kk
-------
∂04-May-84 1940 JMC
To: abadi@SU-SCORE.ARPA
CC: ZM@SU-AI.ARPA
I have not problem with having the MTC qual in the summer. Almost any
time will do. Let me know when you are ready to propose a date.
∂04-May-84 1945 JMC
To: CG
I'll be away till the 11th, so I'll phone the week of the 13th.
∂04-May-84 1947 JMC
To: aaai-office@SUMEX-AIM.ARPA
I'll try for a presidential message.
∂04-May-84 1949 JMC samuel
To: tajnai@SU-SCORE.ARPA
Please phone him 7-3330 and get his approval.
∂04-May-84 2011 JMC missing meeting
To: golub@SU-SCORE.ARPA
Alas, I'll be away this week.
∂04-May-84 2014 JMC
To: golub@SU-SCORE.ARPA
Make my vote a yes on Nilsson.
∂04-May-84 2017 JMC
To: elyse@SU-SCORE.ARPA
I am a member of the faculty club.
∂05-May-84 0155 JMC meeting
To: pack@SU-SCORE.ARPA
Make that 3pm if possible, so SAIL will still be up.
∂05-May-84 1329 JMC meeting
To: grosof@SUMEX-AIM.ARPA
2pm would be best, but if you are just waiting, now is ok.
∂06-May-84 0106 JMC
To: genesereth@SUMEX-AIM.ARPA
restau[s84,jmc] 25 best restaurants, 1984 May
1 Lutece, NY
2 The Four Seasons, NY
3 Le Francais, Wheeling, Ill.
4 Chez Panisse, Berkeley
5 Le Cirque, NY
6 K-Paul's Louisiana Kitchen, New Orleans
7 The Quilted Giraffe, NY
8 Le Perroquet, Chicago
9 Le Cote Basque, NY
10 Commander's Place, New Orleans
11 L'Ermitage, L.A.
12 Coach House, NY
13 La Grenouille, NY
14 Le Bec-Fin, Philadelphia
15 Michael's, Santa Monica
16 Le Lion D'Or, Washington, D.C.
17 Ma Maison, Los Angeles
18 Rex-Il Ristorante, Los Angeles
19 Spago, Los Angeles
20 Valentino, Santa Monica
21 Ernie's, San Francisco
22 Il Nido, NY
23 Felicia, NY
24 Jean-Louis, Washington
25 Parioli Romanissimo, NY
∂12-May-84 1205 JMC
To: PHY
Unfortunately, I must be in L.A. for the next two Wednesdays. May 30
is free, however.
∂12-May-84 1207 JMC
To: lenat@SUMEX-AIM.ARPA, tajnai@SU-SCORE.ARPA
Thurs AA389 Dallas/SFO 6:45pm/8:26pm
03-15 thurs. TI, Dallas, 2 talks, general CRL colloquium
physicists too, also AI group
Harry Tennant, 214 995-0337
3pm, 1 hour
Sheraton Park Central, 12720 Merit Drive, March 14
∂12-May-84 1209 JMC
To: dikran@SU-CSLI.ARPA
You have my permission, but please make it clear that you are
reprinting a net message rather than a formal piece of writing.
∂12-May-84 1212 JMC
To: gavan@MIT-MC.ARPA, jcm@MIT-MC.ARPA
It's not a revision; it's a new paper. I'll send you copies, but I'm
between secretaries at the moment.
∂12-May-84 1935 JMC Goto
To: tanaka@UTAH-20.ARPA
Richard Weyhrauch of Stanford tells me that I should communicate
with you about the desire of someone named Goto of NTT to visit
Stanford. If this is Shigeti Goto of the Musashino laboratory,
I am glad to invite him and will send a letter. If it is someone
else, I would have to no more.
∂13-May-84 1318 JMC re: Soviets boycotting Olympics. Who's responsible? (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - Why the Soviets decided to boycott the Olympics is unclear. It may
be that the organization encouraging defectors was a precipitating factor
in favor of the faction advocating the boycott. One must remember that
the Soviet line concerning defectors doesn't admit that anyone could
honestly prefer another social system or that mistreatment precipitating
defection (such as antisemitism) can exist in the Soviet Union. The line
also won't admit that groups encouraging defection can exist in the U.S.
without being instigated by the Government. Therefore, individual
defections can be explained by coercion the bribery of weak-minded
individuals, attraction to bad aspects of capitalism such as pornography
or the possibility of getting rich, or criminals defecting in order to
prevent detection of their crimes.
There is a strong isolationist faction in the Soviet Union, and it
seems to be increasingly powerful. A rather pure example is the cutting
off (I believe in 1980) of international direct dialing into the Soviet
Union which had existed since (I believe) 1976. The excuse offered at the
time was "technical problems". I offer the following exercise for the
reader. Give the explanation that would be offered by a Soviet official
when asked about this under the following circumstances.
1. When surprised by a question from an American audience.
Remember that lying about the facts is not excluded, especially if
follow-up questions can be avoided.
2. When asked by an ordinary Soviet citizen at a meeting
explaining the party line on some general issue to which this may be
apropos. Remember that intimidation of the questioner by suggesting that
the question is a "provocation" is not excluded. Explain the concept of
"provocation".
3. When asked in private by a "politically mature" individual,
e.g. in a high level party or KGB context. Suppose that the rank of the
questioner is equal to or higher than that of the person answering the
question. Remember that while some cynicism is allowed under these
circumstances, an answer that admits even implicitly that the party is an
oppressive ruling class is probably excluded.
Returning to the Olympic boycott, there is also the possibility
that the motivation is the effect on world public opinion, including
American. Harsh words between the U.S. and the Soviet Union always
increases sentiment for softening the situation by some unilateral U.S.
action.
It is also relevant that the 1988 Olympics are scheduled for South
Korea, another country posing problems for the Soviet Union. The North
Koreans could easily create a situation in which the Soviet Union would
have to boycott or be seen letting down the communist side.
It seems to me that moving the Olympics permanently to Greece is a
good idea provided the money can be found for facilities and provided the
Greek Government can be seen to operate it in a politicaly neutral way.
Greece isn't Switzerland, and its politics are aggressively socialist now
after having been aggressively semi-fascist in the recent past.
∂13-May-84 2101 JMC re: Soviets boycotting Olympics. Who's responsible? (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - I have been to the Soviet Union more than ten times, I read and
speak Russian moderately well, and I went there for the first few times
in a hopeful frame of mind. Let me assure JMM that it really is very
different. He or anyone else interested should read the books written
recently by newspaper correspondents after their return from a few
years in Moscow. They are all very informative about the nature of
Soviet society. What I learned agrees with what they say, but being
professionals, they had the opportunity to learn much more than I did.
Unlike the correspondents of the thirties and forties, they were not
blinded by hope and had an opportunity to learn from dissidents. If
the Soviet Government completes the job of wiping out the dissidents
and restricting the contacts available to correspondents, it will be
interesting to see if they get a more favorable press.
∂13-May-84 2131 JMC
To: KRANE.YKTVMX.IBM@CSNET-RELAY.ARPA
Yet another try.
∂14-May-84 0918 JMC
To: JMC
212 989-3308 391 2396
∂14-May-84 1011 JMC
To: HST
Guy Steele is travelling; he was here yesterday. Alas, Diana has left,
and so has my temporary secretary, so I have none for a while. I'll
try again on pictures.
∂14-May-84 1218 JMC
To: TW
I like Moshe Vardi, but the sign regarding space is the reverse. Namely,
CSLI has agreed to provide space for my visitor Sato. In general, I am
not available for assistance in administrative matters. When I
dismantled the AI Lab, I deliberately gave up using clout.
∂14-May-84 1220 JMC support for concurrent planning workshp
To: aaai-office@SUMEX-AIM.ARPA, pereira@SRI-AI.ARPA,
georgeff@SRI-AI.ARPA
I hereby agree to the Pereira-Georgeff request for $5000 AAAI support
for this workshop. Please let me know when you decide on a date.
Late June is bad for me.
∂14-May-84 1319 JMC
To: georgeff@SRI-AI.ARPA
The date is fine with me, and I have no opinion about location.
∂14-May-84 1411 JMC
To: schreiber@SU-SCORE.ARPA
With regard to programming projects, we have got ourselves into the
mess that might have been expected when we went from the single
programming problem to these projects. I have no suggestions
except to go back to the old way as soon as possible.
∂14-May-84 2240 JMC
To: simonds@CMU-CS-C.ARPA
CC: reddy@CMU-CS-C.ARPA
next visit
I have communicated by mail with Mike Griffiths, and we have agreed to
make my next visit June 25 thru 29. Can you make reservations for us
from Sunday June 24 thru Friday June 29 at the Bristol? How are things
going?
∂15-May-84 0138 JMC
To: HST
I have no memory of the date at all. July 1958 is plausible.
∂15-May-84 1029 JMC
To: RPG
Blasgen of IBM is out till Thursday
∂15-May-84 1040 JMC communication established
To: krane.yktvmx.ibm@CSNET-RELAY.ARPA
Yours got through also.
∂15-May-84 1145 JMC end of New York Times news service
To: su-bboards@SU-AI.ARPA
Because of the use of the NYT News Service outside Stanford, the
New York Times has terminated our use of it. Too bad.
∂15-May-84 1402 JMC Chandrasekharan workshop
To: shortliffe@SUMEX-AIM.ARPA
He has asked AAAI for $5K to support graduate student attendance at this
summer's AIM workshop. Do you support this?
∂15-May-84 1623 JMC All things come to him who waits.
To: gotelli@SU-SCORE.ARPA
The NYT letter says "cc: Profesor Bosack" at the bottom.
∂15-May-84 1705 JMC stolen typewriter
To: su-bboards@SU-AI.ARPA
Between 2pm and 3pm today, a typewriter was stolen from room 358 MJH
(Diana Hall's former office). Anyone who may have seen someone in
MJH carrying anything in which a typewriter might have been concealed
should inform Betty Scott.
∂15-May-84 1710 JMC
To: *
A typewriter was stolen from MJH 358 (Diana Hall's former office) between 2pm
and 3pm today. Did anyone see anyone carrying anything that might conceal it?
∂15-May-84 1717 JMC
To: *
The typewriter was merely borrowed, but without a note or telling Betty Scott.
∂15-May-84 1717 JMC unstolen typewriter
To: su-bboards@SU-AI.ARPA
The typewriter was merely borrowed but without leaving a note or informing
Betty Scott. Since its predecessor was stolen three months ago, we immediately
concluded it was stolen.
∂15-May-84 1901 JMC
To: ME
Channel 35 tty106 has just gone bad, e.g. an l is doubled.
∂16-May-84 0725 JMC paper for Etherington
To: grosof@SUMEX-AIM.ARPA
Yesterday I XGPed a copy of my new circumscription paper for Etherington,
but I never got it from the XGP. Could you retrieve it for him?
∂16-May-84 1633 JMC
To: wunderman@SU-CSLI.ARPA
5-17 is barely possible. 5-21 is somewhat possible. 5/22 is good.
5-23 is impossible.
∂16-May-84 1638 JMC
To: yearwood@SU-SCORE.ARPA
Monday or Tuesday morning from 9:30 to 11:45 are ok, but please mail
me messages as soon as possible about actual schedule.
∂16-May-84 1641 JMC
To: ZM
CC: MA
We are supposed to examine Martin Abadi tomorrow, but I don't have
a time. I need to keep the afternoon free, so I suggest 10am, my office
or yours.
∂16-May-84 1644 JMC
To: JJW
As a rule, TACs don't allow 1200/150, and anyway I think I want to
keep the terminal for one of the summer people, e.g. Bellin or Mason.
So unless they appear by then and don't want it, I'd prefer you
brought it in.
∂16-May-84 1736 JMC my computer account at sri-ai
To: stan@SRI-AI.ARPA
When I try to login it complains I have the wrong password.
Whom should I ask to reset my password?
∂16-May-84 1803 JMC
To: CLT
Include Oliver Selfridge in the tenpage list.
∂16-May-84 1830 JMC
To: elyse@SU-SCORE.ARPA
Please send me the equipment grants brochure.
∂16-May-84 1907 JMC re: Soviets boycotting Olympics. Who's responsible? (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - Entering the country, except for citizens, is not a right at all in
American law. Letting particular foreigners in is a matter of policy, as
it is in almost any country. Once in, whether temporarily or permanently,
a person has certain rights as long as he is in. For example, as far as I
can see, the whole Bill of Rights applies to foreigners. A gray area that
has been a source of considerable litigation concerns foreigners who claim
that their deportation, which would be in accordance with law otherwise,
is illegal because it was motivated by their exercise of rights such as
freedom of speech.
The McCarran act of about 1950 is not very sensible in many of its
aspects and has been got around in various ways. If it were interpreted
literally, it would be impossible to hold international scientific
meetings in the U.S., because certain participants would be unable to
attend. I don't recall the issue arising recently for scientific
meetings, but there is a gray area between scientific meetings and
propaganda exercises which continues to give trouble for officials and
employment to lawyers and journalists.
∂16-May-84 2202 JMC More fuel for the flames.
To: su-bboards@SU-AI.ARPA
n071 1525 16 May 84
BC-OLYMPIC(COX)
Soviets Sought Secret 'No Defections' Pledge
(Exclusive)
By ANDREW J. GLASS
c. 1984 Cox News Service
WASHINGTON - The Soviet Union had sought secret assurances from the
Reagan administration that it would return to Soviet custody any of
their athletes who sought to defect during the Summer Olympic Games
in Los Angeles.
The Soviet request was forwarded to President Reagan in late April,
while he was in China, where it was promptly rejected as morally and
politically objectionable.
Sources familiar with the Soviet position did not assert that a ''no
defectors'' pledge of the sort which Moscow had raised in discussions
with senior U.S. officials would have guaranteed Soviet participation
in the Games. But they suggested that it could have been a critical
factor, given the deeply seated concerns voiced in Moscow over
prospective security arrangements.
The sources said that shortly before the Soviet Olympic pullout was
announced on May 8, a carefully drafted letter was approved for the
president's signature that was aimed at reassuring the Soviets. The
White House plan called for the letter to be delivered to Soviet
President Konstantin Chernenko by Juan Antonio Samaranch, a Spanish
envoy who heads the International Olympic Committee.
The essence of that letter, however, had already been conveyed to
senior Soviet diplomats in Washington on April 27 by Edward J.
Derwinski, a former Illinois congressmen who is coordinating Olympic
matters for the State Department.
In his letter, Reagan vowed that the United States ''is totally
committed to upholding the (Olympic) charter and fulfilling its
responsibilities as the host nation of the Games.''
Derwinski also handed Victor F. Isakov, a Soviet diplomat, a
detailed memorandum, which stated that anti-Soviet emigre groups
would be barred from staging demonstrations near the Olympic villages
housing the athletes or at the competition sites. This pledge, in
effect, met what had been another major demand which had been raised
by the Soviets.
Soviet Olympic athletes and coaches had planned to live on a Russian
ship anchored off Long Beach during the Games. The White House had
waived several procedural rules, including maritime safety
regulations, to enable the Soviets to adopt this arrangement.
Despite all these and other concessions, including unlimited landing
rights for charter flights by Aeroflot, the Soviet state airline,
Moscow further pressed the Reagan administration to make a secret
promise that defectors would be returned.
It was further suggested that such a pledge would be self-limiting,
in that potential defectors would not dare to seek asylum - knowing
they would be handed over to the KGB, the Soviet state security
apparatus. Thus, it was argued, neither Reagan nor Chernenko need
ever have been embarrassed by the proposed arrangement.
Before the May 8 pullout, sources said, top American officials fully
understood the Soviet fears of being humiliated through
demonstrations and defections at the July Games. Accordingly, they
advised Reagan to take a conciliatory tone without, however,
acquiesing to the complete list of Soviet demands. The president,
these sources said, ultimately adopted that course.
U.S. officials had noted privately to their Soviet counterparts that
no Soviet athlete had ever defected during any post-war Olympiad in
which the Soviet Union had participated. But officials said the
Soviets seemed obsessed with the defection issue and made it the
standard by which they intended to judge Reagan's repeated private
assurances that the United States had no desire to seek any political
advantages from hosting the Games.
This view was underscored by F. Don Miller, the executive director
of the U.S. Olympic Committee. ''What (the Soviets) really want(ed)
is a promise to hand over defectors,'' he said.
The Soviet defector issue has been a sensitive one for U.S.
administrations ever since 1970 when a would-be Lithuanian defector,
Simas Kurdirka, was returned to his Soviet fishing vessel off Cape
Cod by the Coast Guard. Kudirka was subsequently granted asylum in
the United States after being held for several years in a Soviet
prison camp.
(Distributed by The New York Times News Service)
nyt-05-16-84 1823edt
***************
∂16-May-84 2220 JMC
To: MA
Sorry! I even replied to your message, agreeing, but forgot to remove
the date from my calendar. Let us know when you are ready.
∂17-May-84 0957 JMC
To: walton@SU-SCORE.ARPA
Basically, I'm not available for random requests for AI information.
Please take a message and tell them I'll call them back if I think I
can help. Also I prefer computer mail about phone messages to the
little blue slips.
∂17-May-84 1210 JMC
To: reddy@CMU-CS-C.ARPA
Here is the letter that you should write to JJSS.
Dear Jean-Jacques:
I have talked with Michael Griffiths and John McCarthy about
the state of the Electronic Library Project, and I am convinced that
it is time to go ahead with ordering the computer and hiring the
people. The official papers permitting these actions are now
needed to permit progress.
cc: John McCarthy, Michael Griffiths
∂17-May-84 1438 JMC
To: JMC
Hyatt SanJose 408 298-0300 649
∂17-May-84 1824 JMC
To: ARK
∂17-May-84 1013 WALTON@SU-SCORE.ARPA
Received: from SU-SCORE.ARPA by SU-AI.ARPA with TCP; 17 May 84 10:12:15 PDT
Date: Thu 17 May 84 10:05:32-PDT
From: Melba Walton <WALTON@SU-SCORE.ARPA>
To: JMC@SU-AI.ARPA
In-Reply-To: Message from "John McCarthy <JMC@SU-AI.ARPA>" of Thu 17 May 84 09:57:00-PDT
Stanford-Phone: (415) 497-2273
Thanks for the info re: random AI requests.
I also prefer mm to blue however it is usually to slow for such a busy
switchboard - any ideas how to "lookup" and send quicker would be a
bonanza for the receptionist.
Melba
-------
∂18-May-84 1122 JMC telephone number
To: tajnai@SU-SCORE.ARPA
Do you have a business telephone number for Oliver Selfridge who attended
the Forum as a representative of (I think) General Telephone and Electronics?
∂18-May-84 1931 JMC workshop
To: amarel@RUTGERS.ARPA
MAIL me a home phone number, and I'll call you this weekend.
My home phone is 415 857-0672, and my office number is 497-4430.
∂18-May-84 1944 JMC
To: faculty@SU-SCORE.ARPA
Hmm. Maybe we should have given Joe Traub more serious consideration.
∂19-May-84 0103 JMC re: Soviets boycotting Olympics. Who's responsible? (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - The news story that the Soviets asked the U.S. for a guarantee that
defectors would be returned suggests an explanation of the boycott.
First of all, no Soviet athletes have ever defected from an Olympics
according to the story, so fear of defections is unlikely as the
reason for the demand for the guarantee. The same issue will arise
in 1988 when the Olympics are scheduled to be in South Korea. Then
if necessary the Soviets and their satellites will boycott again.
By then it is likely that the world will be softened up for the
next step.
It may become a convention that the Olympics, and eventually
any international event, can only be held in a country that will agree
to return defectors. The present Greek leftist government
is a possibility, so that putting the Olympics permanently in Greece,
which has much to recommend it on other grounds, may fit the Soviet
plan. It is interesting to speculate on the third world reaction.
Would JMM favor India making such a guarantee if there were a proposal
to hold the Olympics in India? Does he think Mrs. Gandhi would give
such a guarantee?
The objective is to get the world to accept the Soviet view
that individuals are the property of their country and to accept the
Soviet lie that the security of athletes require that defectors be
returned.
∂19-May-84 0851 JMC
To: golub@SU-SCORE.ARPA
It was a reference to the large raises referred to in Feigenbaum's message.
∂19-May-84 1118 JMC
To: golub@SU-SCORE.ARPA
Sorry, I don't want to reopen it either.
∂19-May-84 1244 JMC new decision procedure
To: JK
Are the features yet to come required for it to do the examples done
by the old decision procedure?
∂20-May-84 1022 JMC new derive
To: JK
bird.lsp[s84,jmc] contains my bird circumscription proof. Step 13 is its
first use of derive, and it runs 15 minutes without output. It ran about
a minute before.
∂20-May-84 1808 JMC re: BBS Confiscation (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - This message may require no explanation in its original context, but
it certainly does now. What's a Sysop? Was the message obscene? Was it
a message exchange relating to drug dealing? Was it about stealing stuff
from Pacific Telephone or breaking into a Pacific Telephone system? How
did Pacific Telephone become involved?
∂20-May-84 1829 JMC
To: JMC
Kathy Faes 7-3687
∂20-May-84 2117 JMC
To: JK
Here's the next problem:
16. failed to derive this time.
5. (define flies2 |∀x.flies2 x ≡ bird x ∧ ¬ostrich x|)
6. (define ab2 |∀z.ab2 z ≡ (∃x.bird x ∧ z = aspect1 x)
∨ (∃x.ostrich x ∧ z = aspect2 x)|)
16. (der |∀x.¬ab2 aspect1 x ⊃ ¬flies2 x| (5 6))
∂20-May-84 2150 JMC
To: JK
bird.lsp now all works.
∂20-May-84 2211 JMC
To: JK
new derive is slow propositionally
bird.lsp now takes approximately 43 seconds, 25 of which is spent on
step 21 which is just the derivation of a conjunction of 5 terms
from the conjuncts. This suggests either some further
improvements in derive or the introduction of pderive specialized
to purely propositional derivations.
∂20-May-84 2227 JMC
To: JK
another peculiarity
I tried deriving the conclusion of step 21 directly (after finishing
the proof) according to the two statements commented out after
step 21. In each case, EKL said it failed to derive some statement,
which was not the statement it was asked to derive nor any of its
conjuncts. Does this indicate a bug or does EKL sometimes announce
its failure to derive some intermediate goal?
∂21-May-84 0048 JMC more slowness
To: JK
lose.lsp[s84,jmc] is adapted from equal.lsp[s84,jmc] and shows DERIVE
taking a long time when asked to show that a simply defined relation
E11 is an equivalence relation. After it ran 5 minutes without result
using the full definition, I tried it on reflexivity, symmetry and
transitivity separately. Reflexivity was immediate, symmetry took
almost two minutes, and it didn't get transitivity in 5 minutes.
∂21-May-84 1954 JMC electronic library project
To: krovetz@NLM-MCS.ARPA
The project is to make an electronic public library. It will begin with
a VAX, phone lines, and French literature without formulas or pictures
that is in the public domain. What's Abacus?
∂21-May-84 2103 JMC mazda repairs
To: elyse@SU-SCORE.ARPA
I had good luck on my RX-2 at V.W. Specialty Center, 904 Industrial Ave.
Palo Alto, 494-7676. Their telephone book ad says they work on Mazdas,
but I haven't had anything done there that is peculiar to a rotary, and
you don't mention that either. If you try them, let me know how it works
out.
∂22-May-84 1204 JMC
To: schreiber@SU-SCORE.ARPA, lamping@SU-SCORE.ARPA
I have just finished my grading, but Lamping and I haven't combined
our grades. Unfortunately, I have another meeting at noon but will
be ready in early afternoon. Sorry about that.
∂22-May-84 1734 JMC IBM 801 talk
To: RPG
∂22-May-84 1728 baskett@decwrl.ARPA IBM 801 talk
Received: from DECWRL.ARPA by SU-AI.ARPA with TCP; 22 May 84 17:26:18 PDT
Received: from acetes.ARPA by decwrl.ARPA (4.22.01/4.7.29)
id AA01520; Mon, 21 May 84 09:52:33 pdt
Received: by acetes.ARPA (4.22.01/4.7.27)
id AA00796; Mon, 21 May 84 09:53:41 pdt
From: baskett@decwrl.ARPA (Forest Baskett)
Message-Id: <8405211653.AA00796@acetes.ARPA>
Date: 21 May 1984 0953-PDT (Monday)
To: jmc@su-ai
Subject: IBM 801 talk
I took out the remarks about intra-company rivalries but I'd still
appreciate your not spreading this around since I promised George Fan
(IBM Yorktown Heights) that I wouldn't.
Forest
------
This is my report on a talk given at Stanford on April 3rd. First the
published abstract:
Albert Chang, IBM Yorktown, VLSI Seminar, Tuesday, April 3 1:15 pm
ERL 401.
Two IBM 801 Chips
Preliminary performance data and the external characteristics of two 801 type
chips will be described. The first chip is a pipelined reduced instruction
set cpu (801). Its measured performance on small kernels will be given and
an indication of the relative power of one of its instructions compared to
a 370 instruction will be discussed. The second chip is a memory management
chip. It supports very large virtual memory at low cost in terms of the real
fixed storage required for page tables. It also provides hardware support for
the locking and journalling that is commonly used in data base programming.
The way in which it is used in an experimental control program will be
described.
Now the report:
Background:
Al Chang got a Ph.D. in EE from Berkeley years ago and has been at
Yorktown Heights for a long time, I think. He started on the 801
project very near it's beginning in 1975. The best background
reading on the 801 project is "The 801 Minicomputer" by George Radin in
The IBM Journal of Research and Development, Vol. 27, No. 3, May 1983.
This article is a slightly expanded version of the paper Radin
presented at the ACM Symposium on Architectural Support for Programming
Languages and Operating Systems in March of 1982. More details are
included in the IBM Journal version.
Motivation:
It came out at the end of the talk that these chips (at least the CPU
chip and I believe the MMU chip, as well) were designed by the IBM
Office Products people in Austin, Texas, the people who brought you the
Displaywriter. (censored part) At any rate, it is widely
believed that this is the chip set that will be in the machine that
is about to be delivered to CMU as part of the big IBM-CMU program. In
addition, even the trade press is starting to figure out what is going
on and is talking about IBM coming out with a proprietary CPU in their
"next generation PC". This project has long had the code name ROMP.
The talk:
The block diagram was
----- -----
| CPU |---------| MMU |-----Memory
----- | -----
|
I/O
This presumably implies I/O instructions rather than memory mapped I/O.
Nothing further about I/O was mentioned. The CPU was described as
RISC, 32 bits, pipelined, with cycle time T. The design target for T
was 200 nanoseconds but the chips seem to run at 160 ns with selected
parts running at 150ns. The MMU supports a large virtual memory (2**40
bytes), has data base support, is pipelined, and has the same pipeline
cycle time. The memory is plain Nmos with an access time of approximately
T and a cycle time of 2T. It can be two way interleaved to reduce
interference. If it is not so interleaved, there could be substantial
speed degradation due to memory interference on successive instructions.
The CPU has 16 registers (the Yorktown Heights people think this is too
few) which are 32 bits. Instructions are either 16 bits or 32 bits.
The 16 bit instructions are of the form:
OP R1, R2
where OP is 8 bits and R1 and R2 are 4 bit register designators. The
32 bit instructions are of the form:
OP R1, R2, D
where D is a 16 bit displacement, sometimes signed, sometimes not.
This form provides 16 bit literal operands and 16 bit displacements for
the load and store instructions. It is a load/store machine in that
only the load and store instructions have storage operands. This has
the usual simplifying effect on the interrupt and virtual memory
implementation. Most instructions are one cycle (if they are not loads
or stores or branches). There are a few microcoded, multiple cycle
instructions. One pair is a load or store multiple registers and another
is load program status (load PSW in IBMese). These instructions are supposed
to help relieve the load on the single port to memory in the same way
that the 16 bit instructions do (less instruction traffic to interfere
with the data traffic). In addition, there are a "small number" of
three register operand instructions of the form:
OP R1, R2, R3
The pipeline organization was diagrammed in what looked to me like the
standard way with a register file with two output ports and two input
ports, an ALU with two input ports and one output port and at least one
bypass from ALU out to ALU in. His diagram only showed one bypass but
it left out a lot so there could be more. It's the standard scheme for
pipelining one cycle ALU operations in a register to register environment.
The extra input port on the register file is for loads.
The Load and Stores were more interesting because they were so slow: 5
cycles with address translation on and 4 cycles with it off, diagrammed
as follows:
| Load | Addr | Xlate | Memory | Reg File |
+------+-------+--------+--------+----------+
When address translation is off, the Xlate cycle is deleted. When it
is on, the CPU waits for the data in order to achieve precise
interrupts if there is a fault. When it is off, the CPU continues
until the data is required. Instruction fetch continues in either
case. The compiler is able to schedule 90% of the loads so that there
is a useful instruction between the load and the use (for the Xlate off
case) and 80% of the loads so that there are two useful instructions
between the load and the use. The memory system was not described so
this explanation could be a little simple compared to reality.
Branches take one cycle if the branch is not taken, 5 cycles if the
branch is taken. There is a "branch and execute" form in which the
instruction following the branch is executed no matter which way the
branch goes. Thus the taken path is effectively 4 cycles for this form.
The load multiple instruction takes 3+ceiling(3R/2) cycles were R is
the number of registers being loaded. The store multiple instruction
takes 3+2R cycles.
The compiler is the PL8 compiler. It mainly compiles PL8, a PL/I like
language for systems programming. It is about 100K lines of code that
compiles into 1.5 megabytes of flattened executable image. It compiles
to about five different machines, including various versions of 801's,
the 370, and the 68000. It also compiles Pascal source and they expect
to be able to compile C source soon. A table of instruction
frequencies was given for three "programs". One was a data base system
they have done that implements the Sequel interface. One was the
"system" which was their monitor and a lot of utilities. The third was
the PL8 compiler itself. These are dynamic instruction frequencies.
Instruction % Sequel System PL8
Register ops 52.9 50.4 52.2
Loads 23.7 19.6 21.6
Stores 9.3 8.9 9.6
Branches taken 9.0 15.7 11.5
Branches not taken 3.8 3.8 4.0
Load/Store multiple 1.3 1.6 1.1 (8 registers, average)
Average cycles per 2.93 3.00 2.86
instruction
Number of instructions 30M 30M 600M
Other effects such as memory interleave interference and translation
buffer misses accounted for 0.3 cycles per instruction in addition to
the above. These numbers together with the 160 ns cycle show you where
the 2.2 MIPS number that John Cocke has been giving people comes from,
I believe.
Al then presented, with some reluctance, the following simple minded
benchmark numbers, probably out of deference to the Berkeley crowd
since he was scheduled to give the talk at Berkeley the next day.
He complained about the relevance of small kernals.
Time in milliseconds
Machine Language Acker Sieve Puzzle
11/780 C (VMS) 4222 116 4625
68000 Pascal 3080 246 9200
(16 MHz)
80286 Pascal 2218 168 9138
(8 MHz)
IBM RISC -- 2400 80 3800
(6.25 MHz) (PL8) (PL8) (Pascal)
Code size in bytes
11/780 152 156 2220
68000 126 228 2940
80286 --- --- ----
IBM RISC 159 202 2454
The numbers not done at IBM (780 and 286) were taken from the relevant
latest versions in Computer Architecture News.
Then, with more enthusiasm, he gave these comparisons for seven
unidentified applications programs.
10**6 instructions executed
PL8 RISC 370
1 103.0 63.7
2 171.9 107.0
3 48.8 40.0
4 88.6 71.0
5 115.8 94.4
6 49.9 38.1
7 85.3 60.5
--- ----- -----
total 663.3 474.7
overall 370/RISC = 0.715 (dynamic)
In the total number of static instructions in these 7 programs, the
IBM RISC and the 370 were about the same. Two thirds of the dynamic
excess of IBM RISC over the 370 were said to be due to the move and
multiply subroutines. The IBM RISC numbers were gathered with a
software trace. The 370 numbers were gathered with a hardware monitor
on a 370/168 which excluded all the CP/CMS system code. These programs
were said not to be commercial - no decimal - and said not to be
scientific - no floating point - but to be representative of systems
code. The average instruction size was also given: 2.85 bytes for the
IBM RISC and 3.5 for the 370.
The MMU chip supports a page size of 2048 bytes. The addresses are
byte addresses. The CPU presents 32 bits of address to the MMU, the
top four of which are a segment ID, the other 28 being an offset within
that segment. The segment registers in the MMU are 12 bits of address
that get put on the top of the 28 bit offset for a 40 bit virtual
address. The segment registers also contain a "special" bit that
controls whether the database stuff is active or not. It also contains
a "key" bit that has something to do with storage keys. I got the
impression that Al didn't want to explain this one because it was
embarrassing. The MMU keeps a 16 x 2 two way set associative translation
look aside buffer. In the monitor they are running, they map files into
segments, one file per segment. They expect most of the files in a
data base system to be relatively small so the 2**28 bytes maximum size
shouldn't be a problem. The special bit is meant to generate
interrupts to the monitor (control program in IBMese) so that locking
and journalling can be supplied by the monitor. The support is such
that one can get an interrupt on the first reference by a transaction
to a 128 byte line of data. The PL8 language has an added "Persistent"
storage class so that PL8 variables may be associated with permanent
segments (or 128 byte lines, I think):
DCL X(100) CHAR(80) PERSISTENT
OPEN (X, 'NAME', options)
X(1) = 'anything'
COMMIT (X)
Compiled code is independent of options such as "shared", "journalled",
"locks", etc. The page table in the MMU is the same organization
as used in the System 38 and gives a reversible map between
virtual and real addresses organized as a hash table as follows. If
one has N real pages (where N = 2**k for some k) then there is a 2N
entry hash table that points to an entry in the N entry page frame
table. The mth entry in the page frame table describes real page m.
That entry contains the virtual page number, a "next" pointer (for hash
collisions), a R/W bit, a TID (transaction id), and 16 lockbits, one
for each 128 byte line of a page. This whole structure was said to take
20N space, where units were not defined but I believe were bytes. It was
never suggested how big N could be. I believe N to be at least 1024.
That would mesh with the CMU talk of a workstation with "a couple of
megabytes".
Al said that, for a TLB miss, they typically got 2.2 references in the
hash table/page frame table, i.e., the probability of no chaining was
close to 90 percent. The R/W, TIB, and Lockbits apply only if the special
bit is on in the segment register. The MMU has a TIDREG (transaction id
register). An interrupt is generated by the MMU for a load/store to line N:
if TID ~= TIDREG (not equal)
if store, R/W = R or Lockbit(N) = 0
if Load, R/W = R and Lockbit(N) = 0
At this point it was decided that the talk had gotten too low level and
that the time was up anyway so we switched to questions. The
technology questions revealed that the chips are Nmos, 2 micron feature
size, with about 7.5 mm per side. Al said it does a bit better than a
4341 in raw CPU power, giving about 1.4 370 MIPS or 2 RISC MIPS. He
also said that the version they are running in Yorktown Heights has a
bug where instructions cross a page boundary and there is a fault on
both instruction and data references so they compile around it.
Additional speculation: APA displays (IBMese for "all points
addressable) have been big inside IBM for some time now and Jim Gosling
(author of the VAX Emacs screen editor) demonstrated a really nice
WYSIWYG (what you see is what you get) editor on a Sun workstation at
the recent Sun users group meeting. It wasn't available since Jim
works for IBM on the CMU project and the Sun was just their interim
development machine. Thus one would reasonably expect the ROMP's that
CMU gets to be like Sun workstations with these chips in them.
After the talk, (censored) people muttered
about how slow these chips were and how much better one could do if one
did things "right". Comments from Al pointed out that this system was
designed to be cheap and that accounted for some of the slowness of the
memory system.
Forest
∂22-May-84 1736 JMC
To: baskett@DECWRL.ARPA
Thanks for the 801 writeup. I'll not publicize it.
∂23-May-84 2108 JMC Procurement Dept Review
To: RPG@SU-AI.ARPA, ME@SU-AI.ARPA, REG@SU-AI.ARPA,
bosack@SU-SCORE.ARPA
∂23-May-84 1420 @SU-SCORE.ARPA:RINDFLEISCH@SUMEX-AIM.ARPA Procurement Dept Review
Received: from SU-SCORE.ARPA by SU-AI.ARPA with TCP; 23 May 84 14:20:44 PDT
Received: from SUMEX-AIM.ARPA by SU-SCORE.ARPA with TCP; Wed 23 May 84 14:15:06-PDT
Date: Wed 23 May 84 13:31:40-PDT
From: T. C. Rindfleisch <Rindfleisch@SUMEX-AIM.ARPA>
Subject: Procurement Dept Review
To: Faculty@SU-SCORE.ARPA, HPP-Exec@SUMEX-AIM.ARPA, McCabe@SUMEX-AIM.ARPA,
Vian@SUMEX-AIM.ARPA,
HPP-Admin: ;,
CSD-Administration: ;
cc: Rindfleisch@SUMEX-AIM.ARPA
Bill Massy and Ken Creighton have set up a committee to review the policies
and procedures of the Procurement Department in support of university
operations. This covers all aspects of procurement including stores,
special contracts (individual purchases), hospital, commissary, and
construction. Out of frustration with the current system, I have agreed
to co-chair this group with Marv Geistlinger (Director of Procurement).
So, this is a chance to gripe. I'd like to hear about the issues that are
most troublesome in dealing with Procurement and those that work well.
Please give examples where possible. These comments should focus on chronic
policy, organizational, and procedural problems rather than on isolated
instances of screw-ups that are not part of a pattern.
Please pass this message on to others not included in the addressees who
may have constructive comments (no bulletin boards please). I would like
to receive your inputs by June 1.
Thanks, Tom R.
-------
∂24-May-84 1512 JMC re: BBS Confiscation (from SAIL's BBOARD)
To: RPG
∂24-May-84 1329 SAMUEL@SU-SCORE.ARPA re: BBS Confiscation (from SAIL's BBOARD)
Received: from SU-SCORE.ARPA by SU-AI.ARPA with TCP; 24 May 84 13:29:21 PDT
Date: Thu 24 May 84 13:26:39-PDT
From: Sam Hahn <Samuel@SU-SCORE.ARPA>
Subject: re: BBS Confiscation (from SAIL's BBOARD)
To: JMC@SU-AI.ARPA
Here's a followup message. I hope it's more informative than the previous.
-- sam hahn
24-May-84 13:09:57-PDT,1284;000000000005
Return-Path: <info-cpm-request@AMSAA.ARPA>
From: ssalzman.es@XEROX.ARPA
Date: 24 May 84 11:11:09 PDT
Subject: Update on MOG-URs BBS Confiscation
Rick Gaitley, who is sysop of Xanadu RCPM/ and a friend of Tom's,
decided to cal the phone company and find out why his system was
confiscated. Apperently, someone left an AT&T calling card number
on Tom's board and it remained there for over a month. It's very
possible that Tom had no idea of that message being there. Ths
same caling card number showed up on Xanadu after Tom left the
message there about what happened to his board. Rick immediatly
deleted it. At any rate, I can see why the phone company might get a little
ticked off. According to the phone company, the reason they took
the system (suuposedly) was to try and find the person who left
the message. The person Rick talked to is the one that actually
took Tom's machine. If anyone cares to find out any more,
the number of Ricks system (Xanadu) is (818)906-1636.
- Isaac Salzman
Ssalzman.es@Xerox
-------
∂24-May-84 1534 JMC
To: mwalker@SU-SCORE.ARPA
CC: RPG@SU-AI.ARPA
Marilyn:
I didn't realize there was any more paperwork to be done
in connection with the promotion of Dick Gabriel. I supposed, as
has been true in the past, that the Department's vote was all the
action required. However, the level of stupid bureaucracy at Stanford
is always increasing. Please transmit my protest and ill wishes to
whoever is asking for the additional paper.
The following is the document that was prepared for the faculty,
although I don't remember if it was actually distributed. Can you
manufacture the necessary paperwork from it, or do you need more? The
resume can be obtained from him.
"We propose to promote Dr. Richard P. Gabriel from Research
Associate to Senior Research Associate in the Formal Reasoning
Group. Dick has performed outstandingly both in research and
in organization. He has led our Common Lisp effort managing
our relations with Livermore and DARPA and has written good
papers that have been accepted for the next Lisp conference
on parallel processing in Lisp and on his measurements of speed
of Lisp implementations. DARPA has accepted several proposals
based on his work, and he is playing a major role in responding
to DARPA's interest in supercomputing for AI.
His resume is attached."
John McCarthy
∂24-May-84 1956 JMC re: UCLA Students Would Reelect Ronald Reagan (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - It always surprises me how much of their own preferences professors
of social science feel free to imbed in their analyses. This one can
only find selfish reasons for voting for Reagan. My impression is that
this situation has gotten worse, and that 40 years ago social scientists
were more objective.
∂25-May-84 0026 JMC re: Clarification of earlier message on BBS's (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - It seems to me that under the usual interpretations of law, in
order to prove the system proprietor guilty under the above codes, it
would have to be at least proved that he knew about this illegal use of his
computer and tolerated it. The seizure of property as evidence on the
basis of a search warrant doesn't even constitute an accusation against
the person from whom it was seized. I believe that evidence can be seized
even from the victim of a crime. It would be interesting to know whether
the telephone credit card number was that of some random individual or
belonged to some institution against which someone had a grudge.
∂25-May-84 1103 JMC
To: CLT
There was a call to say that the cost on the autos was $1177.
∂26-May-84 1844 JMC
To: JK
What is the correct reference to the notion of direct proof?
∂27-May-84 1907 JMC
To: "#GRIPES.TXT[2,2]"
When I mail to su-bboards or other-bboards, I get my mail back that was
addressed to temporarily disconnected computers or computers that
grumble about a lack of subject field. There should be some way of
avoiding this - of having the computer simply forget returned mail
addressed to bulletin boards.
∂28-May-84 1550 JMC programming project
To: schreiber@SU-SCORE.ARPA, ARK@SU-AI.ARPA
Arthur Keller's proposal that a student be allowed to substitute (with
approval) a present style programming project for the blitz test is ok
provided the faculty member is really prepared to evaluate it. I, for
example, would probably not agree to supervise any unless it were something
I knew I wanted anyway. I don't agree with Joan Feigenbaum's objection
that the blitz programming project would tie up the computers too much.
a. It's a worthy cause.
b. It wasn't too bad then.
c. We have more computer facilities now.
∂28-May-84 2110 JMC
To: MA
CC: ZM
11am, May 31 is ok with zm and me.
∂28-May-84 2113 JMC lunch time
To: DEK
I have a seminar at 1pm on Wednesday, so noon or even 11:45 would be
less rushed. However, 12:15 is ok if earlier is inconvenient.
∂29-May-84 0046 JMC
To: aaai-office@SUMEX-AIM.ARPA
presid.2[w84,jmc] Second AAAI presidential message
WE NEED BETTER STANDARDS FOR RESEARCH IN AI
The state of the art in any science includes the criteria
for evaluating research. Like every other aspect of the science,
it has to be developed. The criteria for evaluating AI research
are not in very good shape. I had intended to produce four
presidential messages during my term but have managed only two,
because this one has proved so difficult to write. It kept
threatening to grow into a paper rather than a mere expression
of opinion which is all I now know enough to write.
If we had better standards for evaluating research results
in AI the field would progress faster.
One problem we have yet to overcome might
be called the "Look, ma, no hands" syndrome. A paper reports
that a computer has been programmed to do what no computer program
has previously done, and that constitutes the report. How
science has been advanced by this work or other people
are aided in their work may be unapparent.
Some people
put the problem in moral terms and accuse others of trying to fool
the funding agencies and the public. However, there is no reason
to suppose that people in AI are less motivated than other scientists
to do good work. Indeed I have no information that the average quality
of work in AI is less than that in other fields. In my previous
message I grumbled about there being insufficient basic research,
but one of the reasons for this is the difficulty of evaluating
whether a piece of research has made basic progress.
It seems that evaluation should be based on the kind of
advance the research purports to be. I haven't been able to develop
a complete set of criteria, but here are some considerations.
1. Suppose the research constitutes making the computer solve a problem
that hasn't previously been solved by computer. Let us suppose that there
are no theoretical arguments that the methods are adequate for a class of
problems but merely a program that performs impressively on certain sample
problems together with some explanation of how the program works.
This is a difficult kind of research to explain adequately. The
reader will not easily be able to assure himself that the program is not
overly specialized to the particular example problems that have been used
in developing the program. It has often turned out that other researchers
have not been able to learn much from the paper. Sometimes a topic is so
intractable that this is the best that can be done, but maybe this means
that the topic is too intractable for the present state of the art.
2. A better result occurs when a previously unidentified
intellectual mechanism is described and shown to be either
necessary or sufficient for some class of problems. An example is
the alpha-beta heuristic for game playing. Humans use it, but it
wasn't identified by the writers of the first chess programs. It
doesn't consititute a game playing program, but it seems clearly
necessary, because without it, the number of positions that have to
be examined is sometimes the square of the number when it is used.
3. Experimental work should be repeatable.
In the older experimental sciences, e.g. physics and biology,
it is customary to repeat previous experiments in order to verify
that a phenomenon claimed to exist really does or to verify a claimed value
of an experimentally determined constant. The referees are supposed
to be sure that papers describing experimental work contain enough
of the right details so that this can be done.
Perhaps the most typical problem concerns a piece of experimental
AI research, say a PhD thesis. The general class of problems
that the researcher would like to attack is described, followed by
a description of his program and followed by a description of the
results obtained on his sample problems. Often there is only one
sample problem. The class of problems which it is claimed the
program or the methods it embodies will solve is often not stated.
The reader is free to suspect that the program has been tuned so
that it will solve the specific example described in the paper and
that the author doesn't even know whether it will solve any others.
If we aspire to testable and repeatable work in AI, then
journal authors and referees should require a statement of the
generality of the program. The referee should be able to try out
the program if language, hardware and communication facilities
permit. Moreover, the methods should be described well enough
so that someone skilled in the art can embody them in a program
of his own and test whether they are adequate for the claimed
class of programs.
Repetition of other people's experiments should be as
normal in AI as it is in the other experimental sciences. On the
whole, it should be easier in AI, because
more-or-less standard hardware and programming languages can
be used. Perhaps this is a good apprentice task for beginning
graduate students or people coming into the AI field from
the outside. Students and other newcomers will take pleasure in
trying to find a simple example that the program is supposed to
solve but doesn't.
Stating the generality of piece of work is likely to be
difficult in many cases. It is best done after the program
has solved the example problems, because the researcher
can then understand what compromises he has had to make with
generality in order to make the program do his examples.
He is most likely to make the necessary effort if he knows
that some smart student is likely to look for counterexamples
to his claims.
4. We also need criteria for formalizations
Logic based approaches to AI require that general facts
about the common sense world be expressed in languages of logic
and that reasoning principles (including non-monotonic principles)
be stated that permit determining what a robot should do given
its goals (stated in sentences), the general facts and the facts
of the particular situation. The major criteria for judging
the success of the formalization of such facts are generality
and epistemological adequacy. Generality is partly a property
of the language, and in the case of a first order language, this
means the collection of predicate and function symbols. The original
set of predicates and functions should not have to be
revised when extensions are wanted. It is also partly a property
of the set of axioms. They too should be extendable rather than
having to be changed. The recent development of non-monotonic
formalisms should make this easier.
The author of a paper proposing logical formalisms should
state, if he can, how general they are. The referee and subsequent
critics should try to verify that this is achieved.
Epistemological adequacy refers to the ability to express
the facts that a person or robot in that information system is
likely to be able to know and need to know.
5. The criteria for evaluating methods that purport to
reduce search are perhaps better established than in other fields.
Taking my own experience with game playing programs in the late
1950s and early 1960s, it was possible to demonstrate
how much alpha-beta, the killer
heuristic, and various principles for move ordering reduce search.
6. On the other hand, the evaluation of programs that purport
to understand natural language is worse off. People often simply
don't believe other people's claims to generality.
In this area I can offer two challenge problems. First, I
can provide the vocabulary (sorted alphabetically) of a certain news
story and the vocabulary of a set of questions about it. The
computational linguistic system builder can then build into his
system the ability to "understand" stories and questions involving
this vocabulary. When he is ready, I will further provide the story
and the questions. He can take the questions in natural language
or he can translate them into suitable input for his system. The
limitations of the system should be described in advance. We will
then see what questions are successfully answered and to what
extent the author of the system understood its limitations.
The second problem involves building a system that can
obtain information from databases that purport to interact with
their users in English. Again the vocabulary is given in advance,
and the system builder tunes his system for the vocabulary. It
is then tested as to whether it can answer the questions by
interacting with the database. For example, Lawrence Berkeley
Laboratory has (or had) a database of of 1970 census data. It
would be interesting to know whether a program could be written
that could determine the population of Palo Alto interacting
with the interface this database presents to naive human users.
I think both of these problems are quite hard, and whatever
groups could perform reasonably on them would deserve a lot of
credit. Perhaps this would be a good subject for a prize - either
awarded by the AAAI or someone else.
However, such challenge problems are no substitute
for scientifc criteria for evaluating research in natural
language understanding.
7. Likewise the Turing test, while a challenge problem for
AI, is not a scientific criterion for AI research. The Turing
test, suitably qualified, would be a fine sufficient criterion
for convincing skeptics that human level AI had finally been
achieved. However, we need criteria for evaluating more modest
claims that a particular intellectual mechanism has been identified
and found to be necessary or sufficient for some class of problems.
Incidentally, even as a sufficient condition, the Turing
test requires qualification. The ability to imitate a human
must stand up under challenge from a person advised by someone
who knows how the program works. Otherwise, we are in the
situation of someone watching a stage magician. We can't figure
out the trick, but there must be one. A fortiori, looking at
dialogs and figuring out which one is with a machine isn't
adequate.
∂29-May-84 1119 JMC
To: JK
%3Appendix B%1
Here is an annotated EKL proof that circumscribes the predicate
%2e(x,y)%1 discussed in section 6.
.skip 1
.begin nofill
%7
This proof uses circumscription to maximize the uniqueness of names, through
the circumscription of an equivalence relation E(X,Y), which is to be
interpreted as asserting the equivalence of the objects denoted by X and Y.
1. (AXIOM |INDEX(A)=1∧INDEX(B)=2∧INDEX(C)=3∧INDEX(D)=4|)
(label simpinfo)
Since EKL does not have attachments to determine the equivalence of names,
we establish a correspondence between the names in our domain and some
natural numbers. The label SIMPINFO on this statement and the next
indicates that these statements should be used by the EKL simplifier
whenever appropriate.
2. (AXIOM |¬1=2∧¬1=3∧¬2=3∧¬1=4∧¬2=4∧¬2=4|)
(label simpinfo)
This axiom is used for now, because EKL does not yet have enough number
theory to recognize the uniqueness of integers. We hope to eliminate it
later.
3. (DEFINE EQUIV
|∀E.EQUIV(E)≡
(∀X.E(X,X))∧(∀X Y.E(X,Y)⊃E(Y,X))∧(∀X Y Z.E(X,Y)∧E(Y,Z)⊃E(X,Z))| NIL)
Here we state the conditions necessary for E to be an equivalence relation.
It must be reflexive, symmetric, and transitive.
4. (DEFINE WORLD |∀E.WORLD(E)≡E(A,B)∧EQUIV(E)| NIL)
This axiom describes the world we are considering in this example. There
is an equivalence relation, E, which holds for two names, A and B.
5. (DEFINE WORLD1
|∀E.WORLD1(E)≡
WORLD(E)∧
(∀E1.WORLD(E1)∧(∀X Y.E1(X,Y)⊃E(X,Y))⊃(∀X Y.E(X,Y)≡E1(X,Y)))| NIL)
This is the circumscription formula -- we are circumscribing the equivalence
relation E.
6. (ASSUME |WORLD1(E0)|)
deps: (6)
Since EKL cannot do the circumscription, we assume the result. Many subsequent
statements will list 6 as a dependency because of this.
7. (DEFINE E2 |∀X Y.E2(X,Y)≡X=A∧Y=B∨x=b∧y=a∨X=Y| NIL)
This is what we would like E0 to be in WORLD1. The only equivalent pairs
are [A,B] and pairs in which both elements are the same.
8. (DERIVE |EQUIV(E2)| () ((OPEN EQUIV) (OPEN E2)))
EQUIV(E2)
We infer that E2 is an equivalence relation.
9. (DERIVE |WORLD(E2)| (8) ((OPEN WORLD) (OPEN E2)))
WORLD(E2)
The relation E2 satisfies the definition of WORLD.
10. (RW 6 (OPEN WORLD1))
WORLD(E0)∧(∀E1.WORLD(E1)∧(∀X Y.E1(X,Y)⊃E0(X,Y))⊃(∀X Y.E0(X,Y)≡E1(X,Y)))
deps: (6)
Here we expand the circumscription formula.
11. (DERIVE |WORLD(E0)| (10) NIL)
WORLD(E0)
deps: (6)
This is the first conjunct of the previous statement.
12. (RW 11 (OPEN WORLD))
E0(A,B)∧EQUIV(E0)
deps: (6)
Expanding the definition of WORLD.
13. (RW 12 (OPEN EQUIV))
E0(A,B)∧(∀X.E0(X,X))∧(∀X Y.E0(X,Y)⊃E0(Y,X))∧(∀X Y Z.E0(X,Y)∧E0(Y,Z)⊃E0(X,Z))
deps: (6)
Expanding the definition of EQUIV
14. (DERIVE |∀X Y.E2(X,Y)⊃E0(X,Y)| (13) (OPEN E2))
∀X Y.E2(X,Y)⊃E0(X,Y)
deps: (6)
15. (DERIVE |∀E1.WORLD(E1)∧(∀X Y.E1(X,Y)⊃E(X,Y))⊃(∀X Y.E(X,Y)≡E1(X,Y))|
(10) NIL)
∀E1.WORLD(E1)∧(∀X Y.E1(X,Y)⊃E(X,Y))⊃(∀X Y.E(X,Y)≡E1(X,Y))
deps: (6)
16. (UE ((E1.|E2|)) 15 NIL)
WORLD(E2)∧(∀X Y.E2(X,Y)⊃E(X,Y))⊃(∀X Y.E(X,Y)≡E2(X,Y))
deps: (6)
The last three steps prove that E2 satisfies the circumscription formula.
17. (DERIVE |∀X Y.E(X,Y)≡E2(X,Y)| (9 14 16) NIL)
∀X Y.E(X,Y)≡E2(X,Y)
deps: (6)
Now we can find the set of equivalent pairs.
18. (RW 17 (OPEN E2))
∀X Y.E(X,Y)≡X=A∧Y=B∨X=Y
deps: (6)
19. (DERIVE |E(A,B)| (18) NIL)
E(A,B)
deps: (6)
As before, A and B are equivalent.
20. (DERIVE |¬A=C| (SIMPINFO) NIL)
¬A=C
From the axioms at the beginning, we derive the uniqueness of names.
21. (DERIVE |¬B=C| (SIMPINFO) NIL)
¬B=C
22. (DERIVE |¬E(A,C)| (18 20 21) NIL)
¬E(A,C)
deps: (6)
Here we demonstrate that things not explicitly stated to be equivalent
in our definition of the world are not equivalent.
.end
.skip 1
∂29-May-84 2003 JMC re: Purge of files on SCORE:<CS.*> (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - Files that refer to all users of the machine should be regarded
as facility files. I bet Stuart would find that some of the files
the department is paying for could be given that interpretation. Of
course, the machines still have to generate enough income to pay for
themselves, but flushing files doesn't actually reduce total costs.
∂29-May-84 2007 JMC re: about USNET (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - Frome here USNET is addressed as THEMNET.
∂29-May-84 2310 JMC
To: brachman@SRI-KL.ARPA
The schedule is as expected.
title: What is common sense?
∂30-May-84 1004 JMC
To: dlw@SCRC-STONY-BROOK.ARPA
Baumgart, Bruce 408 378-6038
∂30-May-84 1619 JMC
To: ME
Is the video sythesizer broken? xgpsyn hasn't been working for a while.
∂30-May-84 2213 JMC video synthesizer
To: bosack@SU-SCORE.ARPA
I'm unhappy about its loss and would like to see if we can't get back
what's required to run it in exchange for the remote connections we
discussed.
∂30-May-84 2217 JMC frame problem
To: pylyshyn@CMU-CS-C.ARPA
I evidently forgot the request. However, I have a new paper on
circumscription, one section which uses circumscription to solve the
frame problem. Please tell me again about the volume.
∂31-May-84 0126 JMC
To: HST
I will be at Stanford at the end of July, and you could not get into
a student dormitory which are reserved for specific conferences. How
about the Faculty Club, for which I can pay?
∂31-May-84 0132 JMC Ermakoff's test, etc.
To: rwg@MIT-MC.ARPA
You really should look at that section of Knopp including the remarks
at the end. I have it at home.
∂31-May-84 2107 JMC new computers
To: bosack@SU-SCORE.ARPA, gorin@LOTS-A
This is the first installment of my survey of demand for a new system
in the near future. This message is a supplement to a conversation
in which Brian told me that he had enough compute power for the near
future.
∂31-May-84 2054 reid@Glacier new computers
Received: from SU-GLACIER.ARPA by SU-AI.ARPA with TCP; 31 May 84 20:54:29 PDT
Date: Thu, 31 May 84 20:55:18 pdt
To: JMC@Sail
Subject: new computers
From: Brian Reid <reid@Glacier>
Almost all of the currently-available computers are based on processor
technology of the middle 1970's. The advent of custom VLSI chips and
huge semiconductor memories has radically changed the way that fast
cheap computers should be built. All DEC-20's and all VAXes except the
$5000 MicroVax-2 are built out of discrete components, and organized
around the assumption that there is going to be a lot of off-board
communication in every processor cycle because there isn't much memory
on a board.
Proper use of custom chips requires changes to the architecture to take
advantage of what on-chip processing can do well, so it's not practical
to build a fast implementation of an existing architecture using the
new technology.
For my own part I am quite reluctant to sink any more money right now
computers that will last 5 to 10 years and require maintenance
contracts for 5 to 10 years. DEC, IBM, and various startups are within
a year or two of announcing major new computers built this way.
Brian
∂01-Jun-84 1237 JMC tva
To: FY
I saw the article and also the Jane Jacobs article referred to. The
arguments seem somewhat plausible but hardly conclusive.
∂01-Jun-84 1242 JMC ai qual
To: DBL@SU-AI.ARPA
I would prefer to have my turn in the afternoon, since I want to go
to the organizing meeting of the SRI common sense summer. Incidentally,
the load can be reduced if some research associates can be recruited to
help. I consider Dick Gabriel to be qualified, and HPP has many qualified
people.
∂01-Jun-84 1350 JMC disk charges
To: bosack@SU-SCORE.ARPA
The new SU-ITS rate blurb lists $.01 per kilobyte month for disk storage.
How does this compare with rates on SAIL and SCORE? According to my
estimate, it's cheaper. Also it's stated in units that are worth
remembering, since it's unlikely that the notion of kilobyte will change,
whereas track and block are evanescent in meaning.
∂01-Jun-84 1358 JMC disk charges
To: bosack@SU-SCORE.ARPA
It appears that SAIL charges for disk are 3.5 times those of ITS. This
may be inaccurate, because of not knowing how they treat partially
full blocks. However, our charges are clearly out of line and accordingly
should be greatly reduced.
∂01-Jun-84 1423 JMC Goldblatt
To: barwise@SU-CSLI.ARPA
Feferman taught a course based on Goldblatt's book. Carolyn Talcott's opinion,
and she read the book, was that the book has little to do with computer
programming. Bringing Goldblatt here, in her opinion, with which I am
inclined to concur, would be a charitable enterprise. However, you might
ask Feferman's opinion.
∂01-Jun-84 1504 JMC
To: DC
The xgp is getting rather gray.
∂01-Jun-84 2301 JMC foolog
To: ken@MIT-MC.ARPA
Would it be convenient to MAIL me the text of Martin Nilsson's foolog
interpreter written in Maclisp? If inconvenient I can, of course,
type it in.
∂02-Jun-84 1600 JMC
To: JK
Looks good to me. I'll tryit.
∂02-Jun-84 2053 JMC
To: faculty@SU-SCORE.ARPA
I'm on Brian's side on the Pattis issue.
∂02-Jun-84 2130 JMC Should this fail? It takes only about 1.5 sec to do so.
To: JK
(der-fast)
(der |∀p1 p2 p3 q0 q2.
(∀x y.q2(x,y) ≡ p1(x,y) ∨ p2(x,y) ∨ p3(x,y))
∧ (∀x y.p1(x,y) ⊃ q0(x,y))
∧ (∀x y.p2(x,y) ⊃ q0(x,y))
∧ (∀x y.p3(x,y) ⊃ q0(x,y))
⊃ (∀x y.q2(x,y) ⊃ q0(x,y))|)
∂03-Jun-84 1227 JMC change day of visit
To: LLW@SU-AI.ARPA
Unfortunately, I have to be at Stanford on Wednesday. I'll come Thursday
if that's ok, and I'll check with your secretary tomorrow (or you if I
can catch you).
∂03-Jun-84 1400 JMC thanks for help
To: JK
My unique names proof is now in satisfactory form for the new
circumscription paper. I have two suggestions for printout.
of proofs. First, label statements should be printed in input format.
If they are printed as comments, the proof cannot be fed back in
without repair. Second, a successful DERIVE command should not print
the result any more than ASSUME or DEFINE does. Actually, it would
be best to print it in each of these cases when the user has referred
to parts in his input. It would then be best to reprint the command
itself with the reference to parts left as is. Of course, these last
refinements aren't worth a lot of trouble.
∂03-Jun-84 1532 JMC
To: JK
Thanks for the new proof; I'll study it.
∂03-Jun-84 1811 JMC Mystery File in My Directory
To: grosof@SUMEX-AIM.ARPA
Are you somehow responsible for this?
∂03-Jun-84 1357 SIEGMAN@SU-SIERRA.ARPA Mystery File in My Directory
Received: from SU-SIERRA.ARPA by SU-AI.ARPA with TCP; 3 Jun 84 13:57:34 PDT
Date: Sun 3 Jun 84 13:56:43-PDT
From: Tony Siegman <SIEGMAN@SU-SIERRA.ARPA>
Subject: Mystery File in My Directory
To: su-bboards@SU-SIERRA.ARPA
Mystery file with filename "UNIQUE.NAM" and contents of 1 blank line and
then following line
(!UNIQUE UYNCAA 4252)
has now twice appeared in my directory. Means nothing to me or anyone
I have computer contact with. What's going on?
-------
∂03-Jun-84 2252 JMC bug help needed
To: JK
I made only trivial modifications to your proof in order to fit the
format of my paper. However, step 15 doesn't give the same result.
The last page of equal.lsp[s84,jmc] contains my proof, which I compared
line-by-line with yours. Only the result of step 15 appears to be
different. I re-ran yours adding only label statements and der-slow
and got the same results you got. Most likely, mine contains some
non-trivial change that my proof-reading didn't uncover. Perhaps
the state of ekl would make the problem apparent.
∂04-Jun-84 0108 JMC Proposed Overview of Commonsense Summer
To: pack@SU-SCORE.ARPA
I mentioned to Nils that you were omitted, and he said you should come too.
∂02-Jun-84 1140 HOBBS@SRI-AI.ARPA Proposed Overview of Commonsense Summer
Received: from SRI-AI.ARPA by SU-AI.ARPA with TCP; 2 Jun 84 11:39:15 PDT
Date: Sat 2 Jun 84 11:39:03-PDT
From: HOBBS@SRI-AI.ARPA
Subject: Proposed Overview of Commonsense Summer
To: nilsson@SRI-AI.ARPA, stan@SRI-AI.ARPA, konolige@SRI-AI.ARPA,
georgeff@SRI-AI.ARPA, stickel@SRI-AI.ARPA, appelt@SRI-AI.ARPA,
bmoore@SRI-AI.ARPA, hobbs@SRI-AI.ARPA, rperrault@SRI-AI.ARPA,
barnard@SRI-AI.ARPA, pentland@SRI-AI.ARPA, pereira@SRI-AI.ARPA,
pcohen@SRI-AI.ARPA, blenko@SRI-AI.ARPA, dymetman@SRI-AI.ARPA,
kells@SRI-AI.ARPA, croft@SRI-AI.ARPA, kube@UCB-VAX.ARPA, shoham@YALE.ARPA,
kautz@ROCHESTER.ARPA, jmc@SU-AI.ARPA, dekleer@XEROX.ARPA, brown@XEROX.ARPA,
pat@ROCHESTER.ARPA, james@ROCHESTER.ARPA, mcdermott@YALE.ARPA,
herskovits@UCB-VAX.ARPA
COMMONSENSE SUMMER
Aim:
It has long been agreed that if robots are going to be capable of
intelligent behavior, they will need to have a great deal of knowledge
about the commonsense world. But no one has yet embarked on a
large-scale effort to encode this knowledge. The aim of Commonsense
Summer is to do the first three months of such an effort. We will
attempt to axiomatize in formal logic significant amounts of
commonsense knowledge about the physical, psychological and social
worlds. We will concentrate on eight domains: shape and texture,
spatial relationships, the lexical semantics of cause and possession,
properties of materials, certain mental phenomena, communication,
determiners viewed as relations between textual entities and entities
in the world, and responsibility. We will attempt to make these
axiomatizations mutually consistent and mutually supportive.
We will (at least at first) have one large meeting a week to keep
everyone posted on everyone else's progress and problems. There will
also be smaller meetings as appropriate to discuss problems in
particular domains. People interested in the same problems should
make an effort to get together.
Coordinator: Jerry Hobbs
People Participating:
SRI/CSLI Summer Students:
Tom Blenko (Rochester), Communication, June 4 - August 17
Bill Croft (Stanford), Determiners
Marc Dymetman (Grenoble, France), Lexical Semantics
Greg Hager (Pennsylvania), Materials, June 4 - August 31
Henry Kautz (Rochester), Space, June 4 - August 17
Kathleen Kells (Stanford), Responsibility
Paul Kube (Berkeley), Mental Phenomena
Yoav Shoham (Yale), Shape and Texture, June 7 - August 31
SRI/CSLI: Doug Appelt, Steve Barnard, Phil Cohen, Jerry Hobbs,
Kurt Konolige, Bob Moore, Nils Nilsson, Sandy Pentland,
Fernando Pereira, Ray Perrault, Stan Rosenschein, Mark Stickel
Stanford/CSLI: John McCarthy, Johann van Benthem
Xerox PARC: John Seely Brown, Johann de Kleer
Fairchild: Hector Levesque, Andy Witkin
Berkeley: Annette Herskovits, Len Talmy
Rochester: James Allen (June 18-29), Pat Hayes (June 4-22),
Yale: Drew McDermott (June 18-July 13)
Domains:
We will concentrate on the eight domains listed below; a student is
assigned to each. I have given below a characterization of what I see
the domain to be. I have made these up largely without consulting the
students who will be working on them, so there will undoubtedly be
some more or less significant changes, especially as work begins and
problems are encountered. In all of these domains, I have tried to
define target problems, to give us something to shoot for (or at).
I have also tried to set up a healthy tension between robotics
applications and natural language applications. The aim is to achieve
generality by constructing axiomatizations that would be useful for
both.
If there are other domains that other people would like to axiomatize
in a way consistent with these eight axiomatizations, I would certainly
encourage it, although it's not appropriate for me to twist anyone's
arm.
With each domain I've listed papers I recommended to the students to
provide the necessary background and help create a common culture.
The following papers were recommended to all the students.
For an overview of the entire enterprise:
Patrick Hayes, "The Second Naive Physics Manifesto"
Jerry Hobbs, "Introduction" to "Formal Theories of the
Commonsense World"
On the adequacy of logic:
Robert Moore, "The Role of Logic in Knowledge Representation
and Commonsense Reasoning"
Robert Moore, "Problems in Logical Form"
Jerry Hobbs, "The Logical Notation: Ontological Promiscuity"
Shape and Texture: Yoav Shoham
There has been much recent progress in vision research on the
recognition of shapes and textures, and this area is thus ripe for
investigating the relation between vision and language. Possible
target problems:
(1) Have people describe in English half a dozen pictures that vision
researchers have been using as key examples. Encode the knowledge
that will allow one to go from a characterization of the picture that
a present-day visual component could produce to a suitably idealized
version of the English description. It may be that, as Witkin &
Tennenbaum suggest, regularities in images are understood via causal
accounts of what would produce the regularities. Encode the knowledge
that would be required to infer abstract causal explanations of the
image.
(2) Characterize a number of English shape and texture words, such as
"rough", "bumpy", "undulating", "amorphous", etc., in terms of
visual primitives.
Background reading:
Drew McDermott & Ernest Davis, "Planning Routes through
Uncertain Territory"
Alex Pentland, "Fractal-based Description of Natural Scenes"
Alex Pentland, "Local Shading Analysis"
Andrew Witkin, "Scale Space Filtering: A New Approach to
Multi-scale Description"
Andrew Witkin & Jay M. Tenenbaum, "What is Perceptual
Organization For?"
Stephen Barnard, "Interpreting Perspective Images"
Stephen Barnard, "Choosing a Basis for Perceptual Space"
Space: Henry Kautz
The axiomatization of spatial relationships should span the (probably)
Cartesian view of space that a robot would need to move around the
hallways and the more topological view of language that seems to
underlie natural language. The axiomatization should support spatial
metaphor. A possible concrete goal is to encode the knowledge that
would be required by the proposed KLAUS system to understand
descriptions of desired computer graphics displays. Axiomatize time
as a footnote.
Background Reading:
George Miller & Phillip Johnson-Laird, Sections 6.1 ("Spatial
Relations") and 6.2 ("Temporal Relations") in "Language
and Perception" - state of the art as of mid 1970's.
Annette Herskovits, Chapter 3 ("The Three Basic Topological
Prepositions") and 4 ("The Projective Prepositions") in
"Space and the Prepositions in English: Regularities and
Irregularities in a Complex Domain"
Annette Herskovits, "Comprehension and Production of Locatives"
Leonard Talmy, "How Language Structures Space"
(Herskovits and Talmy are the principal recent advances
over Miller & Johnson-Laird.)
Drew McDermott & Ernest Davis, "Planning Routes through
Uncertain Territory" (what a robot would have to know)
Cause, Motion and Perception: Marc Dymetman
George Miller and Phillip Johnson-Laird, in their book "Language and
Perception", have written the (so far) definitive book on lexical
semantics. Their rules for decompositions of verbs can be viewed
as axioms. The problem is to recast their rules into a form that is
consistent with the other axiomatizations being developed for
commonsense summer.
Background Reading:
George Miller & Phillip Johnson-Laird, Sections 6.3 ("Causal
Relations"), 7.1 ("Verbs of Motion"), and 7.2 ("Verbs of
Possession") in "Language and Perception"
Materials: Greg Hager
Consider a variety of materials: metal, wood, concrete, rubber, cloth,
sand, powder, chocolate, water, smoke, air. What states can they be
in and what changes do they undergo when subjected to certain physical
actions, such as motion, deformation, cutting, loss of support. Under
what conditions do they expand or contract, erode, decay, etc.? How
do these properties relate to some of their principal functions. An
extension of Pat Hayes's liquids paper.
Background Reading:
Patrick Hayes, "Naive Physics I: Ontology for Liquids"
Mental Phenomena: Paul Kube
There has been much progress lately in models of belief, desire and
intention. There has also been psychological work on people's metaphors
for the structure of mind. In terms of these two, characterize the
folk view of a number of mental phenomena, such as "remember", "forget",
"suspect", "think of", etc.
Background Reading:
Robert Moore, "A Formal Theory of Knowledge and Action"
Kurt Konolige, "Belief and Incompleteness"
C. Raymond Perrault & James Allen, "A Plan-based Analysis of
Indirect Speech Acts"
George Miller & Phillip Johnson-Laird, Sections 7.3 ("Verbs of
Vision") in "Language and Perception"
Communication: Tom Blenko
Continuation of work by Perrault, Cohen, Allen, Levesque and others
on planning speech acts. Carry this work to one more level of detail.
Axiomatize what the proposed KLAUS system must know about the
communication it will have with the user. Another source of problems
arises in a situation I have devised involving complex interaction in
a simple blocks world, allowing crisp formulations of such acts as
indirect offers, suggestions, tricks, negotiation, etc.
Background Reading:
C. Raymond Perrault & James Allen, "A Plan-based Analysis of
Indirect Speech Acts"
Philip Cohen & Hector Levesque, "Speech Acts and the
Recognition of Shared Plans"
Douglas Appelt, "Planning English Referring Expressions"
Martha Pollack, "Generating Expert Answers through Goal
Inference"
Robert Moore, "A Formal Theory of Knowledge and Action"
George Miller & Phillip Johnson-Laird, Sections 7.4 ("Language
of Communication") in "Language and Perception"
Text and Determiners: Bill Croft
The meanings of many determiners can be characterized as a relation
between linguistic objects, viz. noun phrases, and things in the world.
Axiomatize knowledge about texts and a simple theory of meaning, in
which these relations can be captured. This effort follows up on
previous linguistic work by Bill Croft and is an attempt to cast it
into a logical, AI framework.
Communicative Continuity and Morality: Kathleen Kells
Speech acts iterated over time. How does the fact that communication
will continue affect one's moment-by-moment behavior? This leads
to notions of responsibility, especially the responsibility for the
honesty of one's assertions and the sincerity of one's requests.
This responsibility is the glue that holds social life together.
Leads to problems of characterizing certain aspects of morality.
Background Reading:
Jaime Carbonell, "Computer Models of Human Personality Traits"
C. Raymond Perrault & James Allen, "A Plan-based Analysis of
Indirect Speech Acts"
Philip Cohen & Hector Levesque, "Speech Acts and the
Recognition of Shared Plans"
Interface Problems:
I will be responsible for trying to insure that the various
axiomatizations are mutually consistent and mutually supportive.
My guess is that the interface problems will be localized within
the four physical domains and the four psychosocial domains. For
example, reasoning about where to grasp an irregularly shaped,
rough-textured bowl containing a heap of round objects and moving
it to another location requires knowledge of shapes and textures,
spatial relations, and the behavior of materials. Reasoning about
deception requires knowledge of the other's beliefs, communicative
behavior and sense of responsibility.
In part, coordinating these efforts will involve negotiating interface
predicates between the domains. The theory of materials will need
a containment relation with certain properties, so these properties
must be axiomatized in the theory of spatial relations.
Methodology (Sort Of):
There is no real, known methodology for axiomatizing domains. I have
found three methods useful:
(1) Proceed bottom-up: What is the ontology of the domain? What kinds
of entities are there? What properties do they have? What classes do
they fall into? What relations are there among them? Is there some
set of concepts which seem more "primitive" in terms of which others can
be characterized?
(2) Proceed top-down: Write down in English the facts that seem to be
needed for the class of behaviors or the class of texts of interest.
Gradually refine these statements into rationally arranged predicate
calculus axioms.
(3) Work toward the solution of a target problem, axiomatizing
everything that seems to be required.
The discovery of certain methodological principles for axiomatizing
domains could be one of the most important contributions of Commonsense
Summer.
Monitoring Progress:
I think we need some way of making sure that adequate progress is
being made. One possible mechanism is to have the students write
up a page or so each week about what they've done and where they are.
The first few weeks this might be a clearer formulation of what their
domain is. A bit later it might be statements in English of what the
basic concepts are and the principal facts that must be captured.
Toward the end it should be lists of axioms together with a discussion
of problems the axiomatization doesn't cover. The progress reports
might be viewed as successively expanded and refined drafts of a paper.
The term "progress" should be loosely construed. The discovery of
an unexpected difficulty is progress. The reduction of a page of
axioms to a half page of axioms is progress.
At the end of the summer, we should put together a report with papers
by me, the students, and whoever else wishes to contribute something.
Some Black Holes:
This is a partial list of problems that could easily dominate
discussions for the entire summer. It is hoped these issues will be
avoided or bypassed in some simple manner.
1. Is first-order predicate calculus adequate? Answer: Maybe not,
but let's give it a shot.
2. Notational disputes. Answer: I propose we take a flat,
ontologically promiscuous first-order notation as the lingua franca.
People should then be free to work in any notation they feel
comfortable with, as long as they define translation rules from their
notation to the lingua franca.
3. Before axiomatizing a certain phenomena, a new representation
language has to be devised and implemented. Answer: Not true. This
is the classic waste of time in AI.
4. Semantic disputes. Are there such things as events, actions,
possible worlds, etc.? Is our view of time based on points or
intervals? Answer: Be ontologically promiscuous. Take the union
of the two competing ontologies, and see what the axiomatizations
really need.
5. Definition. Searching for necessary and sufficient conditions
for concepts. Answer: Aim for "characterization", not definition.
We can't expect to discover necessary and sufficient conditions, just
lots of necessary conditions and lots of sufficient conditions.
6. One counterexample refutes a theory. Answer: Theories are not
refuted, only replaced by better theories. One should try to maximize
coverage, settling, where necessary, for axioms that are usually true,
confident that higher processes will plug the holes in the theories.
7. Context-independence. Attempting to characterize concepts in
ways that are independent of the context of use. Answer: Knowledge
will be used in a context-dependent fashion. This assumption allows
us to be loose where being strict would be a black hole. Assume
circumscriptive reasoning is available.
8. Efficiency. Decidability. Answer: For over two decades, AI
researchers have worried about these issues. It won't hurt to ignore
them for three months while we worry about getting the content right.
9. Fuzziness. Uncertainty. Most commonsense concepts are inherently
fuzzy, so a formalism for dealing with fuzzy concepts must first be
adopted or worked out. Answer: Either adopt the idealization that the
world has sharp boundaries or assume a theory of granularity that
maps the world into such an idealized world.
10. Reduction of folk concepts to mathematical structures. For example,
how close is near? Answer: The relation between between folk concepts
and mathematical concepts, e.g., between cognitive topology and
Cartesian space, is in general complex and context-dependent. Relations
between them can be characterized, but the mathematical should not be
taken as primitive.
11. Is there any limit to what we know? Answer: Who cares? Every
little bit of knowledge should still help.
A Gray Hole:
A theory of granularity is needed by every domain. For example, what is
texture at one grain-size is shape at another. At one grain-size, one
can be "at" the post office, while at another one is "in" the post
office "at" the stamp window. The difference between a leak and an
outlet is one of grain-size. This is probably not a problem that can
be sidestepped. One possible solution: Index predicates by grain-size.
-------
∂04-Jun-84 0118 JMC omissions
To: hobbs@SRI-AI.ARPA
Jerry:
In the main I agree with your view of common sense, but there
are some exceptions. First, two topics I consider important for
axiomatization are the consequences of actions and other events
and the relation between objects and phenomena on the one hand
and observations of them on the other. Incidentally, you might
include my old MI4 paper with Pat Hayes, Some Philosophical Problems
from the Standpoint of Artificial Intelligence, among the references.
Second, common sense reasoning is as important as common sense
knowledge, and it isn't just logical deduction. Some first order formalizations
presuppose that non-monotonic reasoning will also be used.
Also you left out Leslie Pack, LEP@SU-AI, off your list of
addressees. SRI has just hired her to work on common sense.
John
∂05-Jun-84 1339 JMC
To: DFH
Please Scribe or tech gaffne.1[let,jmc].
∂05-Jun-84 1411 JMC
To: konolige@SRI-AI.ARPA
I will be in, but I will first have to check that I am satisfied with
the final version.
∂05-Jun-84 1419 JMC
To: wouk@BRL-VGR.ARPA
You have stated the problem correctly up to a point, but the additional
event occurs that S states that now he knows the numbers. The
problem for you is to determine the numbers. Here is the solution
in a form you can decode if you want. I have written the numbers
each taking two digits preceded by 1 getting six digits and then
converted the result to octal, which is 313261. If the numbers had
been 8 and 19, I would have written 323127. I don't know who originated
the problem.
∂05-Jun-84 1427 JMC
To: schreiber@SU-SCORE.ARPA
McGraw-Hill I think.
∂06-Jun-84 0057 JMC
To: HST
I don't have Edwards, but you can try National Security Agency,
Fort George Meade, Maryland. For Silver I have
Silver, Roland 505 586-0197, Box 111, San Cristobal, New Mexico,
but this is several years old.
∂06-Jun-84 0201 JMC
To: DFH
Please arrange to make R. William Gosper a visiting scholar. He will
phone you.
∂07-Jun-84 1723 Mailer failed mail returned
To: JMC
In processing the following command:
MAIL JMB-S1-A
The following message was aborted because of a command error,
namely, nonexistent recipient(s):
JMB-S1-A
------- Begin undelivered message: -------
∂07-Jun-84 1723 JMC IBM 801 talk
∂22-May-84 1728 baskett@decwrl.ARPA IBM 801 talk
Received: from DECWRL.ARPA by SU-AI.ARPA with TCP; 22 May 84 17:26:18 PDT
Received: from acetes.ARPA by decwrl.ARPA (4.22.01/4.7.29)
id AA01520; Mon, 21 May 84 09:52:33 pdt
Received: by acetes.ARPA (4.22.01/4.7.27)
id AA00796; Mon, 21 May 84 09:53:41 pdt
From: baskett@decwrl.ARPA (Forest Baskett)
Message-Id: <8405211653.AA00796@acetes.ARPA>
Date: 21 May 1984 0953-PDT (Monday)
To: jmc@su-ai
Subject: IBM 801 talk
I took out the remarks about intra-company rivalries but I'd still
appreciate your not spreading this around since I promised George Fan
(IBM Yorktown Heights) that I wouldn't.
Forest
------
This is my report on a talk given at Stanford on April 3rd. First the
published abstract:
Albert Chang, IBM Yorktown, VLSI Seminar, Tuesday, April 3 1:15 pm
ERL 401.
Two IBM 801 Chips
Preliminary performance data and the external characteristics of two 801 type
chips will be described. The first chip is a pipelined reduced instruction
set cpu (801). Its measured performance on small kernels will be given and
an indication of the relative power of one of its instructions compared to
a 370 instruction will be discussed. The second chip is a memory management
chip. It supports very large virtual memory at low cost in terms of the real
fixed storage required for page tables. It also provides hardware support for
the locking and journalling that is commonly used in data base programming.
The way in which it is used in an experimental control program will be
described.
Now the report:
Background:
Al Chang got a Ph.D. in EE from Berkeley years ago and has been at
Yorktown Heights for a long time, I think. He started on the 801
project very near it's beginning in 1975. The best background
reading on the 801 project is "The 801 Minicomputer" by George Radin in
The IBM Journal of Research and Development, Vol. 27, No. 3, May 1983.
This article is a slightly expanded version of the paper Radin
presented at the ACM Symposium on Architectural Support for Programming
Languages and Operating Systems in March of 1982. More details are
included in the IBM Journal version.
Motivation:
It came out at the end of the talk that these chips (at least the CPU
chip and I believe the MMU chip, as well) were designed by the IBM
Office Products people in Austin, Texas, the people who brought you the
Displaywriter. (censored part) At any rate, it is widely
believed that this is the chip set that will be in the machine that
is about to be delivered to CMU as part of the big IBM-CMU program. In
addition, even the trade press is starting to figure out what is going
on and is talking about IBM coming out with a proprietary CPU in their
"next generation PC". This project has long had the code name ROMP.
The talk:
The block diagram was
----- -----
| CPU |---------| MMU |-----Memory
----- | -----
|
I/O
This presumably implies I/O instructions rather than memory mapped I/O.
Nothing further about I/O was mentioned. The CPU was described as
RISC, 32 bits, pipelined, with cycle time T. The design target for T
was 200 nanoseconds but the chips seem to run at 160 ns with selected
parts running at 150ns. The MMU supports a large virtual memory (2**40
bytes), has data base support, is pipelined, and has the same pipeline
cycle time. The memory is plain Nmos with an access time of approximately
T and a cycle time of 2T. It can be two way interleaved to reduce
interference. If it is not so interleaved, there could be substantial
speed degradation due to memory interference on successive instructions.
The CPU has 16 registers (the Yorktown Heights people think this is too
few) which are 32 bits. Instructions are either 16 bits or 32 bits.
The 16 bit instructions are of the form:
OP R1, R2
where OP is 8 bits and R1 and R2 are 4 bit register designators. The
32 bit instructions are of the form:
OP R1, R2, D
where D is a 16 bit displacement, sometimes signed, sometimes not.
This form provides 16 bit literal operands and 16 bit displacements for
the load and store instructions. It is a load/store machine in that
only the load and store instructions have storage operands. This has
the usual simplifying effect on the interrupt and virtual memory
implementation. Most instructions are one cycle (if they are not loads
or stores or branches). There are a few microcoded, multiple cycle
instructions. One pair is a load or store multiple registers and another
is load program status (load PSW in IBMese). These instructions are supposed
to help relieve the load on the single port to memory in the same way
that the 16 bit instructions do (less instruction traffic to interfere
with the data traffic). In addition, there are a "small number" of
three register operand instructions of the form:
OP R1, R2, R3
The pipeline organization was diagrammed in what looked to me like the
standard way with a register file with two output ports and two input
ports, an ALU with two input ports and one output port and at least one
bypass from ALU out to ALU in. His diagram only showed one bypass but
it left out a lot so there could be more. It's the standard scheme for
pipelining one cycle ALU operations in a register to register environment.
The extra input port on the register file is for loads.
The Load and Stores were more interesting because they were so slow: 5
cycles with address translation on and 4 cycles with it off, diagrammed
as follows:
| Load | Addr | Xlate | Memory | Reg File |
+------+-------+--------+--------+----------+
When address translation is off, the Xlate cycle is deleted. When it
is on, the CPU waits for the data in order to achieve precise
interrupts if there is a fault. When it is off, the CPU continues
until the data is required. Instruction fetch continues in either
case. The compiler is able to schedule 90% of the loads so that there
is a useful instruction between the load and the use (for the Xlate off
case) and 80% of the loads so that there are two useful instructions
between the load and the use. The memory system was not described so
this explanation could be a little simple compared to reality.
Branches take one cycle if the branch is not taken, 5 cycles if the
branch is taken. There is a "branch and execute" form in which the
instruction following the branch is executed no matter which way the
branch goes. Thus the taken path is effectively 4 cycles for this form.
The load multiple instruction takes 3+ceiling(3R/2) cycles were R is
the number of registers being loaded. The store multiple instruction
takes 3+2R cycles.
The compiler is the PL8 compiler. It mainly compiles PL8, a PL/I like
language for systems programming. It is about 100K lines of code that
compiles into 1.5 megabytes of flattened executable image. It compiles
to about five different machines, including various versions of 801's,
the 370, and the 68000. It also compiles Pascal source and they expect
to be able to compile C source soon. A table of instruction
frequencies was given for three "programs". One was a data base system
they have done that implements the Sequel interface. One was the
"system" which was their monitor and a lot of utilities. The third was
the PL8 compiler itself. These are dynamic instruction frequencies.
Instruction % Sequel System PL8
Register ops 52.9 50.4 52.2
Loads 23.7 19.6 21.6
Stores 9.3 8.9 9.6
Branches taken 9.0 15.7 11.5
Branches not taken 3.8 3.8 4.0
Load/Store multiple 1.3 1.6 1.1 (8 registers, average)
Average cycles per 2.93 3.00 2.86
instruction
Number of instructions 30M 30M 600M
Other effects such as memory interleave interference and translation
buffer misses accounted for 0.3 cycles per instruction in addition to
the above. These numbers together with the 160 ns cycle show you where
the 2.2 MIPS number that John Cocke has been giving people comes from,
I believe.
Al then presented, with some reluctance, the following simple minded
benchmark numbers, probably out of deference to the Berkeley crowd
since he was scheduled to give the talk at Berkeley the next day.
He complained about the relevance of small kernals.
Time in milliseconds
Machine Language Acker Sieve Puzzle
11/780 C (VMS) 4222 116 4625
68000 Pascal 3080 246 9200
(16 MHz)
80286 Pascal 2218 168 9138
(8 MHz)
IBM RISC -- 2400 80 3800
(6.25 MHz) (PL8) (PL8) (Pascal)
Code size in bytes
11/780 152 156 2220
68000 126 228 2940
80286 --- --- ----
IBM RISC 159 202 2454
The numbers not done at IBM (780 and 286) were taken from the relevant
latest versions in Computer Architecture News.
Then, with more enthusiasm, he gave these comparisons for seven
unidentified applications programs.
10**6 instructions executed
PL8 RISC 370
1 103.0 63.7
2 171.9 107.0
3 48.8 40.0
4 88.6 71.0
5 115.8 94.4
6 49.9 38.1
7 85.3 60.5
--- ----- -----
total 663.3 474.7
overall 370/RISC = 0.715 (dynamic)
In the total number of static instructions in these 7 programs, the
IBM RISC and the 370 were about the same. Two thirds of the dynamic
excess of IBM RISC over the 370 were said to be due to the move and
multiply subroutines. The IBM RISC numbers were gathered with a
software trace. The 370 numbers were gathered with a hardware monitor
on a 370/168 which excluded all the CP/CMS system code. These programs
were said not to be commercial - no decimal - and said not to be
scientific - no floating point - but to be representative of systems
code. The average instruction size was also given: 2.85 bytes for the
IBM RISC and 3.5 for the 370.
The MMU chip supports a page size of 2048 bytes. The addresses are
byte addresses. The CPU presents 32 bits of address to the MMU, the
top four of which are a segment ID, the other 28 being an offset within
that segment. The segment registers in the MMU are 12 bits of address
that get put on the top of the 28 bit offset for a 40 bit virtual
address. The segment registers also contain a "special" bit that
controls whether the database stuff is active or not. It also contains
a "key" bit that has something to do with storage keys. I got the
impression that Al didn't want to explain this one because it was
embarrassing. The MMU keeps a 16 x 2 two way set associative translation
look aside buffer. In the monitor they are running, they map files into
segments, one file per segment. They expect most of the files in a
data base system to be relatively small so the 2**28 bytes maximum size
shouldn't be a problem. The special bit is meant to generate
interrupts to the monitor (control program in IBMese) so that locking
and journalling can be supplied by the monitor. The support is such
that one can get an interrupt on the first reference by a transaction
to a 128 byte line of data. The PL8 language has an added "Persistent"
storage class so that PL8 variables may be associated with permanent
segments (or 128 byte lines, I think):
DCL X(100) CHAR(80) PERSISTENT
OPEN (X, 'NAME', options)
X(1) = 'anything'
COMMIT (X)
Compiled code is independent of options such as "shared", "journalled",
"locks", etc. The page table in the MMU is the same organization
as used in the System 38 and gives a reversible map between
virtual and real addresses organized as a hash table as follows. If
one has N real pages (where N = 2**k for some k) then there is a 2N
entry hash table that points to an entry in the N entry page frame
table. The mth entry in the page frame table describes real page m.
That entry contains the virtual page number, a "next" pointer (for hash
collisions), a R/W bit, a TID (transaction id), and 16 lockbits, one
for each 128 byte line of a page. This whole structure was said to take
20N space, where units were not defined but I believe were bytes. It was
never suggested how big N could be. I believe N to be at least 1024.
That would mesh with the CMU talk of a workstation with "a couple of
megabytes".
Al said that, for a TLB miss, they typically got 2.2 references in the
hash table/page frame table, i.e., the probability of no chaining was
close to 90 percent. The R/W, TIB, and Lockbits apply only if the special
bit is on in the segment register. The MMU has a TIDREG (transaction id
register). An interrupt is generated by the MMU for a load/store to line N:
if TID ~= TIDREG (not equal)
if store, R/W = R or Lockbit(N) = 0
if Load, R/W = R and Lockbit(N) = 0
At this point it was decided that the talk had gotten too low level and
that the time was up anyway so we switched to questions. The
technology questions revealed that the chips are Nmos, 2 micron feature
size, with about 7.5 mm per side. Al said it does a bit better than a
4341 in raw CPU power, giving about 1.4 370 MIPS or 2 RISC MIPS. He
also said that the version they are running in Yorktown Heights has a
bug where instructions cross a page boundary and there is a fault on
both instruction and data references so they compile around it.
Additional speculation: APA displays (IBMese for "all points
addressable) have been big inside IBM for some time now and Jim Gosling
(author of the VAX Emacs screen editor) demonstrated a really nice
WYSIWYG (what you see is what you get) editor on a Sun workstation at
the recent Sun users group meeting. It wasn't available since Jim
works for IBM on the CMU project and the Sun was just their interim
development machine. Thus one would reasonably expect the ROMP's that
CMU gets to be like Sun workstations with these chips in them.
After the talk, (censored) people muttered
about how slow these chips were and how much better one could do if one
did things "right". Comments from Al pointed out that this system was
designed to be cheap and that accounted for some of the slowness of the
memory system.
Forest
------- End undelivered message -------
∂07-Jun-84 1724 JMC IBM 801 talk
To: JMB@S1-A.ARPA
∂22-May-84 1728 baskett@decwrl.ARPA IBM 801 talk
Received: from DECWRL.ARPA by SU-AI.ARPA with TCP; 22 May 84 17:26:18 PDT
Received: from acetes.ARPA by decwrl.ARPA (4.22.01/4.7.29)
id AA01520; Mon, 21 May 84 09:52:33 pdt
Received: by acetes.ARPA (4.22.01/4.7.27)
id AA00796; Mon, 21 May 84 09:53:41 pdt
From: baskett@decwrl.ARPA (Forest Baskett)
Message-Id: <8405211653.AA00796@acetes.ARPA>
Date: 21 May 1984 0953-PDT (Monday)
To: jmc@su-ai
Subject: IBM 801 talk
I took out the remarks about intra-company rivalries but I'd still
appreciate your not spreading this around since I promised George Fan
(IBM Yorktown Heights) that I wouldn't.
Forest
------
This is my report on a talk given at Stanford on April 3rd. First the
published abstract:
Albert Chang, IBM Yorktown, VLSI Seminar, Tuesday, April 3 1:15 pm
ERL 401.
Two IBM 801 Chips
Preliminary performance data and the external characteristics of two 801 type
chips will be described. The first chip is a pipelined reduced instruction
set cpu (801). Its measured performance on small kernels will be given and
an indication of the relative power of one of its instructions compared to
a 370 instruction will be discussed. The second chip is a memory management
chip. It supports very large virtual memory at low cost in terms of the real
fixed storage required for page tables. It also provides hardware support for
the locking and journalling that is commonly used in data base programming.
The way in which it is used in an experimental control program will be
described.
Now the report:
Background:
Al Chang got a Ph.D. in EE from Berkeley years ago and has been at
Yorktown Heights for a long time, I think. He started on the 801
project very near it's beginning in 1975. The best background
reading on the 801 project is "The 801 Minicomputer" by George Radin in
The IBM Journal of Research and Development, Vol. 27, No. 3, May 1983.
This article is a slightly expanded version of the paper Radin
presented at the ACM Symposium on Architectural Support for Programming
Languages and Operating Systems in March of 1982. More details are
included in the IBM Journal version.
Motivation:
It came out at the end of the talk that these chips (at least the CPU
chip and I believe the MMU chip, as well) were designed by the IBM
Office Products people in Austin, Texas, the people who brought you the
Displaywriter. (censored part) At any rate, it is widely
believed that this is the chip set that will be in the machine that
is about to be delivered to CMU as part of the big IBM-CMU program. In
addition, even the trade press is starting to figure out what is going
on and is talking about IBM coming out with a proprietary CPU in their
"next generation PC". This project has long had the code name ROMP.
The talk:
The block diagram was
----- -----
| CPU |---------| MMU |-----Memory
----- | -----
|
I/O
This presumably implies I/O instructions rather than memory mapped I/O.
Nothing further about I/O was mentioned. The CPU was described as
RISC, 32 bits, pipelined, with cycle time T. The design target for T
was 200 nanoseconds but the chips seem to run at 160 ns with selected
parts running at 150ns. The MMU supports a large virtual memory (2**40
bytes), has data base support, is pipelined, and has the same pipeline
cycle time. The memory is plain Nmos with an access time of approximately
T and a cycle time of 2T. It can be two way interleaved to reduce
interference. If it is not so interleaved, there could be substantial
speed degradation due to memory interference on successive instructions.
The CPU has 16 registers (the Yorktown Heights people think this is too
few) which are 32 bits. Instructions are either 16 bits or 32 bits.
The 16 bit instructions are of the form:
OP R1, R2
where OP is 8 bits and R1 and R2 are 4 bit register designators. The
32 bit instructions are of the form:
OP R1, R2, D
where D is a 16 bit displacement, sometimes signed, sometimes not.
This form provides 16 bit literal operands and 16 bit displacements for
the load and store instructions. It is a load/store machine in that
only the load and store instructions have storage operands. This has
the usual simplifying effect on the interrupt and virtual memory
implementation. Most instructions are one cycle (if they are not loads
or stores or branches). There are a few microcoded, multiple cycle
instructions. One pair is a load or store multiple registers and another
is load program status (load PSW in IBMese). These instructions are supposed
to help relieve the load on the single port to memory in the same way
that the 16 bit instructions do (less instruction traffic to interfere
with the data traffic). In addition, there are a "small number" of
three register operand instructions of the form:
OP R1, R2, R3
The pipeline organization was diagrammed in what looked to me like the
standard way with a register file with two output ports and two input
ports, an ALU with two input ports and one output port and at least one
bypass from ALU out to ALU in. His diagram only showed one bypass but
it left out a lot so there could be more. It's the standard scheme for
pipelining one cycle ALU operations in a register to register environment.
The extra input port on the register file is for loads.
The Load and Stores were more interesting because they were so slow: 5
cycles with address translation on and 4 cycles with it off, diagrammed
as follows:
| Load | Addr | Xlate | Memory | Reg File |
+------+-------+--------+--------+----------+
When address translation is off, the Xlate cycle is deleted. When it
is on, the CPU waits for the data in order to achieve precise
interrupts if there is a fault. When it is off, the CPU continues
until the data is required. Instruction fetch continues in either
case. The compiler is able to schedule 90% of the loads so that there
is a useful instruction between the load and the use (for the Xlate off
case) and 80% of the loads so that there are two useful instructions
between the load and the use. The memory system was not described so
this explanation could be a little simple compared to reality.
Branches take one cycle if the branch is not taken, 5 cycles if the
branch is taken. There is a "branch and execute" form in which the
instruction following the branch is executed no matter which way the
branch goes. Thus the taken path is effectively 4 cycles for this form.
The load multiple instruction takes 3+ceiling(3R/2) cycles were R is
the number of registers being loaded. The store multiple instruction
takes 3+2R cycles.
The compiler is the PL8 compiler. It mainly compiles PL8, a PL/I like
language for systems programming. It is about 100K lines of code that
compiles into 1.5 megabytes of flattened executable image. It compiles
to about five different machines, including various versions of 801's,
the 370, and the 68000. It also compiles Pascal source and they expect
to be able to compile C source soon. A table of instruction
frequencies was given for three "programs". One was a data base system
they have done that implements the Sequel interface. One was the
"system" which was their monitor and a lot of utilities. The third was
the PL8 compiler itself. These are dynamic instruction frequencies.
Instruction % Sequel System PL8
Register ops 52.9 50.4 52.2
Loads 23.7 19.6 21.6
Stores 9.3 8.9 9.6
Branches taken 9.0 15.7 11.5
Branches not taken 3.8 3.8 4.0
Load/Store multiple 1.3 1.6 1.1 (8 registers, average)
Average cycles per 2.93 3.00 2.86
instruction
Number of instructions 30M 30M 600M
Other effects such as memory interleave interference and translation
buffer misses accounted for 0.3 cycles per instruction in addition to
the above. These numbers together with the 160 ns cycle show you where
the 2.2 MIPS number that John Cocke has been giving people comes from,
I believe.
Al then presented, with some reluctance, the following simple minded
benchmark numbers, probably out of deference to the Berkeley crowd
since he was scheduled to give the talk at Berkeley the next day.
He complained about the relevance of small kernals.
Time in milliseconds
Machine Language Acker Sieve Puzzle
11/780 C (VMS) 4222 116 4625
68000 Pascal 3080 246 9200
(16 MHz)
80286 Pascal 2218 168 9138
(8 MHz)
IBM RISC -- 2400 80 3800
(6.25 MHz) (PL8) (PL8) (Pascal)
Code size in bytes
11/780 152 156 2220
68000 126 228 2940
80286 --- --- ----
IBM RISC 159 202 2454
The numbers not done at IBM (780 and 286) were taken from the relevant
latest versions in Computer Architecture News.
Then, with more enthusiasm, he gave these comparisons for seven
unidentified applications programs.
10**6 instructions executed
PL8 RISC 370
1 103.0 63.7
2 171.9 107.0
3 48.8 40.0
4 88.6 71.0
5 115.8 94.4
6 49.9 38.1
7 85.3 60.5
--- ----- -----
total 663.3 474.7
overall 370/RISC = 0.715 (dynamic)
In the total number of static instructions in these 7 programs, the
IBM RISC and the 370 were about the same. Two thirds of the dynamic
excess of IBM RISC over the 370 were said to be due to the move and
multiply subroutines. The IBM RISC numbers were gathered with a
software trace. The 370 numbers were gathered with a hardware monitor
on a 370/168 which excluded all the CP/CMS system code. These programs
were said not to be commercial - no decimal - and said not to be
scientific - no floating point - but to be representative of systems
code. The average instruction size was also given: 2.85 bytes for the
IBM RISC and 3.5 for the 370.
The MMU chip supports a page size of 2048 bytes. The addresses are
byte addresses. The CPU presents 32 bits of address to the MMU, the
top four of which are a segment ID, the other 28 being an offset within
that segment. The segment registers in the MMU are 12 bits of address
that get put on the top of the 28 bit offset for a 40 bit virtual
address. The segment registers also contain a "special" bit that
controls whether the database stuff is active or not. It also contains
a "key" bit that has something to do with storage keys. I got the
impression that Al didn't want to explain this one because it was
embarrassing. The MMU keeps a 16 x 2 two way set associative translation
look aside buffer. In the monitor they are running, they map files into
segments, one file per segment. They expect most of the files in a
data base system to be relatively small so the 2**28 bytes maximum size
shouldn't be a problem. The special bit is meant to generate
interrupts to the monitor (control program in IBMese) so that locking
and journalling can be supplied by the monitor. The support is such
that one can get an interrupt on the first reference by a transaction
to a 128 byte line of data. The PL8 language has an added "Persistent"
storage class so that PL8 variables may be associated with permanent
segments (or 128 byte lines, I think):
DCL X(100) CHAR(80) PERSISTENT
OPEN (X, 'NAME', options)
X(1) = 'anything'
COMMIT (X)
Compiled code is independent of options such as "shared", "journalled",
"locks", etc. The page table in the MMU is the same organization
as used in the System 38 and gives a reversible map between
virtual and real addresses organized as a hash table as follows. If
one has N real pages (where N = 2**k for some k) then there is a 2N
entry hash table that points to an entry in the N entry page frame
table. The mth entry in the page frame table describes real page m.
That entry contains the virtual page number, a "next" pointer (for hash
collisions), a R/W bit, a TID (transaction id), and 16 lockbits, one
for each 128 byte line of a page. This whole structure was said to take
20N space, where units were not defined but I believe were bytes. It was
never suggested how big N could be. I believe N to be at least 1024.
That would mesh with the CMU talk of a workstation with "a couple of
megabytes".
Al said that, for a TLB miss, they typically got 2.2 references in the
hash table/page frame table, i.e., the probability of no chaining was
close to 90 percent. The R/W, TIB, and Lockbits apply only if the special
bit is on in the segment register. The MMU has a TIDREG (transaction id
register). An interrupt is generated by the MMU for a load/store to line N:
if TID ~= TIDREG (not equal)
if store, R/W = R or Lockbit(N) = 0
if Load, R/W = R and Lockbit(N) = 0
At this point it was decided that the talk had gotten too low level and
that the time was up anyway so we switched to questions. The
technology questions revealed that the chips are Nmos, 2 micron feature
size, with about 7.5 mm per side. Al said it does a bit better than a
4341 in raw CPU power, giving about 1.4 370 MIPS or 2 RISC MIPS. He
also said that the version they are running in Yorktown Heights has a
bug where instructions cross a page boundary and there is a fault on
both instruction and data references so they compile around it.
Additional speculation: APA displays (IBMese for "all points
addressable) have been big inside IBM for some time now and Jim Gosling
(author of the VAX Emacs screen editor) demonstrated a really nice
WYSIWYG (what you see is what you get) editor on a Sun workstation at
the recent Sun users group meeting. It wasn't available since Jim
works for IBM on the CMU project and the Sun was just their interim
development machine. Thus one would reasonably expect the ROMP's that
CMU gets to be like Sun workstations with these chips in them.
After the talk, (censored) people muttered
about how slow these chips were and how much better one could do if one
did things "right". Comments from Al pointed out that this system was
designed to be cheap and that accounted for some of the slowness of the
memory system.
Forest
∂07-Jun-84 1726 JMC
To: DFH
No pay, no quarters. I just want to give him an account and library privileges.
∂07-Jun-84 1808 JMC
To: grosof@SU-SCORE.ARPA
I dunno about IBM; I'll call you later.
∂07-Jun-84 1817 JMC
To: ARK
jmc - The following restaurants, not mentioned in previous Yumyums are quite
reasonable: A la Carte, Theo's, Fresco - all Palo Alto. Gosper recommends
Pedro's in th Stanford Shopping Center. Also Playboy (June) rates Chez Panisse
as the 4th best restaurant in the U.S.
∂07-Jun-84 1855 JMC foonly
To: golub@SU-SCORE.ARPA, bosack@SU-SCORE.ARPA
While an F1-B might be quite appropriate for LOTS, I don't think CSD should
go in that direction. If necessary, I'll elaborate.
I mention it becaue of the Stanford Report story.
∂08-Jun-84 0134 JMC
To: llw@S1-A.ARPA
.<<opinio[w84,jmc] More on Science and Technology Opinion>>
.require "memo.pub[let,jmc]" source;
.cb Announcing %2Science and Technology Opinion%1
%2Science and Technology Opinion%1 is a new journal dedicated
to developing public opinion in the science and technology
communities. It will be edited and written by scientists and engineers rather
than by journalists. Submissions of articles and letters on
public issues related to science and technology are solicited.
A certain amount of commentary by scientists and technologists
on general public issues even where science is not explicitly
involved will also be published.
The editors will normally decide themselves whether an article
will be published, but referees will be used when it seems appropriate
to the editors or is requested by an author.
The editors hope to achieve a higher standard of objectivity and
fairness than is usual in general publications and will not hesitate
to require changes in articles if required for this. Authors are expected
to try to make their articles comprehensible to a wide audience, but
technical material is welcome when relevant.
The editor is: -- Todorovich
The publisher is:
Editorial and subscription correspondence address:
6 issues per year at a price of:
Contents of first issue: (None of the authors mentioned has been asked yet.
This is the kind of thing we might hope for. The actual announcement will
contain reference to articles that have been written and accepted).
Indoor pollution, the neglected hazard: Henry Hurwitz
Towards a scientific public opinion: the editors
Scientists and the Common Defense: Edward Teller
Issues concerning artificial hearts: a symposium
Spending too much on a safety measure may cost lives: John McCarthy
∂08-Jun-84 0135 JMC
To: llw@S1-A.ARPA
opinio[f83,jmc] Science and Technology Opinion
1983 December 18
Science and Technology Opinion is to be a journal, a monthly
would be desirable, but a quarterly may be more feasible, that
would provide a forum for other than liberal views about science.
It would be written by scientists and would counteract
the journalists who run the News and Comment section of
Science.
The project has been discussed with Miro Todorovich, David
and Gregory Chudnovsky, and Irving Kristol. David thinks
working capital of $250,000 is required. Kristol thinks
a quarterly can be done for $100,000 or even for $50K to $60K
per year subsidy provided the editors are amateur. He suggests
talking to Walter Lammi of the Institute for Contemporary Studies
in San Francisco. Miro will talk to him. Miro is interested in
being editor. SE2 and UCRA might be sponsors.
!Draft statement of purpose
Science and Technology Opinion will provide %2news%1 about
scientific and technological events affecting the public interest
in general and science policy in particular. It will also provide
a forum for %2comment%1, especially by scientists and engineers
on policy issues. News and comment will be kept separate, and
an appropriate high standard fairness in both will be maintained
and enforced by the editors. A major objective is to raise
the standards for discussion of social and political issues
towards those already achieved in discussion of technical scientific
issues.
The atmosphere of public discussion of scientific and
technological issues suffers from moralistic intimidation.
This has produced an inhibition of rational discussion of
numerous issues. For example, rational discussion of the level
of damage that would be caused by nuclear war and how it might
be mitigated by civil defense is almost non-existent by arguments
that anyone who asserts that certain bad things might not happen
is in favor of nuclear war. Similar inhibitions affect discussions
of the nature and inheritance of intelligence and the nature of the
differences between the sexes. While Science and Technology Opinion
will not over-emphasize such issues, it won't avoid them either.
The editorial board is not a random sample of scientific
opinion and wishes to announce its collective opinion in certain
matters, although we still welcome contributions from people who
are not in complete agreement.
#. Science provides the main way of understanding the world
including the social world.
#. Science is important both as an expression of the collective
curiosity of mankind and for its contributions to human health,
longevity, prosperity, public order and enjoyment. These benefits
are realized through technology.
#. Human progress since the renaissance is real, has been
beneficial to humanity, and can and should continue.
#. Technology has been the prime contributor to human progress
and there is much more to come.
#. Social science is very difficult, but successful and applicable
social science is possible and some even exists. However, it is
necessary to be modest and tentative in asserting the likelihood that
any particular strategy for solving a social problem will work. It is
even necessary to be modest and tentative about one's identification
of the problems.
#. The defense of the United States is important, and scientists
and technologists should d their best to increase its effectiveness and
reduce its cost.
#.
!Possible contents
Forum on CO2.
Forum on the effects of nuclear war.
Forum on "nuclear winter".
Forum on the facts about racial, sexual and genetic equality.
Biases in scientific journalism.
Quality standards for scientific journalism.
When aren't safety measures cost effective.
Indoor pollution
Political intrusions on scientific consensus.
(The New York City Council has ruled that Haitians are not especially
at risk from AIDS).
Appropriate technology
Forum on what level and kind of secrecy will help the country.
Forum on Star Wars.
The state of nuclear energy
The search for alternate energy sources
Prospects for the future in computing, AI, energy, manufacture
Forum on what would be good to invent. Announce with quotes from
Gabor and Stent
The evidence on yellow rain.
Verification of arms agreements
Strengths and weaknesses of Soviet science - Chinese.
History of Marxist attitudes to science
Reviews of Gould and other books where science and ideology interact
The importance of space exploration to humanity
Why more research on nuclear weapons
Spelling out why the U.S. should be supported in its conflicts with the
USSR.
Politics and institutions related to the support of science.
Are scientists hostile to technology? History of this hostility.
Civil defense
Moralistic bullying vs. rationality. Attacks on rationality.
Forum on the prospects for longer human life. (Prof. Roy Walford, UCLA).
!possible contributors
Bobby Inman
Daniel Graham
Freeman Dyson
Edward Teller
Henry Hurwitz
Bernard Cohen
Bernard Davis
Andrei Sakharov
Zhores Medvedev
John McCarthy
Igor Shafarevich
David Chudnovsky
Gregory Chudnovsky
Tom Connolly
George Keyworth
Hans Mark
Hans Bethe
Frederick Seitz
Bruce Ames
Petr Beckmann
Arthur Kantrowitz
Lowell Wood
Robert Forward
Jerry Pournelle
Richard Meehan
Elizabeth Pate-Cornell
Chauncey Starr
William Shockley
Arthur Jensen
Hans Eysenck
Roy Walford
∂08-Jun-84 0901 JMC
To: CLT
The card is under the checkbook in the drawer.
∂08-Jun-84 1551 JMC visit
To: AGE.YKTVMT.IBM@CSNET-RELAY.ARPA,
BLASGEN.YKTVMT.IBM@CSNET-RELAY.ARPA
Unfortunately, Dick Gabriel found out that his band has to play
Friday evening and can't come. Therefore, I propose to reschedule
our proposed Friday visit for some time later. I'll phone.
∂08-Jun-84 1826 JMC
To: llw@S1-A.ARPA
The Chudnovskys are U.S. citizens as of last Fall.
∂08-Jun-84 1927 JMC
To: DFH
Please phone Nancy Steege at 7-2547 and tell her that I intend to eat
all four free meals at the CSLI Workshop in practical reasoning
∂08-Jun-84 2244 JMC
To: faculty@SU-SCORE.ARPA, library@SU-SCORE.ARPA
jmc - Arthur, since our library has gone into hopeless chicken shit mode, it
is not worthwhile to comment about trivia.
∂09-Jun-84 1716 JMC re: Computer Use Seen As Outpacing Legal Safeguards (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - This is one of the what-if-ier pieces of superstition I have seen
in some time. The idea that a company or the Government achieves control
over people by collecting trivial personal information is like the Voodoo
believe that power over someone is obtained by fingernail clippings and
hair. Like many such articles, it refers to "increasing concern in the
last few years". The what-if articles were at least equally prevalent
in the late 1950s and early 1960s, but the speculations were wilder,
because the authors new less about computers. The country will suffer
more from the "public policy community"'s need to find ever new
menaces to hold conferences about, get contracts to study, write articles about,
file lawsuits about, and get jobs regulating. New legal ground was broken
for the public policy community last year when the Sierra Club lost a
a lawsuit, but the judged ordered some utility to pay them $90,000 legal
expenses, because the case raised interesting legal issues.
∂10-Jun-84 0152 JMC (→16460 16-Jun-84)
To: "#___JMC.PLN[2,2]"
At M.I.T. Endicott House till June 16.
∂16-Jun-84 0000 JMC Expired plan
To: JMC
Your plan has just expired. You might want to make a new one.
Here is the text of the old plan:
At M.I.T. Endicott House till June 16.
∂16-Jun-84 1738 JMC
To: DFH
I have been unable to reach George Johnson, and I give the 10am
meeting at SRI higher priority. I expect to be back from there
about 1pm.
∂16-Jun-84 1740 JMC
To: bscott@SU-SCORE.ARPA
I will be available most times the week of June 18. Fran has my calendar.
∂16-Jun-84 1741 JMC
To: ken%oz@MIT-MC.ARPA
I'd prefer the readable version.
∂16-Jun-84 1742 JMC
To: YOM
Please recommend a grade.
∂16-Jun-84 1749 JMC
To: VRP@SU-AI.ARPA, buchanan@SUMEX-AIM.ARPA
CC: golub@SU-SCORE.ARPA
I still get very angry about the loss of keys every time it comes to
mind. The last time it came to mind was when I was at IBM Yorktown
and found their library available at midnight. It has made Stanford
much less attractive to me at no demonstrable gain to Harry Llull
except to his sense of proprietorship. Unless I get a key I am
considering refusing service on certain committees.
∂16-Jun-84 2004 JMC re: DoD plans for Software Engineering Institute (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - I doubt that any part of Stanford would express interest in
anything that involved having a secure building on the Stanford
campus.
∂17-Jun-84 2356 JMC
To: DFH
Please print PRESS.NS[1,JMC] and mail it to my daughter Susan; then delete it.
∂18-Jun-84 0008 JMC axioms
To: LEP
Putting a bit more effort into reading the axioms made them clear enough.
I don't know quite what to think about the Prolog style you are using.
Perhaps I'll have a comment later. 2.1 seems to be missing an & symbol.
∂18-Jun-84 0030 JMC meeting about Foonly
To: bosack@SU-SCORE.ARPA
Unfortunately, I have another meeting at that time. However, I'm
still opposed to CSD acquiring one.
About another matter. I had a discussion with Dave Cheriton, who
would be interested in having CSD-CF maintain Unix for Vaxen. This
would involve hiring a Unix wizard and would require getting enough
of the Vaxen to pay for it. It seems to me that the right number
is that which corresponds to the number of users of SCORE or SAIL
so that the maintenance expenses would correspond to about the
same per user.
When you have time, let's discuss it. I have a list of people
who might be invited to a meeting about it.
∂18-Jun-84 0140 JMC
To: rwg@MIT-MC.ARPA
I assume Knopp eventually arrived.
∂18-Jun-84 0843 JMC
To: bscott@SU-SCORE.ARPA
I will be in this afternoon.
∂18-Jun-84 0855 JMC
To: buchanan@SUMEX-AIM.ARPA
Please don't forward to me messages from Llull.
∂18-Jun-84 1256 JMC
To: DFH
Gunther on 12th St., San Jose
∂18-Jun-84 1910 JMC prioritized circumscription
To: LEP@SU-AI.ARPA, grosof@SUMEX-AIM.ARPA
priori[s84,jmc] and priori.lsp[s84,jmc] are an exposition of the
use of ordering to rehabilitate Leslie's style of axioms and an
ekl version of these axioms. They should be looked at carefully
before anyone attempts to do the ekl proof that
(all x (equiv (flies0 x) (bird x))).
Ben, if you feel an urge to hack ekl, I'll arrange for a SAIL account.
∂18-Jun-84 2333 JMC re: Article on BBOARD abuse (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - Maybe they suspect that he knew the number was on his bulletin
board and didn't promptly delete it. Whether that would be a misdemeanor
isn't obvious to me. Another issue is how long police can hold something
seized as evidence. When a camera lens stolen from my house was found
in the home of a fence, the police didn't return it to me for about a
year. I would guess that this is a matter of tradition and common law
rather than statute law.
∂19-Jun-84 1709 JMC
To: aaai-office@SUMEX-AIM.ARPA
agenda[s84,jmc] Agenda for AAAI meeting
Claudia: Here is what I presently have in mind. I would like to think
a bit before making this final. Also I would like to know about any
other items you think should be covered. After this much is fixed, please
ask Woody for his items.
The first part of the meeting will be chaired by John McCarthy
who will then turn the chair over to Woody Bledsoe who will continue
with his agenda items.
Part I of the Agenda
1. Report on workshops sponsorship by AAAI - John McCarthy
2. The AI Magazine -
3. Administration of the AAAI - Claudia Mazzetti
4. Finances - Rich Fikes
5. The on-line library and abstracts project. - Mike Genesereth
6. Proposal to make president's term two years - John McCarthy
∂20-Jun-84 0034 JMC Keys for library
To: golub@SU-SCORE.ARPA
I am often in the office late at night, and somehow it seems that
this is the most common time when I want something from the library.
The ability to use it late at night was one of the more valuable
aspects of being at Stanford. Many thanks for putting on the pressure
for this.
∂20-Jun-84 1410 JMC
To: RWW
∂20-Jun-84 1354 DFH
To: JMC, DFH
Dan Slotnick (?) from U. of Illinois called. Asked me to take this
message:
Looking for a section head for the concurrent processing and AI section in
the center at Goddard, on whose advisory committee you agreed to serve.
Job is slated at GSl4-l5. Can go up to the mid 50's.
Looking for a PhD in AI-concurrent processing with some experience who can
direct a small good research group and be active himself.
If you could suggest to me with a phone number, give me a call.
301 338 7518 at J. Hopkins where he is for summer.
If you can solicit a resume, even better. Please send it.
∂20-Jun-84 1856 JMC
To: bscott@SU-SCORE.ARPA
I have two summer graduate students in 353.
∂20-Jun-84 2031 JMC Veronica Dahl workshop
To: aaai-office@SUMEX-AIM.ARPA
Claudia:
Veronica Dahl has written me giving adequate reason for
contributing $2200 to her workshop about which she asked previously.
I'll send you a copy of the letter. Unfortunately, the present
letter doesn't repeat the name of the workshop. Do you have
correspondence about it? If not please phone her at Simon Fraser
University in Burnaby, B.C Canada. The number is given as 291-4277
and I suppose the area code is 604, since the university is has
been located in Vancouver, so I suppose Burnaby is a suburb.
∂20-Jun-84 2033 JMC
To: DFH
Please ask me about Livermore bill.
∂21-Jun-84 0240 JMC re: softwar, a new weapon to deal with the Soviets ? (from SAIL's BBOARD)
To: OTHER-SU-BBOARDS@SU-AI.ARPA
jmc - I believe this article is a work of imagination. Indeed I don't
even believe in the existence of the book referred to.
∂21-Jun-84 1718 JMC
To: bscott@SU-SCORE.ARPA
I think Gertrud is the best of the lot and would be satisfactory. So
please check her references and make sure she understands about 40 hours.
∂21-Jun-84 1928 JMC Homebrew Computer Club Meets 6/20
To: dlw@MIT-MC.ARPA
Here is an item that might interest you. I have some questions about
the Symbolics Dialnet. Will old 3600s be upgradable to your new modem?
What is the name of the new modem? Does your Dialnet follow the Crispin
protocols or is it entirely new?
∂21-Jun-84 1446 CSL.ALLISON@SU-SIERRA.ARPA Homebrew Computer Club Meets 6/20
Received: from SU-NAVAJO.ARPA by SU-AI.ARPA with TCP; 21 Jun 84 14:46:37 PDT
Received: from Diablo by Navajo with TCP; Thu, 21 Jun 84 14:41:34 pdt
Received: from SU-SIERRA.ARPA by Diablo with TCP; Thu, 21 Jun 84 14:38:13 pdt
Date: Thu 21 Jun 84 14:39:39-PDT
From: Dennis Allison <CSL.ALLISON@SU-SIERRA.ARPA>
Subject: Homebrew Computer Club Meets 6/20
To: bayboards@SU-AIMVAX.ARPA, su-bboards@SU-SIERRA.ARPA,
bboard%LOTS-A@SU-SIERRA.ARPA
HOMEBREW COMPUTER CLUB
Turing Auditorium -- Jordan Quad -- Stanford University
7:30 -- 10:30 pm ** June 20, 1984
MEETING LOCATION -- This month's meeting is scheduled in the Turing Auditorium,
Poly Hall Room 111 located at the Jordan Quad, Stanford University. Meeting
date is June 20-th. Turing Auditorium is an excellent facility with video
display terminals located throughout the auditorium.
SPEAKER: Dr. Hank Magauski of Gamma Technology, Inc.
TOPIC: 9600 Bit/Second Modem. Gamma Technology has developed a 9600
bit/second modem board for the IBM PC and IBM PC-XT. This modem works
on dial-up telephone lines and has several modes of operation.
On-screen graphics allow the user to program the modem easily. This
is a very advanced product that utilizes the telephone dial-up service
very efficiently. Dr. Magnuski will discuss this new product and also
explain some of the problems in high speed data communications on
ordinary telephone lines.
Dr. Magnuski has many years experience in computer technology and
communications systems. His hobbies include amateur radio. He set up
the first amateur packet radio system in this area and he is currently
very active in amateur satellite communications experiments.
-------
∂22-Jun-84 0557 JMC arriving Sunday
To: simonds@CMU-CS-C.ARPA
Carolyn and I will arrive Sunday and check into the Bristol, and
I'll come in Monday am.
∂22-Jun-84 0933 JMC clt
To: RWW
Carolyn was in the office all night and just got home and went to bed.
She'll call you when she's fit.
∂22-Jun-84 1427 JMC Marseilles
To: simonds@CMU-CS-C.ARPA
I would propose to spend two days there arriving one morning
and leaving the following evening. Which days are at their
(i.e. Mike Griffiths's) and your convenience.
∂22-Jun-84 1516 JMC prioritized circumscription
To: pack@SRI-AI.ARPA
The current section of circum[f83,jmc] on prioritized circumscription
now seems to contain the formalism required to avoid "sneak interpretations"
of your tower axioms. I'll be away for a week, and I'll leave a copy of
the current version of the paper for you with Fran. I am hoping to improve the
priority section and put it into a form that keeps the axioms involving
different phenomena including the axioms about priorities in a suitably
modular form.
∂22-Jun-84 1943 JMC (→16477 1-Jul-84)
To: "#___JMC.PLN[2,2]"
Carolyn and I will be in Paris and Marseilles leaving Saturday June 23
and arriving back Sunday July 1. We can be reached in case of necessity
at the Hotel Bristol or at the Centre Mondial. Their phone numbers are,
counting foreign access,
(Bristol Hotel, Paris 011 33 1 266-9145)
(Centre Mondial: 011 33 1 268-1100).
∂22-Jun-84 2338 JMC flight time
To: RWW
In case Sarah can't pick us up, we will arrive Sunday, July 1 on
PA125 at 3:25pm. Needless to say, you should check the actual
arrival time before going to the airport.