perm filename MAY76.IN[LET,JMC] blob
sn#217770 filedate 1976-06-02 generic text, type C, neo UTF8
COMMENT ⊗ VALID 00242 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00024 00002 ∂27-FEB-76 1103 FTP:FEIGENBAUM at SUMEX-AIM RETRANSMISSION OF MSG
C00027 00003 ∂27-FEB-76 1715 WD
C00028 00004 ∂27-FEB-76 1824 CCG AI-Qual
C00029 00005 ∂28-FEB-76 0624 REM
C00030 00006 ∂28-FEB-76 1132 FTP:FEIGENBAUM at SUMEX-AIM (Response to message)
C00032 00007 ∂02-MAR-76 0803 LCW PDP-10 FORTRAN
C00034 00008 ∂03-MAR-76 1041 LES ARPA Final Report
C00036 00009 ∂04-MAR-76 0448 REM
C00040 00010 ∂04-MAR-76 0651 JRA disappearance
C00041 00011 ∂04-MAR-76 2111 REM via AMET FORMAT OF SPINDLE AND METHOD OF UPDATING, PROPOSAL
C00044 00012 ∂05-MAR-76 0233 DCL Frieder von Henke
C00046 00013 ∂05-MAR-76 1202 REM via AMET MAXIMUM SIZE OF A SPINDLE
C00047 00014 ∂05-MAR-76 1301 REM via AMET SPINDLE FORMAT PROPOSAL, EDITED
C00048 00015 ∂07-MAR-76 2315 REM via AMET Keeping .MSG[2,2] files in E format
C00050 00016 ∂08-MAR-76 0547 REM via AMET Crunching POX.FAI
C00051 00017 ∂08-MAR-76 0559 REM via AMET Addends to crunching POX.FAI
C00052 00018 ∂08-MAR-76 0608 REM via AMET More
C00053 00019 ∂10-MAR-76 1448 100 : QUEENIE TALKS IN CS224
C00054 00020 ∂10-MAR-76 1451 100 : QUEENIE CARL ENGLEMAN'S VISIT FROM MITRE CORPORATION
C00055 00021 ∂10-MAR-76 1540 DCL
C00057 00022 ∂11-MAR-76 0644 FTP:PRATT at MIT-AI
C00058 00023 ∂11-MAR-76 1057 FTP:ALAN COLE(A617AC01)@CMUB o: JMC@SAIL
C00060 00024 ∂11-MAR-76 1304 FTP:BEN at MIT-AI ''Computer Power and Human Reason''
C00061 00025 ∂11-MAR-76 1736 FTP:TK at MIT-AI
C00062 00026 ∂12-MAR-76 2241 WTL 226
C00063 00027 ∂13-MAR-76 0017 REM via AMET C.R. for .FAI files
C00065 00028 ∂13-MAR-76 0126 REM via AMET More statistics on c.r. for .FAI[S,SYS]
C00067 00029 ∂15-MAR-76 0122 REM Crunch and Spindle, a reply.
C00069 00030 ∂14-MAR-76 2353 JMC crunching and spindling
C00071 00031 ∂16-MAR-76 0641 JRA lisp book
C00072 00032 ∂16-MAR-76 1417 DRB
C00074 00033 ∂17-MAR-76 1022 REF via CMUA
C00077 00034 ∂17-MAR-76 1219 FTP:MINSKY at MIT-AI
C00078 00035 ∂17-MAR-76 1234 FTP:MINSKY at MIT-AI
C00079 00036 ∂18-MAR-76 1357 REM via AMET
C00080 00037 ∂18-MAR-76 2113 LES
C00082 00038 ∂19-MAR-76 1202 FTP:MINSKY at MIT-AI
C00083 00039 ∂19-MAR-76 1203 FTP:MINSKY at MIT-AI
C00084 00040 ∂19-MAR-76 1904 FTP:LEDERBERG at SUMEX-AIM book review
C00085 00041 ∂20-MAR-76 0110 ME
C00086 00042 ∂21-MAR-76 2049 REM HOTER.BAY
C00091 00043 ∂21-MAR-76 2106 KRD errata in your weizenbaum review
C00095 00044 ∂24-MAR-76 2046 JMC
C00097 00045 ∂25-MAR-76 0901 JMC*
C00098 00046 ∂25-MAR-76 0136 JMC
C00099 00047 ∂26-MAR-76 0722 REG
C00104 00048 ∂26-MAR-76 0727 REG
C00107 00049 ∂27-MAR-76 0256 FTP:MASINTER at PARC-MAXC file-understanding
C00110 00050 ∂28-MAR-76 1513 FTP:MARG at MIT-AI
C00111 00051 ∂28-MAR-76 2050 REM via AMET Progress update on Spindle project
C00112 00052 ∂29-MAR-76 2006 FTP:RMS at MIT-AI
C00113 00053 ∂29-MAR-76 2020 FTP:Jrobinson at SRI-AI Dinner Wednesday
C00114 00054 ∂29-MAR-76 2044 DSB
C00115 00055 ∂30-MAR-76 0140 LES Forrest Howard
C00116 00056 ∂30-MAR-76 0856 FTP:BEN at MIT-AI
C00118 00057 ∂30-MAR-76 1959 FTP:LEDERBERG at SUMEX-AIM WEIZENBAUM book, p.s. [jmc,eaf
C00119 00058 ∂30-MAR-76 2003 DSB
C00120 00059 ∂30-MAR-76 2022 DSB chess
C00121 00060 ∂30-MAR-76 2044 DSB chess
C00123 00061 ∂30-MAR-76 2156 WTL 224 readings
C00124 00062 ∂31-MAR-76 0935 FTP:MINSKY at MIT-AI
C00125 00063 ∂31-MAR-76 1651 DCL EXAMPLES FOR VERIFICATION
C00126 00064 ∂31-MAR-76 1953 FTP:CERF at USC-ISI Fancy CRT's
C00128 00065 ∂02-APR-76 1533 DCO Verification Seminar
C00129 00066 ∂03-APR-76 1016 DEK
C00133 00067 ∂03-APR-76 2133 HPM quack1
C00134 00068 ∂04-APR-76 1317 FTP:MINSKY at MIT-AI
C00135 00069 ∂05-APR-76 0652 FTP:MINSKY at MIT-AI
C00136 00070 ∂05-APR-76 0859 FTP:REM at MIT-MC
C00137 00071 ∂05-APR-76 1039 RWW FOL
C00138 00072 ∂05-APR-76 1917 WTL IEEE Proceedings
C00139 00073 ∂05-APR-76 1929 WTL CS224 Lecture
C00141 00074 ∂06-APR-76 1059 FTP:STEFIK at SUMEX-AIM Lighthill Film
C00142 00075 ∂07-APR-76 0059 REM via AMET Spindle status
C00143 00076 ∂07-APR-76 1020 REG
C00144 00077 ∂07-APR-76 1416 KRD via ML WEIZENBAUM
C00145 00078 ∂08-APR-76 0727 100 : Weizenbaum via MITT Mental←qualities
C00148 00079 ∂08-APR-76 1245 REG "VERIFY"
C00150 00080 ∂08-APR-76 1542 LES
C00152 00081 ∂09-APR-76 0903 FTP:BUCHANAN at SUMEX-AIM Robert Blum
C00153 00082 ∂09-APR-76 1125 REG
C00154 00083 ∂12-APR-76 0011 DWP Midnight Maurading
C00155 00084 ∂12-APR-76 1008 REM via AMET Progress report on spindle
C00156 00085 ∂12-APR-76 1735 PLW side effects paper
C00159 00086 ∂12-APR-76 1945 DCO Verification Seminar
C00161 00087 ∂13-APR-76 0641 REM via AMET Progress rpt.
C00162 00088 ∂13-APR-76 0821 REG
C00163 00089 ∂13-APR-76 1350 LES
C00165 00090 ∂13-APR-76 1442 RAK Thursday talk
C00166 00091 ∂13-APR-76 1446 REM via AMET CRU3.FAI fixed!
C00168 00092 ∂13-APR-76 1553 100 : REM via AMET
C00169 00093 ∂13-APR-76 1601 FTP:REM at MIT-MC More info about CRU3
C00170 00094 ∂13-APR-76 1604 FTP:REM at MIT-MC
C00171 00095 ∂13-APR-76 1658 JMC
C00172 00096 ∂14-APR-76 0105 REM via AMET Bug I found in CRU3
C00174 00097 ∂14-APR-76 0108 REM via AMET Proposed preliminary default for trees
C00175 00098 ∂14-APR-76 0115 JMC
C00179 00099 ∂14-APR-76 1501 RLD TIME CHANGE FOR THURSDAY TALK
C00181 00100 ∂14-APR-76 1619 FTP:PHW at MIT-AI
C00183 00101 ∂14-APR-76 1656 FTP:PHW at MIT-AI
C00184 00102 ∂14-APR-76 1757 DCL
C00185 00103 ∂14-APR-76 2347 DCO THURSDAY SEMINAR
C00186 00104 ∂15-APR-76 1028 TOB arpa image understanding
C00188 00105 ∂15-APR-76 1427 FTP:PHW at MIT-AI
C00189 00106 ∂15-APR-76 1627 REM via AMET
C00190 00107 ∂15-APR-76 1822 REM via AMET
C00191 00108 ∂15-APR-76 2321 REM via AMET That size of crunched file bug (total size of spindle)
C00192 00109 ∂16-APR-76 0123 REM via AMET New CRU3.DMP[1,REM]
C00193 00110 ∂16-APR-76 0211 REM via AMET
C00194 00111
C00195 00112 ∂16-APR-76 1016 FTP:gjs@mit-ai at MIT-AI
C00196 00113 ∂17-APR-76 0712 RWG
C00197 00114
C00198 00115 Can you send me a copy of your draft paper on computer message
C00210 00116 ∂18-APR-76 0742 FTP:JOSEPH at MIT-MC
C00214 00117 ∂18-APR-76 2348 KRD outlandish idea
C00219 00118 ∂19-APR-76 0741 FTP:PANKO at OFFICE-1 (Response to message)
C00221 00119 ∂20-APR-76 0514 REM via AMET BUG IN SYSTEM MAKING CRU3 BREAK!!!!
C00223 00120 ∂20-APR-76 0731 FTP:REM at MIT-ML
C00224 00121 ∂21-APR-76 1140 DCO THURSDAY VERIFICATION SEMINAR
C00225 00122 ∂21-APR-76 1544 FTP:REM at MIT-MC
C00228 00123 ∂21-APR-76 2136 KRD distinction
C00230 00124 ∂22-APR-76 0723 FTP:BEN at MIT-AI Book recommendation
C00233 00125 ∂22-APR-76 1322 FTP:BEN at MIT-AI
C00237 00126 ∂22-APR-76 1507 REG Benchmarks
C00238 00127 ∂22-APR-76 1639 REM Beehive programs running on Datamedia in ME's office easy!
C00240 00128 ∂23-APR-76 1252 LES Benchmarks
C00241 00129 ∂23-APR-76 1951 DCL
C00242 00130 ∂23-APR-76 2320 REM via AMET
C00243 00131 ∂24-APR-76 0036 REM via AMET Double and Triple occurrance words
C00245 00132 ∂24-APR-76 0147 FTP:REM at MIT-DMS
C00246 00133 ∂24-APR-76 0323 REM via AMET Success at making CRU5.SAI optimize output so that SSORT doesn't thrash.
C00247 00134 ∂24-APR-76 1029 FTP:SIGART(X180LE03) at CMUB Weizenbaum's review, etc.
C00250 00135 ∂24-APR-76 1529 DCL
C00252 00136 ∂24-APR-76 1623 REM via AMET New SSORT desirable
C00253 00137 ∂24-APR-76 1638 FTP:REM at MIT-MC
C00255 00138 ∂24-APR-76 1640 FTP:REM at MIT-MC
C00258 00139 ∂24-APR-76 1737 REM "ONE COPY OF EACH WORD"
C00261 00140 ∂24-APR-76 1932 REM CRU4.WRU
C00265 00141 ∂24-APR-76 2147 REG
C00267 00142 ∂26-APR-76 0834 REM via AMET Working on word-dictionary crunching scheme...
C00269 00143 ∂27-APR-76 0910 REG
C00270 00144 ∂27-APR-76 1424 DML Byte magazine
C00271 00145 ∂29-APR-76 0011 DCL
C00272 00146 ∂29-APR-76 0539 RWG at TTY41 0539
C00273 00147 ∂29-APR-76 0645 REM via AMET First results for word-dictionary crunching algorithm
C00274 00148 ∂29-APR-76 0734 REM via AMET
C00275 00149 ∂29-APR-76 0738 REM via AMET
C00276 00150 ∂29-APR-76 0914 MLM AUTOLOGOUT
C00277 00151 ∂29-APR-76 0923 FTP:REM at MIT-MC
C00279 00152 ∂29-APR-76 1018 BPM ~autologout(x)
C00280 00153 ∂29-APR-76 1910 RDA AUTOLOGOUT
C00281 00154 ∂30-APR-76 0723 REM via AMET Finishing touch on word-oriented crunching estimates for HOTER.ESS
C00283 00155 ∂30-APR-76 1551 RAK via AMET Autologou
C00285 00156 ∂30-APR-76 1949 ZM referees
C00286 00157 ∂30-APR-76 1140 JMC
C00287 00158 ∂01-MAY-76 0958 REM Improvement in CRU4.SAI caching algorithm
C00289 00159 ∂01-MAY-76 1301 REM Final results for word-dictionary crunch on NOTICE[UP,DOC]
C00291 00160 ∂01-MAY-76 1316 REM DIALOG
C00293 00161 ∂02-MAY-76 0428 REM
C00294 00162 ∂02-MAY-76 0833 FTP:JOSEPH at MIT-MC
C00298 00163 ∂02-MAY-76 1228 REG CSD Comprehensive Exam
C00299 00164 ∂02-MAY-76 1336 FTP:JOSEPH at MIT-MC
C00302 00165 ∂02-MAY-76 1636 REM via AMET Results for WUTHER
C00304 00166 ∂02-MAY-76 1742 REM via AMET Addenda
C00305 00167 ∂02-MAY-76 1916 REM via AMET
C00306 00168 ∂03-MAY-76 0420 100 : DON Auto-logout
C00307 00169 ∂02-MAY-76 2220 JMC spindle and crunch
C00309 00170 ∂02-MAY-76 2210 JMC
C00310 00171 ∂03-MAY-76 0550 REM via AMET Timing for CRU3 (actually C3, the new version)
C00311 00172 ∂03-MAY-76 0911 REG
C00312 00173 ∂03-MAY-76 0917 REG
C00313 00174 ∂04-MAY-76 0628 FTP:JM at MIT-MC
C00314 00175 ∂05-MAY-76 0932 FTP:Dehall at SRI-AI Abstracts and Papers
C00319 00176 ∂05-MAY-76 1702 RCM dinner invitation
C00321 00177 ∂06-MAY-76 0339 REM via AMET New features in CRU3
C00324 00178 ∂06-MAY-76 0352 REM via AMET New version of CRU3.DMP[1,REM] new spindle program.
C00326 00179 ∂06-MAY-76 2233 100 : REM via AMET
C00327 00180 ∂07-MAY-76 0007 REM via AMET Bug in CRU3.DMP from 76.5.06 0338 to 76.5.07 0005
C00328 00181 ∂07-MAY-76 0134 REM via AMET New CRU3, hopefully with all bugs fixed...
C00329 00182 ∂07-MAY-76 1324 REM via AMET
C00330 00183 ∂07-MAY-76 1343 WD lock
C00331 00184 ∂08-MAY-76 0142 JMC bug
C00335 00185 ∂08-MAY-76 1055 REM via AMET
C00337 00186 ∂08-MAY-76 2322 FTP:REM at MIT-ML
C00341 00187 ∂09-MAY-76 0145 REM via AMET
C00342 00188 ∂09-MAY-76 0147 REM via AMET Proposed entry in NEWS = NOTICE[UP,DOC]
C00343 00189 ∂09-MAY-76 0204 REM via AMET Enhancements to SPINDL proposed by JFR
C00344 00190 ∂08-MAY-76 1230 JMC
C00346 00191 ∂09-MAY-76 1412 FTP:JOSEPH at MIT-MC
C00353 00192 ∂09-MAY-76 1441 FTP:JOSEPH at MIT-MC
C00355 00193 ∂09-MAY-76 1633 FTP:JOSEPH at MIT-MC
C00357 00194 ∂09-MAY-76 1705 FTP:JOSEPH at MIT-MC
C00359 00195 ∂09-MAY-76 1749 FTP:JOSEPH at MIT-MC
C00360 00196 ∂09-MAY-76 1756 FTP:JOSEPH at MIT-MC
C00361 00197 ∂09-MAY-76 1844 FTP:JOSEPH at MIT-MC
C00362 00198 ∂09-MAY-76 2151 WD spindl
C00363 00199 ∂09-MAY-76 2312 REM via AMET
C00364 00200 ∂09-MAY-76 2354 100 : REM via AMET @LICKLI.LE6[LET,JMC]
C00365 00201 ∂10-MAY-76 1034 FTP:JOSEPH at MIT-MC
C00366 00202 ∂10-MAY-76 1234 REM via AMET
C00368 00203 ∂10-MAY-76 1243 JMC
C00370 00204 ∂09-MAY-76 2331 JMC
C00371 00205 ∂09-MAY-76 1209 JMC
C00372 00206 ∂11-MAY-76 0937 FTP: Dertouzos at MIT-Multics
C00373 00207 ∂14-MAY-76 0420 FTP:GARY GOODMAN(A610GG01) at CMUB thesis update
C00378 00208 ∂15-MAY-76 1305 REM ARCHIVES AT MIT-DM
C00380 00209 ∂16-MAY-76 0043 BZM Identification
C00382 00210 ∂16-MAY-76 0339 REM SPINDL.REM
C00383 00211 ∂16-MAY-76 0349 REM P.S.
C00384 00212 ∂17-MAY-76 1041 QIB New Telephone Number
C00385 00213 ∂18-MAY-76 0622 REM SSORT (String-Sort)
C00386 00214 ∂18-MAY-76 0810 FTP:SIGART(X180LE03) at CMUB[Lee] Weizenbaum review.
C00387 00215 ∂18-MAY-76 1210 EJG via AMET System Downtime
C00388 00216 ∂18-MAY-76 1427 EJG via AMET System downtime (revised)
C00389 00217 ∂18-MAY-76 2104 FTP:CARL at MIT-AI
C00390 00218 ∂19-MAY-76 0807 PJ
C00391 00219 ∂20-MAY-76 0653 REM
C00392 00220 ∂20-MAY-76 1345 FTP:GLS at MIT-AI
C00393 00221 ∂20-MAY-76 1430 REG
C00394 00222 ∂21-MAY-76 0725 100 : David Roode via AMET DEC-20 [LOTS]
C00395 00223 ∂22-MAY-76 0910 JRA lisp
C00398 00224 ∂23-MAY-76 1216 WD Lock
C00399 00225 ∂24-MAY-76 1230 FTP:SIGART(X180LE03) at CMUB Weizenbaum review.
C00400 00226 ∂24-MAY-76 2230 FTP:SIGART(X180LE03) at CMUB Acknowledgment of receipt of Weizenbaum review.
C00401 00227 ∂24-MAY-76 2306 RSC Cartwright's Thesis Orals
C00402 00228 ∂24-MAY-76 2353 LES
C00408 00229 ∂25-MAY-76 1313 FTP:SIGART(X180LE03) at CMUB Yet more on the JW review.
C00410 00230 ∂25-MAY-76 1319 FTP:Omalley at SRI-AI MEETING
C00412 00231 ∂25-MAY-76 1344 100 : DREW HI THERE
C00413 00232 ∂25-MAY-76 1501 FTP:Ray Reiter Theorem Proving Meeting
C00414 00233 ∂26-MAY-76 0625 100 : REM via AMET What I have been up to
C00417 00234 ∂26-MAY-76 2205 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
C00419 00235 ∂26-MAY-76 2218 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
C00421 00236 ∂26-MAY-76 2242 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
C00423 00237 ∂26-MAY-76 2230 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
C00425 00238 ∂26-MAY-76 2252 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
C00427 00239 ∂26-MAY-76 2311 FTP:SIGART(X180LE03) at CMUB[Lee Erman] Weizenbaum's Response.
C00459 00240 ∂27-MAY-76 0153 KRD
C00460 00241 ∂29-MAY-76 0722 WD
C00462 00242 ∂31-MAY-76 1409 FTP:SIGART(X180LE03) at CMUB[Lee Erman] A question about the Weizenbaum review (with candidate correction).
C00464 ENDMK
C⊗;
∂27-FEB-76 1103 FTP:FEIGENBAUM at SUMEX-AIM RETRANSMISSION OF MSG
Date: 27 FEB 1976 1050-PST
From: FEIGENBAUM at SUMEX-AIM
Subject: RETRANSMISSION OF MSG
To: GREEN at SU-AI, MCCARTHY at SU-AI
27-FEB-76 10:47:51-PST,1133;000000000000
Date: 27 FEB 1976 1047-PST
From: FEIGENBAUM
Subject: AI QUAL
To: GREEN at SU-AI, MCCARTHY at SU-AI
cc: WINOGRAD at PARC-MAXC, FEIGENBAUM
THERE ARE ABOUT 5 PEOPLE WHO WISH TO TAKE THE AI QUAL THIS SPRING.
I HAVE ALREADY INQUIRED INFORMALLY OF EACH OF YOU WHETHER YOU WOULD BE
AGREEABLE TO A CHANGE OF FORMAT THIS
YEAR:NAMELY, THAT INSTEAD OF A WRITTEN
OR AN ORAL EXAMINATION, EACH STUDENT BE ALLOWED TO CHOOSE THE OPTION OF
WRITING A NUMBER OF (SAY 8 OR 10) ARTICLES FOR THE EMERGING "AI HANDBOOK".
THE PARTICULAR ARTICLES, AND THEIR SPREAD OVER THE MANY SUBFIELDS OF AI,
WOULD BE DESIGNATED BY US AS THE EXAMINING COMMITTEE FOR EACH STUDENT.
SUCH A SCHEME WOULD MAKE THE EXAM MORE THAN JUST AN EDUCATIONAL EXPERIEMCE,
IN THAT THE FRUITS OF A STUDENT'S
PREPARATION FOR THE EXAM WOULD BE
USEFUL TO A BROAD COMMUNITY.
BECAUSE I AM PUSHING THIS, I VOLUNTEER
TO CHAIR THE AI QUAL COMMITTEE AND
ORGANIZE THIS ACTIVITY, ALONG WITH
ANY TRADITIONAL EXAM FORMAT THAT SOME
STUDENTS MIGHT OPT FOR.
TW HAS ALREADY AGREED TO THIS PLAN.
WHAT ABOUT JMC AND CG?
HOPE TO HEAR FROM YOU SOON...ED F
-------
I am willing to agree under the condition that a traditional option
also be available, and that the "ai handbook option" have a due date
early enough so that a student's effort in that direction can be
judged inadequate in time for him to pursue the traditional opition.
-------
∂27-FEB-76 1715 WD
I left your tape recorder on your desk at home.
∂27-FEB-76 1824 CCG AI-Qual
To: Feigenbaum @ SUMEX-AIM, JMC
I accept your gracious offer to chair the exam and offer the handbook
option. McCarthy's conditions seem prudent. How will you guarantee the
quality of the student's papers,and suficient breadth in subject? It
seems experimental,and i wonder how a student passing such a qual might
do on a conventional exam. I am not objecting, but am just skeptical.
Cordell
Further thought: Eight to ten papers seems an excessive requirement;
I'd settle for less. I imagine we are sacrificing evidence of breadth,
but I go along anyway.
∂28-FEB-76 0624 REM
The file on my ISR ideas is 76228A.WRU[1,REM]
∂28-FEB-76 1132 FTP:FEIGENBAUM at SUMEX-AIM (Response to message)
Date: 28 FEB 1976 1131-PST
From: FEIGENBAUM at SUMEX-AIM
Subject: (Response to message)
To: JMC at SU-AI, feigenbaum, CCG at SU-AI, TW at SU-AI
In response to the message sent 27 FEB 1976 1834-PST from JMC @ SU-AI
FRIENDS,
I HAD ALWAYS INTENDED A TRADITIONAL OPTION..THAT OPTION LIKELY BEING A TWO HOUR
ORAL LIKE LAST YEAR. IA ACCEPT JMC'S IDEA THAT THE HANDBOOK OPTION
BE COMPLETED IN TIME FOR A STUDENT TO GO BACK TO THE TTRADITIONAL OPTION.
CONCEERNING CCG'S QUESTION, THE SELECTION OF TOPICS FOR BREADTH, AND THE
"GRADING" OF THESE MULTIPLE EXAMLETS
FOR QUALITY CONTROL IS THE
RESPONSIBILITY OF HTE AI FACULTY,I.E. US, PLUS ANY HELP WE WANT TO RECRUIT
,E.G. BRUCE,DAVE LUCKHAM,TOM BINFORD, ZOHAR, INDEED ANY OF THE FAIRLY SENIOR PEOPLE
WHOSE AREAS OF INTEREST ARE COINCIDENT WITH THE PARTICULAR HANDBOOK TOPIC THE
STUDENT HAS WRITTEN ABOUT,E.G.
TERRY WILL CEERTAINLY WANT TO "GRADE"
THOSE HAVING TO DO WITH NATURAL LANGUAGE PROCESSING.
BEST, ED
-------
∂02-MAR-76 0803 LCW PDP-10 FORTRAN
We were unable to find a good optimizing Fortran compiler for the PDP-10;
the local compiler, for example, does not even optimize over Fortran statement
boundaries. However, we have already done the comparison that you mention.
Bliss-10 and Fortran-H are both highly optimizing compilers. The input to
Fortran-H was coded in a way that masked the Fortran control structure
deficiencies (for example, we coded the second loop with GOTOs since Fortran
does not allow a negative step in a DO loop). Therefore the BLISS-10 and
FORTRAN-H outputs adequately demonstrate the code density acheivable by
optimizing compilers compiling the same source code for the PDP-10 and
360.
I know the local compiler is unoptimized. I thought it might win anyway.
∂03-MAR-76 1041 LES ARPA Final Report
To: JMC, RWW, DCL, CCG, TW, TOB, REG
Remember the Annual Report that you wrote a few months ago?
Perhaps you didn't notice, but it hasn't been published yet.
I made the mistake of waiting for Feigenbaum to finish his part.
Perhaps this is just as well, however, since we are starting a
new ARPA contract and are therefore obligated to write a "Final Report"
on the old one. I propose to use the stuff that you have already written.
I suggest that you review it and, if necessary, edit to make it complete
through the end of 1975.
If you still have your copy, use it, else poke around in [R,LES].
The nominal publication date will be January 1976. I request that
you finish no later than Wednesday, March 10, so that we can get it
printed and distributed to ARPA by 1 April, to satisfy a request of
theirs.
∂04-MAR-76 0448 REM
COMSAT@MIT-ML 02/26/76 20:41:13 Re: Msg of 8:30 PM
To: REM at MIT-ML
FAILED for JMC at SU-AI; Queued msg failed.
Recipient name apparently rejected; last reply was:
{451 Mail to JMC failed, protection violation}
(Losing message follows):
------
Date: 26 FEB 1976 2030-EST
From: REM at MIT-ML
To: JMC at SU-AI
The SU-AI system is down at the moment, which gives me an opportunity to
suggest this item for HOTER.ESS -- Not only can you netmail from your
home-port (in you case and mine, SU-AI) but if you have your own home
terminal which is not tied to your home-port and your home port is down
you can dial up another computer and netmail from it. Thus you don't lose
as much time as if you were tied to one system. You can create initial
versions of text files by mailing them to yourself. The only thing you can't
do is access already-existing files on your home port, but someday every file
will exist in two different places, and whenever one site goes down the other
place will copy the file to a third site to maintain redundancy, so that you
will be able to do work even if several computers including allof your
favorites are momentarily down. -- Many businesses are mostly paperwork,
hence can be done from the home terminal, with the advantage that you can
do your laundry, play with children, cook, eat, take catnaps, listen to
loud music, etc. while working, without bothering other workers or offending
your boss, providing you get the work done. And it cuts down on commute
traffic. You can even be "on call", ready to do 10 minutes emergency
maintenance on some project of yours then get back to sleep or play.
Much energy can be saved by (1) not commuting to work (2) not maintaining
air conditioning in two places, home and work, each of which is only used
half the time, yet costs as much to heat up or cool down or air out
for a half-day as it does to just keep things running all the time hence
today things are kept on 24 hours.
-------
∂04-MAR-76 0651 JRA disappearance
goto has gone...
i was unable to find papers in files or card in index. do you have
other likely places to look? if not, could you please gieve me his full
name and mailing address.
john
∂04-MAR-76 2111 REM via AMET FORMAT OF SPINDLE AND METHOD OF UPDATING, PROPOSAL
To: REG
CC: LES, JMC
Page 3 of CRUNCH.PLN[1,REM] details the proposed format of a spindle.
The following features will be noted: Each file is crunched or copied
into the spindle page-by-page, with its directory item pointing to each
page separately. Updating is done in such a way to minimize chance of
lossage due to system crash in the middle of an update (the only time
a crash can mung a spindle is if it happens during the single dump-mode
output that writes the new directory -- earlier crashes result in a spindle
that wastes space but otherwise is completely valid). Bubbling is limited
to a very tiny amount of data in most cases (when space is needed for
expanding the directory, only one page usually, not the entire file,
is bubbled). ** The total size of a spindle is limited
to 2↑18 words, i.e. 512 pages, 256 k. That should be capable of holding
512 k of text files that have been crunched 50%. If you believe that
larger spindles are desirable, I can change the format of the directory
to allow 2↑36 words, at the expense of increasing the size of the directory.
Let me know your opinion on this matter before I begin coding! 512 k
of text should be enough to hold one book or many small files.
∂05-MAR-76 0233 DCL Frieder von Henke
To: LES, JMC
V. Henke has written to say he would like to return to the AI Lab.
for a reasonable period of time-six to nine months. Dates are not yet
clear but I have asked him for dates. I would be very much in favor of
getting him back on the Verification project here (also LCF and other
related work) if at all possible. He contributed very well last time
for very little cost to us since he had a NATO fellowship. Please let
me know what you think about this possibility.
-David
My first priority at present is to get a full time LISP maintainer and
improver - something I should have done some years ago. I am assuming
v. Henke would cost us money this time. How is your proposal coming?
∂05-MAR-76 1202 REM via AMET MAXIMUM SIZE OF A SPINDLE
To: JMC, LES
REG's opinion is that 256k (512 pages) is too restrictive.
If you agree, I'll use larger than 18 bit addresses in the SFD (spindle
file directory), probably 36 bit addresses.
I leave it to REG.
∂05-MAR-76 1301 REM via AMET SPINDLE FORMAT PROPOSAL, EDITED
To: REG
CC: LES, JMC
If the format proposed in CRUNCH.PLN[1,REM](3) (which I've just finished
editing) is acceptable, I can begin coding it. Let me know as soon as you
have thoroughally studied it. Any questions, suggestions, bug-predictions?
∂07-MAR-76 2315 REM via AMET Keeping .MSG[2,2] files in E format
To: REG
CC: LES, JMC, ME
Using E to edit mail encourages users to get free disk space not subject
to purge merely by never deleting or transfering incoming mail.
The following users have abused this priviledge:
RHT 41.5K
DCL 41.0K
RWW 31.2K
SGK 23.0K
RWG 21.3K
ASCXXX 18.5K
TED 13.4K
EAF 11.5K
TVR 10.2K
and numerous others with fewer than 10k but more than 4k.
The total for *.MSG[2,2] is 382.6 k.
For all other users who go over their allocation, they risk having their
files transferred to a purge tape whenever an emergency purge is done.
These [2,2] hackers are exempt from that, and shouldn't be.
∂08-MAR-76 0547 REM via AMET Crunching POX.FAI
With no optimization at all, I made up a list of tokens for POX.FAI[1,REM]
and had it scanned in history mode, making the HIST.POL and HUFFS.POL files.
The c.r. was 46%, reducing POX.FAI to 19338 words. Sounds pretty ok.
∂08-MAR-76 0559 REM via AMET Addends to crunching POX.FAI
Next I used the very same tables to crunch POX.FAI[CSP,SYS] which is an
older version of the same file. The c.r. was 46%, verifying our prediction.
∂08-MAR-76 0608 REM via AMET More
Next, I used the code for POX.FAI to crunch NRAID.FAI[CSP,SYS].
C.R. = 64% (NOTE THAT USES BOTH THE TOKENS AND THE FREQUENCIES FROM POX.FAI,
WHICH IS AN EXTREME TEST SINCE I DIDN'T EVEN CONDUCT A SURVEY)
∂10-MAR-76 1448 100 : QUEENIE TALKS IN CS224
CORDELL WOULD LIKE FOR YOU TO GIVE A TALK ON ONE OF THE FOLLOWING:
(1) REPRESENTATION THEORY (2) GAME PLAYING OR (3) YOUR CHOICE
∂10-MAR-76 1451 100 : QUEENIE CARL ENGLEMAN'S VISIT FROM MITRE CORPORATION
CARL ENGLEMAN WANTS TO TALK TO YOU AND CORDELL AT THE SAME TIME DURING HIS
VISIT HERE ON THE 17TH OF MARCH. I'M TRYING TO SET UP A SCHEDULE - COULD
YOU GIVE ME A DEFINITE TIME WHEN YOU MIGHT BE AVAILABLE DURING THE 17TH -
THEN I'LL COORDINATE WITH CORDELL. THANK YOU - QUEENIE
∂10-MAR-76 1540 DCL
To: GROUP.LST[VCG,DCL]:;
***********************************************************************
NOTE : TIME CHANGE--VERIFICATION GROUP MEETINGS START AT 3.30 PM.
VERIFICATION GROUP MEETING THURSDAY 11th MARCH
TIME 3.30 pm.
Derek Oppen "Complexity of Proof Methods for Presburger Arithmetic"
This talk ought to lead into a discussion of the special purpose prover
used by the verifier and whether or not similar complexity results hold
for it. There are many untackled questions about logics of the sort we
are using in the verifier. Also we might be able to discuss whether
similar theories have anything to do with the verification of sorting
programs.
**************************************************************************
∂11-MAR-76 0644 FTP:PRATT at MIT-AI
Date: 11 MAR 1976 0942-EST
From: PRATT at MIT-AI
To: JMC at SU-AI
Here's advance notice of a recommendation I
am today sending you by mail. The body of the recommendation goes:
I have known Wendy Peikes since she took my algorithms course in Spring 1975.
She performed well, coming in the first third and getting an A. She was fairly
quiet in class, and no opportunity arose for me to judge whether she had any above-average
talents. She has a strong personality which should carry her through graduate school.
- Vaughan Pratt
-------
∂11-MAR-76 1057 FTP:ALAN COLE(A617AC01)@CMUB o: JMC@SAIL
From: ALAN COLE(A617AC01)@CMUB
Date: 11 Mar 1976 1357 EST
Subject:
To: JMC@SAIL
- - - -
From: ALAN COLE(A617AC01)@CMUB
Date: 11 Mar 1976 1353 EST
Subject: ENERGY RESEARCH PROPOSAL DINNER MEETING
To: JMC@SAIL
- - - -
JOHN,
SOMEONE FROM BILL REYNOULDS OFFICE CALLED ME ABOUT A DINNER MEETING
TO PLAN STRATEGY FOR A LARGE RESEARCH PROPOSAL TO ERDA. HOPEFULLY
SOMEONE HAS CONTACTED YOU ABOUT IT. I WILL BE WILLING TO COME TONIGHT
ONLY IF YOU WON'T BE ABLE TO MAKE THE MEETING. CALL ME IN BERKELEY
AT 642-7453 TODAY IF YOU HAVE ANY QUESTIONS OR IF YOU CAN'T ATTEND THAT
MEETING YOURSELF.
JIM HIERONYMUS
-------
-------
∂11-MAR-76 1304 FTP:BEN at MIT-AI ''Computer Power and Human Reason''
Date: 11 MAR 1976 1602-EST
From: BEN at MIT-AI
Subject: "Computer Power and Human Reason"
To: jmc at SU-AI
Marvin lent you my copy of Joe Weizenbaum's book, which
you have probably taken with you back to Stanford.
Can you send it back? Thanks.
Ben Kuipers
-------
∂11-MAR-76 1736 FTP:TK at MIT-AI
Date: 11 MAR 1976 2034-EST
From: TK at MIT-AI
To: JMC at SU-AI
SO ARE WE.... IF I FIND ANYONE WITH AN INSATIABLE DESIRE
TO LIVE IN CALIFORNIA, I WILL SPREAD THE WORD.
-------
∂12-MAR-76 2241 WTL 226
I just wanted to be sure you knew that I was taking 226 pass/fail. I'm not
sure if I ever made that point explicit.
OK, you are taking CS226 fail/pass.
∂13-MAR-76 0017 REM via AMET C.R. for .FAI files
To: JMC, REG
One more experiment -- DPYSER[S,SYS] is 47.3 k
After crunching it is 16.1 k
Not counting the deletion of directory page and nulls etc., c.r.=47%
This was done using ALL tokens above the threshhold, using history method,
with no manual optimization or automatic optimization at all, with the
Huffman-code at each node computed by CRU1.DMP automatically after reading
back in the list of reversed tokens for left-context.
Preliminary conclusion: most large .FAI files can be crunched so that
c.r. is between 45% and 50% without requiring manual optimization, and without
generating a word dictionary.
∂13-MAR-76 0126 REM via AMET More statistics on c.r. for .FAI[S,SYS]
To: REM, JMC
CC: REG
Using the code designed for DPYSER[S,SYS] mentionned in preceeding note,
not optimized, not including survey for any file other than DPYSER[S,SYS],
I have crunched several other files on [S,SYS] with these results:
FILE C.R. OLDSIZE CRUSIZE
DPYSER 46% 47.3 16.1
TTYSER 59% 40.0 18.8
UUOCON 57% 39.2 17.5
DSKSER 58% 27.0 12.6
IMPUUO 58% 15.5 6.2
SCHEDU 62% 18.2 8.5
COMCSS 58% 30.3 12.8
XGPSER 73% 19.8 12.1
CLKSER 60% 37.3 17.3
CORE 62% 21.2 11.0
DDT 70% 36.5 20.9
Each crunched file included the polish history and Huffman info.
C.R. doesn't include compression obtained by deletion of directory, nulls,
and <cr>'s before <lf>'s, thus the C.R. is pessimistic, as witnessed by the
much better CRUSIZE/OLDSIZE.
CONCLUSION -- It is feasible to crunch several .FAI files using the same
code exactly. Only DDT and XGPSER weren't very good above. They would
presumably require their own history/huffman info, or at least a code that
was developed from surveying more than just one file as above.
∂15-MAR-76 0122 REM Crunch and Spindle, a reply.
Since .FAI programs seem to crunch ok with just the history method,
without requiring a dictionary as I had previously thought, I am postponing
the careful study and research into word-dictionary methods until after I
get a working version of the new spindler up. The spindler was delayed for
a while while we worked out the design carefully, then for another while
while I fixed some bugs and added some fetures in POX for RWG, but now I
have actually started writing code (at present it can write an empty spindle,
and it can read in a spindle and recognize a valid one from a random file)
and hope to be able to spindle but not crunch within a week. After that,
history-crunch uncrunch will be included, then I'll be at a resting place
willing to explore various additional methods of crunching. Ok with you?
∂14-MAR-76 2353 JMC crunching and spindling
Do you have all you need to make the utility programs now?
[REM -- Unless I hit an unforseen snag, the spindler can be written fairly
easily with only moderate hair involved in allocating directory space.
The CRU2 program to crunch and uncrunch is up for users who want to play
with it (I think I have fixed most of the bugs in it). The CRU1 program i
is rather awkward at present, you must collect output using PTYJOB or the like
and edit it with TECO before it can be further processed. I plan to overhaul
it in the near future so random people will be able to use it. No apparant
ok
snags in any of that at the present.]
∂16-MAR-76 0641 JRA lisp book
the other day you menntioned that you were planning to read my manuscript.
was that a "threat or a promise"? Mcgraw-Hill wants to know who's reading
it. If you will indeed read it, give me an idea of how much time you
plan to spend and McGraw-Hill will supply at least a small payment.
Since I didn't hear from you about Goto, I finally wrote directly to him.
John
∂16-MAR-76 1417 DRB
To: LES, DBL, RCB, DES, CCG, BPM, JMC, TOB, PAW, RF
thought you might be interested in this:
∂16-MAR-76 1333 BLF via NMC visitors
hi david. i just had two visitors from the fbi come
ask me questions about the trip to russia.
they asked me about getting visas, and i told them
about going to the consulate, and they asked with whom,
and i gave them your name. sorry about that.
so you may have some visitors one of these days. they are fishing
for contacts with russians. i told them all of our
research was uncclassified. anyway, they seem to enjoy
fishing.
let me know when you are around here.
bil;
∂17-MAR-76 1022 REF via CMUA
Relative to your offer, sure, thanks.
If you still want me to call, let me know.
I've had a thought about a question you raised last Dec.,
and, though you've probably already considered it, I'll mention it
anyway.
The question under consideration is, that, given tthe Godel
incompleteness theorem, how can a pred. calc. based a.i. system
answer "I don't know" to some question.
It seems to me that we are imposing a stronger criterion of
'know' on the machine than we would on a human. Consider, for
example, that I have the proof of some theorem that you don't know.
Let us say, I've found an n satisfying Fermat's last theorem. If I
ask you, do you know such an n, you will quickly answer, something to
the effect of 'I don't know any'. But there is one (we're assuming),
one that within your internal axiom system, is in all senses
understandable. And therefore, by completeness, given enough time,
you could have found it. But as a human, you use 'I don't know' in a
search limit way -- I have done as much search on this problem as i
want, and don't have an answer yet.
There is another form of 'I don't know' (similiar, but I feel
there is a distinction) that humans use, as an answer to questions
like 'What is Harry's phone number.' ('Whose's Harry might be the
human response). The theorem proving machine rumbles to an 'I don't
know' in this case not because its search limit has been reached, but
rather, because it has run out of leads on phone numbers (hopefully).
It would seem perfectly reasonable to have the machine answer 'I
don't know ' in this case, too.
So much for my small thought for the morning.
Let me know if there's anything you'd like me to do.
Thanks again.
See you in May.
bob
∂17-MAR-76 1219 FTP:MINSKY at MIT-AI
Date: 17 MAR 1976 1517-EST
From: MINSKY at MIT-AI
To: jmc at SU-AI
yes, it is from the end of McCulloch's
"circuitry of ethical robots" -- in McCulloch's collected MIT press book, for instance.
McCulloch, in that article points out the problems of circular preferences -- which exist -- to utility models,
and notes that some machines would rather play than win. (I once had trouble
with one of my kids who wouldnt make the winning tic-tac-toe move because
"that would end the game-- no more fun".
-------
Thanks
∂17-MAR-76 1234 FTP:MINSKY at MIT-AI
Date: 17 MAR 1976 1532-EST
From: MINSKY at MIT-AI
To: jmc at SU-AI
reference: "Toward some circuitry of ethical robots..."(very long title)
ACTA BIOTHEORETICA XI 147-156, 1956. In fact thjis is a brave attempt
to find a dignified sense of freedom within the
psychological determinism morass.
-------
∂18-MAR-76 1357 REM via AMET
My second month is almost over -- is a third month ok (I suspect it'll be half POX half spindle)
Les is discussing it with you.
∂18-MAR-76 2113 LES
∂18-MAR-76 1811 100 : REM via AMET Planning regarding crunch/spindle
By the end of the third month I should have a working spindle program
that includes crunching/uncrunching, assuming minimal interferance from POX.
Also there may be some time to work on some of the variant crunching ideas
that JMC has proposed beyond that which is already known to work, but probably
such research would stretch into a fourth month if we go beyond simple
gathering of statistics and estimation of code lengths. This third month,
however, is the critical one in terms of getting ONE system for crunch/spindle
fully implemented (except for minor changes such as command syntax which can
be done by myself or anyone else after the guts are working). The spindle
program is being written in such a way that it will be easy to include any
particular crunching scheme in it after-the-fact (without requiring drastic
change in either file format or program design, merely adding program modules
and interfacing them in a simple way) once we have demonstrated its feasibility
and debugged its routines in a separate crunch/uncrunch-only program.
∂19-MAR-76 1202 FTP:MINSKY at MIT-AI
Date: 19 MAR 1976 1500-EST
From: MINSKY at MIT-AI
To: jmc at SU-AI
The review is for Physics Today. I like your draft a lot.
-------
∂19-MAR-76 1203 FTP:MINSKY at MIT-AI
Date: 19 MAR 1976 1501-EST
From: MINSKY at MIT-AI
To: jmc at SU-AI
Incidentally, I am replying from my 2500 console!
-------
∂19-MAR-76 1904 FTP:LEDERBERG at SUMEX-AIM book review
Date: 19 MAR 1976 1904-PST
From: LEDERBERG at SUMEX-AIM
Subject: book review
To: jmc at SU-AI
Sure, by all means copy it.
I leave soonest for a week, so must apologize for having done
that review rather hastily. I must say, if I had known how
irritated I would be by Joe's moralizing, I would not have
taken it on.
Josh
-------
Thanks.
∂20-MAR-76 0110 ME
The Z command in E now takes an argument and does an L command with
the argument, if any. I haven't actually tried this on an Imlac
myself, but it should work. Let me know if there is any problem.
It works on my Imlac.
∂21-MAR-76 2049 REM HOTER.BAY
Question -- I have heard that you don't have the authority to set up
a commercial (even if non-profit) club and grant it access to
the Arpanet. Therefore you cannot come through on your promise
regarding such access. Therefore most of the few potential users
won't have access to your club (because their only way of accessing
the west coast at a reasonable price is over the Arpanet) or won't want
to join (because they wanted the fringe benifit of access to Arpanet, and
don't consider the club worth the money without it).
Comment -- Financing is a problem. People don't have money to
pay for your club if they are out of work. One of my ideas is to
bootstrap a general information system from a mostly-employment-info
system. Also a flat rate for membership is bad. An accounting system
that charges only for services used would allow a person with fewer
resources (money) to get into the club at a level he can afford.
In general, I don't like the idea of large quantum jumps between levels
of income or expenses. I prefer graduated income and expenses, the more a
person can work the more he gets; the more money a person has the more
services he can buy; rather than cutting a person out completely because
he can't work fulltime or because he can't afford $75/month for club membership.
-- One further idea, beyond straight accounting of charges, that might
be investigated, is my idea for service level determined by $/S ratio.
Persons may voluntarily contribute money from time to time, for the
purpose of increasing their priority. Persons use services from time to
time, for the perpose of getting the services. At any moment the ratio
between the total money and total service the person has ever used
determines his priority. Anyone beyond the capacity of the system is
kicked off and advised to come back 3-6 am or whatever time is less busy.
Thus rich people buy time during busy time when it is expensive, poor
people get cheap services at "inconvient" times, and the system is
self-regulating, no ad-hoc decisions have to be made as to when prime time
is, such decisions are automatically made online in response to supply
and demand. The only hard decisions are how many Wams/Bams to count for
various types of service (connect time, cpu time, file space, ...).
You are right about the ARPA net. The club would have to be local.
I also agree with your point about linear charges. The $75 was intended
to be an average not a minimum. I expected it to attract the
hobbyists and the well-to-do - more the former than the latter.
∂21-MAR-76 2106 KRD errata in your weizenbaum review
To: JMC, KRD
(page, line)
(6,5): guarantee ==> guaranteed
(5, last paragraph): don't see why word "however" appears, seems out of place
(6,81): double quote misplaced
(6,152): word seems to be missing, "morally <what> .."
(6,157): do you mean "production system models" ?
(7,77): two "a"'s in a row
(8, 17-18): two "by"'s in a row
(8,67): words seem to be omitted
(8,76): "[it] seems apparent.." (word omitted)
(8,103): "the its .." ("the" extraneous)
I read the review very quickly (and have not even read the book yet),
so have little technical comment. But you seem to be indulging in
"hopeful thinking" in discussing the last two questions about AI near
the end. Yes, it would be very nice if, when AI finally arrives,
those who turn on the machine have the sense to ask it about its own
use. But:
if `high level AI' is intelligence comparable to man's, then
what about our own inability to determine consequences (what would
the long term impact be on the economy and society if the current
"right to a job/federal employment" bills are passed?). We can't now
predict consequences, why should human-level AI machines be able to
predict the consequences of its own use? [It gets a bit recursive,
but a more to the point example: here is a book that makes
predictions about the long term impact of AI on society and man. It
has generated much controversy because there are vastly different
views. That is, a collection of those who know AI very well can't
agree on a prediction of its impact. Why should a machine "only" as
smart as they be able to answer that same question? Yet a machine
"only" as smart as they are would have profound consequences, yes?]
and even if it could, on what do you base the belief that those
who turn it on will have the sense to ask? On what do you base the faith
that they who ask will not ignore the answer?
Such an important question needs a more substantive answer, and
perhaps not one that can easily be written in a short review. The one
that's there at the moment just seems a bit too facile.
Thanks for the corrections and the comments. You are right that
there is too to be discussed for a review. Your guess that
Weizenbaum "makes predictions" is mistaken. He says less about
it than I did on that one page. Let me solicit an article for
McCarthy's electronic magazine on the subject.
∂24-MAR-76 2046 JMC
How is crunchingnd spindling?
[REM -- Right now the system is too slow to work, which is typical of
al except 4am-7am nowadays, making my work go rather slowly. Also, I
haven't gotten word from you on to what extent I am authorized to spend
time making POX capable of doing complicated math stuff like RWG is
doing, and to what extent I can resuse to do such work on grounds
that it interferes with crunch/spindl too much. -- To date, after that
lengthy design of the new spindler, I have written some of the code.
It is capable of making a empty spindle, verifying that an
existing spindle is valid, and almost of putting a new file in a spindle.
No major problems so far, except lack of compute time etc.]
Put at least 2/3 of the time into crunch and spindle.
The KL-10 will be here in a few days, and I suppose the
computer time situation will get worse for a month or
more. It has been a bit better in the last few days,
because of inter-quarter break.
∂25-MAR-76 0901 JMC*
Teller
∂25-MAR-76 0136 JMC
Put at least 2/3 of the time into crunch and spindle.
The KL-10 will be here in a few days, and I suppose the
computer time situation will get worse for a month or
more. It has been a bit better in the last few days,
because of inter-quarter break.
[REM -- OK, will do.]
∂26-MAR-76 0722 REG
You have exceeded your disk quota.
The files listed below have been purged to reduce your disk
area to your quota of 1600
Before purging, your files occupied 1804
BACKUP.TMP[258,JMC]
BACKUP.TMP[F75,JMC]
BACKUP.TMP[LET,JMC]
BACKUP.TMP[W76,JMC]
RELATI.TEM[CUR,JMC]
QQPUB.RPG[LET,JMC]
QQPUB.RPG[W76,JMC]
SIGART.LST[LET,JMC]
LEADER.LST[LET,JMC]
ENERGY.LST[LET,JMC]
GTREE.LST[206,JMC]
NEWELL.LST[LET,JMC]
WISEMA.DMP[F75,JMC]
CRYPT.DMP[ 2,JMC]
BLOCKS.DMP[W76,JMC]
CODE.DMP[ 2,JMC]
SOURC2.LAP[ 1,JMC]
REVA.LAP[206,JMC]
INSANE.LAP[206,JMC]
SEARCH.LAP[206,JMC]
INSAN2.LAP[206,JMC]
INSAN3.LAP[206,JMC]
TICTA2.LAP[206,JMC]
TICTA3.LAP[206,JMC]
TIC3D.LAP[206,JMC]
GTREE.LAP[206,JMC]
GAME.LAP[206,JMC]
TICTAC.LAP[206,JMC]
BASIC.LAP[206,JMC]
REPRES.PRO[ 1,JMC]
FUNS[ 1,JMC]
EPIS[ 1,JMC]
P1[ 1,JMC]
PART[ 1,JMC]
PERM[ 1,JMC]
PATH2[ 1,JMC]
PATH[ 1,JMC]
ANTI[ 1,JMC]
ANTIN[ 1,JMC]
SYLL[ 1,JMC]
MEET[ 1,JMC]
SEMAN[ 1,JMC]
NONDUP[ 1,JMC]
COMPIL[ 1,JMC]
TIMES[ 1,JMC]
TESTA.SAI[ 1,JMC]
TESTC.SAI[ 1,JMC]
TESTB.SAI[ 1,JMC]
TESTD.SAI[ 1,JMC]
DADDA[ 1,JMC]
ORDIN[ 1,JMC]
PROB1[ 1,JMC]
ROTAT.SAI[ 1,JMC]
ROTA.SAI[ 1,JMC]
ROTB.SAI[ 1,JMC]
ROTC.SAI[ 1,JMC]
SEMAA[ 1,JMC]
SEMAB[ 1,JMC]
INTER[ 1,JMC]
FLOYD.SAI[ 1,JMC]
CONVER.SAI[ 1,JMC]
RELREP[ 1,JMC]
MC[245,JMC]
FWGC[245,JMC]
PATH[245,JMC]
PATH2[245,JMC]
CONTRO[ 1,JMC]
DEMO[ 1,JMC]
LISPAD[ 1,JMC]
TESTE.SAI[ 1,JMC]
ICOM[ 1,JMC]
TRANS1[ 1,JMC]
PLAY[ 1,JMC]
SIMUL[ 1,JMC]
COMPU2[ 1,JMC]
SOURC2[ 1,JMC]
SOURCE[ 1,JMC]
PUZZ.SAI[ 1,JMC]
TCLUBA[ 1,JMC]
TCLUB[ 1,JMC]
OPEN[ 1,JMC]
TOREAD[ 1,JMC]
NEWDOC[ 1,JMC]
BOARDS.SAI[ 1,JMC]
NEEDS.CSD[ 1,JMC]
FAC.PRO[ 1,JMC]
LCOMT.LSP[206,JMC]
LIB.RLS[206,JMC]
REFER.ENC[225,JMC]
ECHO.FAI[206,JMC]
OUTLIN[206,JMC]
MCCRAC.LET[ 1,JMC]
PATHS.RLS[225,JMC]
SYMFUN.RLS[206,JMC]
PAUSD.CMN[ 1,JMC]
PARKER[ 1,JMC]
GRPDAT.RLS[225,JMC]
GRPALG.RLS[225,JMC]
GROPER.RLS[225,JMC]
ROOSTE[ 1,JMC]
DRUMME[ 1,JMC]
DIMEN.RLS[225,JMC]
CHAR.RLS[225,JMC]
PERMU2.RLS[206,JMC]
LARRY2[ 1,JMC]
COMPU2.LET[ 1,JMC]
COMPU.LET[ 1,JMC]
DEG5.IN[225,JMC]
REPRES.RLS[225,JMC]
S4.REP[225,JMC]
DEG6.IN[225,JMC]
SLOMAN.REF[ 1,JMC]
R42.IN[225,JMC]
PROP.ECO[ESS,JMC]
CKSUM.DAT[ 1,JMC]
PARA.AI[ESS,JMC]
PUZZ.SAI[225,JMC]
PUZZA.SAI[225,JMC]
PANIC.SOS[225,JMC]
ANNOUN[225,JMC]
CHILD.REV[ESS,JMC]
PUZZB.SAI[225,JMC]
ENCYC1.PRO[ESS,JMC]
RACRUL.MEM[ESS,JMC]
PUZZE.SAI[225,JMC]
TURIN1.LEC[ESS,JMC]
DRIVE.DIR[ AI,JMC]
MINE.DIR[ AI,JMC]
PTCH2.DIR[ AI,JMC]
SORT.LSP[206,JMC]
DIFF.LSP[206,JMC]
KNOW2.AI[ESS,JMC]
ARCH.PRO[ESS,JMC]
ARTNA1.ART[ESS,JMC]
STUD[206,JMC]
LOGREQ.AI[ESS,JMC]
RECOG.LET[ 1,JMC]
TSS.PRO[ESS,JMC]
HEADIN[206,JMC]
III[ 1,JMC]
MEMMTC.QUA[ESS,JMC]
LANG1.AI[ESS,JMC]
S5.QUA[ESS,JMC]
MTC71.QUA[ESS,JMC]
SERVIC.PRO[ESS,JMC]
R1301.ART[ESS,JMC]
R1303.ART[ESS,JMC]
KNOW3.AI[ESS,JMC]
N30[ 1,JMC]
Now, why don't you really delete some files?
∂26-MAR-76 0727 REG
If you want Nick Littlestone's files preserved over the NEXT purge ...
∂27-MAR-76 0256 FTP:MASINTER at PARC-MAXC file-understanding
Date: 27 MAR 1976 0256-PST
From: MASINTER at PARC-MAXC
Subject: file-understanding
To: JMC at SAIL
My apologies for my slow response.
Here, for what it is worth, are some ideas related to our discussion:
The problem is to design a language of primative concepts of file
formats, such that it is easy and intuitive to write descriptions
of file formats in, say, predicate calculus.
Design/write a program which, given a request for information
and a set of files with descriptions of their formats, will
answer that request.
It seems, offhand, that such a program would not be interesting
or useful or intelligent unless it could infer file formats; and that
inferring file formats is, in general, fairly difficult.
For example, I personally am capable, generally, of inferring the
format of a telephone directory file from the first couple of entries;
but that capability is partially based on my knowledge that a given
name is probably a first name or a last name or maybe a street name.
It is unclear how much world-knowledge would be required for
more general tasks than telephone-directory lookup.
===========
personally, I am pretty loaded down with projects for the next
couple of months & am not anxious to jump into programming/design.
I would be more than glad to continue discussing this via sndmsg.
Yes?
Larry
-------
∂28-MAR-76 1513 FTP:MARG at MIT-AI
Date: 28 MAR 1976 1812-EST
From: MARG at MIT-AI
Sent-by: MARG0 at MIT-AI
To: JMC at SU-AI
WHAT IS FILENAME OF WEIZEN REVIEW?
-------
Weizen review is weizen.rev[pub,jmc]. This is a PUB source file
and is still being "improved". Copies of the "final" version
will be mailed to Minsky, but I don't know when this will be.
However, the weizen.rev copy is public.
∂28-MAR-76 2050 REM via AMET Progress update on Spindle project
CRU3.FAI seems now to be able to spindle and unspindle files (no crunch)
successfully, although there are a few changes I want to make before
letting other people use it or adding the crunch/uncrunch code to it.
CRUNCH.PLN is up-to-date now insofar as the status reported on the last page.
OK, got it.
∂29-MAR-76 2006 FTP:RMS at MIT-AI
Date: 29 MAR 1976 2218-EST
From: RMS at MIT-AI
To: JMC at SU-AI
I can't seem to recall Forest Howard at all (although
it's likely that I once knew him if he was at HARV-10).
My memory for people isn't first-class. Sorry.
-------
OK, thanks anyway.
∂29-MAR-76 2020 FTP:Jrobinson at SRI-AI Dinner Wednesday
Date: 29 MAR 1976 2021-PST
From: Jrobinson at SRI-AI
Subject: Dinner Wednesday
To: jmc at SU-AI
cc: friedman
Joyce and I delighted to accept.
Jane
See you then.
-------
∂29-MAR-76 2044 DSB
Scientific purpose is improvement of logical thinking capabilities,
and exercising creativity. System load is not great. I was waiting
for other things to finish while playing game so I will stop.
∂30-MAR-76 0140 LES Forrest Howard
I take it that you haven't discussed salary with Howard. And I suppose
that we should offer to pay his moving expenses to get here.
I have not discussed salary. He is unmarried, and I am agreeable
to any algorithm you may devise concerning salary and moving expenses,
but he has already had some experience working for the commercial
world and may not be agreeable to a very large cut. You could ask
him or Cheatham what he is getting now.
∂30-MAR-76 0856 FTP:BEN at MIT-AI
Date: 30 MAR 1976 1119-EST
From: BEN at MIT-AI
To: jmc at SU-AI
John,
You would enjoy reading "In Praise of Technology",
and article by Samuel Florman which appeared in Harper's Magazine
in November, 1975. He also has a recent book entitled
"The Existential Pleasures of Engineering," which covers much
the same ideas at greater length. He is defending technology
against the attacks of Mumford, Ellul, et al, quite successfully,
I think.
Ben Kuipers
-------
Thanks.
∂30-MAR-76 1959 FTP:LEDERBERG at SUMEX-AIM WEIZENBAUM book, p.s. [jmc,eaf
Date: 30 MAR 1976 1815-PST
From: LEDERBERG at SUMEX-AIM
Subject: WEIZENBAUM book, p.s. [jmc,eaf
To: jmc at SU-AI
cc: FEIGENBAUM
I wish I had articulated the following point clearly enough before
to have included it in my review. It was lurking just below conscious-
ness as what was really strange about the book.
Joe says don't LOVE the machine. {E.G. don't animalize or personify it}
But it is inconsistent then to HATE it.
Josh
-------
∂30-MAR-76 2003 DSB
I apologize for the tone of my last note which, after some
thought, I conclude may have sounded rather rude. That was not
intentional. Just to show you there're no hard feelings (or something
like that), I challenge you to a game of chess. I am only a Class B
player so you will probably whip me but the game should be
interesting anyway.
Well, I don't think I'm even class B, but I accept.
∂30-MAR-76 2022 DSB chess
Shall we set a time when you're here so we can use DCHESS for
the board or what ? (Have you got a chess board ?)
I'm old fashioned and prefer to play across the boad.
∂30-MAR-76 2044 DSB chess
I don't mind playing over a board; I don't have one though, so what
do you suggest ?
I have one in the Lab, so catch me sometime there.
∂30-MAR-76 2156 WTL 224 readings
In your box you should find a copy of the revised reading list for 224. One of
the changes is the inclusion of the paper by yourself and Hayes on philosophical
problems from the standpoint of A.I. The students will have read this paper by
the time you give your lecture in that class.
∂31-MAR-76 0935 FTP:MINSKY at MIT-AI
Date: 31 MAR 1976 1234-EST
From: MINSKY at MIT-AI
To: jmc at SU-AI
i like weizen rev. have told physics today about it and am sending copy of
my letter and theirs to you.
-------
I look forward.
∂31-MAR-76 1651 DCL EXAMPLES FOR VERIFICATION
To: GROUP.LST[VCG,DCL]:;
EXMPLS.VER[VCG,DCL] contains a list of problems for people
to work on. Anyone may add to this list another peoblem.
There are two categories of difficulty.
∂31-MAR-76 1953 FTP:CERF at USC-ISI Fancy CRT's
Date: 31 MAR 1976 1952-PST
From: CERF at USC-ISI
Subject: Fancy CRT's
To: jmc at SU-AI, pbaran, anderson at RAND-ISD
cc: cerf
I understand that Dr. ART Collomeyer of CALMA COrp, 408-245-7522
may come out with a fancy 1 megabit CRT using CCD memory
in 16K bit chips (64 of them). The refresh rate is about 40/sec
and the CCD memory is organized as a 3 MB 16-bit parallel shift
register. I am not certainof the price and the announcement date
is probably late in the year. This is basically rumor from a
student I know working on the project, so Collomeyer may be
somewhat defensive about responding to your inquiries.
Vint
-------
∂02-APR-76 1533 DCO Verification Seminar
To: GROUP.LST[VCG,DCL]:;
"User-Defined Data Types as an Aid to Verifying Lisp Programs"
by Corky Cartwright
The verification seminar will be on Tuesday at 330 p.m., not
Thursday. Corky will discuss his thesis work on proving non-trivial
properties of non-trivial LISP programs.
∂03-APR-76 1016 DEK
i found the long letter i typed to you re janet roberts sitting on
my terminal. sure enough, the message was followed by ↑C instead of ↑Z,
so it had died without my awareness. a new kind of dead letter.
thanks for acting as you did when you finally did hear from me.
Please let me know of any reservations you might have about appointing
the yaos as assistant professors. we have to get the appointment papers
to the provost soon or we lose the slots; ill be happy
give you a resume of what i said about the matter at the end of the
faculty meeting (after you had left), and about the conference
floyd feigenbaum and i had with the dean yesterday. basically we were
reassured that this doesnt close our options for the future, and that
franceses appointment doesnt fill a slot its an additional gift (affirmative
action) that is actually a gift of another part of stanford to H&S,
so it doesnt count in the statistics. this is unclear, but id rather
talk to you in person about it if you really have some reservations
about this appointment. i have no doubt that andy is the brightest
young computer scientist to have gotten a phd during the last several
years, and he continues to develop at a fantastic pace.
sincerely, ↑Z, don.
Well, it seems like too much of one thing. Andy struck me as quite
narrowly focussed, Frances less so. At least she could see where I
wasn't following Andy and put in the missing words. I hold it
somewhat against Andy that he didn't know about the missing
neutrinos, which suggests that when he left physics, he didn't retain
enough interest even to read Physics Today or even Scientific
American. Well maybe this leads to the best and most concentrated
work on analysis of algorithms, but it certainly doesn't lead to good
judgment about what is important in computer science. If Frances is
genuinely free and doesn't compete with our possibility of hiring,
say Pratt, or Manna, thenI will go along. However, if the Yao's
remain as one-sided as they are, you will find me balking at the
tenure step no matter how well they do in the narrow field. I guess
I am also doubtful about whether there will be enough students with
such a mathematical concentration and talent to keep them busy -
unless they can get students of pure mathematics also.
By the way, it's your move on Jack Schwartz; what are you
doing there. It isn't fair to just leave the matter in suspension.
∂03-APR-76 2133 HPM quack1
On reading it again, it seems pretty likely that the part which annoyed
you most was the last paragraph or two of the introduction. I can see
your point. It it was intended to focus attention, not as a personal
attack, but the same words seem to have a stronger connotation
written than spoken. Taken too seriously, they are unfair and
uncalled for. I plead lack of skill and apologize. Any suggestions
on how to tone them down, or eliminate them will be accepted.
∂04-APR-76 1317 FTP:MINSKY at MIT-AI
Date: 4 APR 1976 1616-EST
From: MINSKY at MIT-AI
To: JMC at SU-AI
HEWITT SUGGESTS USING SOME OTHER PROPERTY THAN "NEW LEFTIST"
THAT IS MORE DESCRIPTIVE AND LESS "PARTISAN"
-------
I guess I'll have to think of something, because Hewitt's is the
third comment to that effect.
∂05-APR-76 0652 FTP:MINSKY at MIT-AI
Date: 5 APR 1976 0950-EST
From: MINSKY at MIT-AI
To: jmc at SU-AI
crypto-neo-calvinists
-------
∂05-APR-76 0859 FTP:REM at MIT-MC
Date: 5 APR 1976 1154-EST
From: REM at MIT-MC
To: JMC at SU-AI
Well, the 2am-9am gap that had been shortened to 3:30am-8am recently
has now been completely filled. Tonight the system was busy steadily
all morning. At 3 am the busy factor was 4 (i.e. it took 4 minutes of
realtime to do a DI[1,3]/Q/G which usually takes 1 minute if the system
isn't loaded), by 5:45 am it had increased to 5, and as the
daytime crowd began showing up it had increased to 10.
Any estimate of when if eveer the Librascope will be back up?
-------
∂05-APR-76 1039 RWW FOL
NEW FOL UP WITH REALLOCATION WORKING
∂05-APR-76 1917 WTL IEEE Proceedings
Do you recieve copies of the IEEE proceedings? If so, could we use your April
issue when it arrives for 224 for a few days? We would like to make copies of
Reddy's paper for everyone in the class, but it seems that the university libraries
are not recieving their copies.
Yes, if it comes in paper, but I have just tried to switch
my IEEE stuff to microfiche.
∂05-APR-76 1929 WTL CS224 Lecture
This is to inform you of a room change for CS224. It now meets in room 300 of
the outer quad (near the Engineering corner.) I would also like to remind you
that you are scheduled to speak on "Formal Problem Representation" next Tuesday,
April 13 from 1:15-2:30 pm. Let me know if you need any audio-visual aids
(overhead projector, slide projector, movie projector, etc.). Thanks.
∂06-APR-76 1059 FTP:STEFIK at SUMEX-AIM Lighthill Film
Date: 6 APR 1976 1058-PST
From: STEFIK at SUMEX-AIM
Subject: Lighthill Film
To: McCarthy at SU-AI
Prof. McCarthy,
I am interested in finding a copy of the film of the
"Lighthill Debate" to show in Prof. Feigenbaums's course
"Computers: Their Nature, Use and Impact" this quarter. Do
you have a copy of this film or know how we could get access
to one ?
Thank you,
Mark Stefik (STEFIK@SUMEX)
-------
Disregard previous message. Nilsson at SRI has it.
∂07-APR-76 0059 REM via AMET Spindle status
CRU3.FAI is now capable of all the usual operations having to do
with spindle-without-crunch. See CRUNCH.PLN (last page) for details.
Next to incorporate the crunching stuff...
∂07-APR-76 1020 REG
To: TAG, TED, JMC
There remains some evidence that the IMLAC in Polya really isn't feeling
well. It has been observed many times that the Imlac is being cooked
sitting there by the window. Perhaps we should remove it until such time
as a reflective window coating is provided.
∂07-APR-76 1416 KRD via ML WEIZENBAUM
I INADVERTENTLY TOOK IT WITH ME WHEN I CAME EAST FOR A VISIT. IF
YOU NEED IT RIGHT AWAY, I'LL BE HAPPY TYO MAIL IT BACK TO YOU (1ST
CLASS), BUT I'LL BE BACK IN PALO ALTO BY A WEEK FROM TODAY.;;
I SUSPOECT THAT MIGHT BE FAST THAN EWVEN 1ST CLASS MAIL. LET ME KNOW IF
YOU WANTED IT MAILED; SORRY IF THISD INCONVENIENCES YOU. I SHOULD HAVE
REMEMBERED TO ASK BEFORE LEAVINGT TOWN. (TERMINAL ACTING UP, OR
THE TYPING WOULDN'T BE QUITE SO BAQD).
I guess it won't hurt me to wait a week.
∂08-APR-76 0727 100 : Weizenbaum via MITT Mental←qualities
John:
Your material arrived. Many thnaks.
with respect to your review:
Have to think about it. My present feeling is that we are simply
talking past one another. Hope to havbe more to say later (maybe).
With respect to your paper:
To the extent I understand it, I find nothing to grumble aboiut.
I agree, certainly, that the use of mentalistic language and imagery
is essential to understanding what a computer is doing and for
dealing with it in most practical situations. I also think thjat the
inferences you draw from that are correct and reasonable.
Before you go much further in this reseacrch, you should aquaint
yourself with the work of the philosopher Daniel C. Dennett. (tufts)
May I show him your paper? I thnik he would be intrigued and quite
texxx sympathetic.
The mailbox situation on multics seems to be screwed up somehow. I
will try to get a file directory on ITS for communication purposes.
Meanwhile, should you want to send me messages, perhaps you skhould
rout them via one of your ITS frie nds, sayWinston or Sussman.
(sorry for the bad typing - I will have to discover what the various
controld characters, e.g. delete, are on thisd thing.)
Thanks again.
Joe.
∂08-APR-76 1245 REG "VERIFY"
I am concerned by the presence of multiple users running VERIFY, which
I believe is Suzuki's verifier. These jobs are typically larger than
140 pages and compute bound. The presence of one such job on the system
for the past thirteen hours has substantially degraded system throughput
for other users.
I think that there's really no place in a timesharing system for jobs that
are that large and that cpu intensive. VERIFY is frequently run under the
"Batch" program which suggests that it's not interactive and should therefore
do its computing on a non-interactive system.
Our system is the only one we have, and VERIFY is an important project.
We can restrict the amount of time it uses if we need to. I will check
on how much time it used in March.
∂08-APR-76 1542 LES
To: JMC, REG, ME
∂08-APR-76 1057 BH via MITT DECUS REPORT
(SORRY ABOUT GETTING THAT WRONG THE FIRST TIME) THERE WAS A
SESSION ON EDITORS THIS MORNING (ENTITLED "EVERYONE
NEEDS A GOOD SOS") AT WHICH CLARK WILCOX AND I TOOK
TURNS HARANGUING THE CROWD ABOUT DISPLAYS. TURNS OUT SOMETHING
LIKE 90% OF THE PEOPLE AT THE MEETING WERE FROM INSTALLATIONS
WITH DISPLAY TERMINALS BUT WERE USING SOS ANYWAY FOR LACK
OF AVAILABLE SOFTWARE AND/OR FOR LACK OF KNOWING ANY BETTER.
CLARK WAS SUFFICIENTLY GROSSED OUT BY THIS THAT HE IS GOING TO TRY
TO GET UP A VANILLA-TOPS-10 VERSION OF THE TENEX TVEDIT. MAYBE
WE WILL TAKE OVER THE WORLD AFTER ALL. THERE IS A BIG EDITOR
MEETING IN MAYNARD TOMORROW WHICH I HAVEN'T DECIDED WHETHER TO
TRY TO ATTEND OR NOT. THE DEC PERSON IN CHARGE OF
THEIR FORTHCOMING "EDITOR STANDARD" HAS DONE AN EXHAUSTIVE
SURVEY OF HIS NAVEL AND DECIDED THAT DPYS ARE BEST TREATED AS HARDCOPY TTYS!!!
STAY TUNED...
∂09-APR-76 0903 FTP:BUCHANAN at SUMEX-AIM Robert Blum
Date: 9 APR 1976 0903-PST
From: BUCHANAN at SUMEX-AIM
Subject: Robert Blum
To: mccarthy at SU-AI
What was the Adm. Committee's decision on Bob Blum (the medical resident at
Kaiser Hospital)? Is there any more information I could supply that would
still be helpful?
Ed mentioned another medical person whose folder was set aside for
later consideration -- Jim Brinkely. What decision was made about him?
Thanks,
Bruce
-------
∂09-APR-76 1125 REG
You are exceeding your disk quota.
Files that occupy space beyond your quota are subject to purging!
If you don't delete some of your files, the purger will.
Your disk quota is: 1600
Your files occupy 1835
∂12-APR-76 0011 DWP Midnight Maurading
John - I have your book `A History of π' . If you want it soon
leave a message for me at 327-2207. D. Poole
∂12-APR-76 1008 REM via AMET Progress report on spindle
The code to crunch-by-pages-and-spindle, including copying trees
into the spindle if not there already, seems to mostly work (we won't know
for sure until the uncrunch code is put into CRU3.FAI and we try to
uncrunch-unspindle the file that was crunch-spindled).
OK, I am looking forward to it.
∂12-APR-76 1735 PLW side effects paper
CC: JMC
i just read your comments on my paper. i agree with all of your
conclusions: that it raises interesting questions, that it is worth
a research project, and that unfortuneately the paper comes to no
solid conclusions. with respect to the research project, i am
doing a research project this quarter aimed at developing my
version of the ultimate programming language (supervised by prof Knuth)
where i will be tackling the side effect question from the point of
view of developing "good" constructs for expressing programs with
side effects. the question of a mathematical characterization of
which side effects can be made virtually invisible is also of interest
to me. if i can find the time, i would like to investigate that
subject. depending on how my workload shapes up later this quarter,
i may ask you to supervise an independent study project on that
subject, if you would be willing.
i just wrote my name and address on paper for you in your office.
in case it is more convenient to have it on-line i will repeat it:
phil wadler
250 curtner ave #28
palo alto, ca 94306
phone: 494-0529
i hope to hear from you soon on the possibility of working at the
lab this summer (i've been dieing to work at the lab since before i
came to stanford)
-- phil wadler
∂12-APR-76 1945 DCO Verification Seminar
To: GROUP.LST[VCG,DCL]:;
" Verifying Quicksort " Thursday 4 p.m.
Scot Drysdale will talk this Thursday on verifying various
versions of Quicksort.
∂13-APR-76 0641 REM via AMET Progress rpt.
The code to uncrunch-by-pages-and-unspindle is implemented and
seems to be working. The uncrunched-unspindled output is SRCCOM
identical to the original input file (5 pages, POX.PLN[1,REM] if you're
curious). Unfortunately earlier tonight I managed to introduce a bug
in the crunch-by-pages-and-spindle code that was previously working,
so to do that you must use a backup of the program, whereas to do the former
you must use the latest version. Also there are a few details of format
I want to clean up, so the system won't be ready to play with for another
day appx.
∂13-APR-76 0821 REG
∂12-APR-76 2300 100 : ERIC SEALE via ALOT GUGST ACCOUNT
I AM VERY DISAPPOINTED THAT THE GUEST ACCOUNT HAS BEEN TERMINATED.
THOUGH I DO UNDERSTAND THAT IS OFTEN MISUSED, I FEEL THAT IT COULD
BE VERY HELPF]L IN LEARNING ABOUT THE STANFORD SYSTEM. I AM NOW A STUDENT
AT MOANALUA HIGH SCHOOL IN HONOLULU HAWAII AND CURRENTLY USE THE BCC-508
(EXCUSE ME, BCC-500) TO LEARN. I FEEL VHAT HAVING EXPERIENCE ON MORE THAN ONE
COMPUTER WOULD HELP ME LEaRN EVEN MORE. MAHALO--THANK YOU.
∂13-APR-76 1350 LES
∂13-APR-76 1256 FTP:HOWARD at HARV-10
Date: 13 Apr 1976 1555-Edt
From: HOWARD at HARV-10
To: LES at SU-AI
LES:
THANK YOU FOR YOUR INTEREST. I HAVE DECIDED, HOWEVER,
TO TAKE MY SAVINGS AND GO TO EUROPE, WHERE I PLAN TO REMAIN UNTIL
I AM OUT OF FUNDS. WORK AT STANFORD IS A VERY INTERESTING PROPOSITION,
HOWEVER, AFTER EIGHT YEARS OF HACKING WITHOUT ANY REAL RELIEF, I FEEL
IT IS TIME TO GET AWAY FROM CODING FROM A WHILE.
I SUSPECT THAT YOU SHOULD HAVE NO PROBLEM IN GETTING SOMEONE WHO
WILL BE GOOD FOR YOU, AS WELL AS THE TYPE OF PERSON THAT WOULD
BENIFIT FROM THE ATMOSPHERE. I REALIZE THAT AT THIS MOMENT, THE
APPLICATION OF THESE TWO CRITERIA TO MY RESTLESSNESS WOULD NOT
REALLY FIT.
I EXPECT TO BE BACK SOMETIME IN ABOUT A YEAR -- IF YOU
HAVE A OPENING AT THAT TIME, I HOPE I WILL BE PERMITTED TO THEN FOREWARD
MY VITA.
THANKS AND BEST WISHES
FORREST HOWARD
∂13-APR-76 1442 RAK Thursday talk
To: GROUP.LST[VCG,DCL]:;
If it's OK with everyone, how about we start at 1 pm? This has the nice
side effect that it doesn't conflict with Richard Weyrauch's talk on
Thursday.
Comments→DCO or me.
Dick
∂13-APR-76 1446 REM via AMET CRU3.FAI fixed!
OK, I have fixed the bug, which was due to confusion over two
different UNREAD routines. Using POX.PLN (5 pages, 3.1 K total) as test
data, I did a Spindle followed by an Unspindle, and a Crunch-by-pages-and-spindle
followed by an Uncrunch-by-pages-and-unspindle. In both cases the final
output was BINCOM identical to the original input!! Thus the whole
program seems to be essentially working and is available for you to
play with FOR EXPERIMENTAL PURPOSES ONLY (don't delete the originals of
files you spindle, and when unspindling don't overwrite the original,
use an alternate file name, when it says:
FILE ON DSK: ALREADY EXISTS, OVERWRITE?
>
you type A<cr> and it will ask you for an alternate file name).
I will put up this current version as CRU3.DMP[1,REM] now.
Note that * is the prompt for a top-level command, and > is the prompt
for particular arguments that it has just asked for.
∂13-APR-76 1553 100 : REM via AMET
I have just finished mailing you a 20 line message detailing
lots of useful info about how to use CRU3 and various limitations to watch for.
Unfortunately the pty under the telnet server got wedged by a bug in MAIL or something,
and the message was lost, and I cannot recreate it. shit shit.
∂13-APR-76 1601 FTP:REM at MIT-MC More info about CRU3
Date: 13 APR 1976 1902-EST
From: REM at MIT-MC
To: JMC at SU-AI
ATTEMPT TO RECREATE MOST OF THE MESSAGE THAT WAS LOST A FEW MINUTES AGO.
SUBJECT: More info about CRU3
MSG:
(1) The format of the crunched-by-pages files in a spindle is likely to be changed.
The format of a plaintext-spindled file is unlikely to change. --no echo --
-------
∂13-APR-76 1604 FTP:REM at MIT-MC
Date: 13 APR 1976 1904-EST
From: REM at MIT-MC
To: JMC at SU-AI
ASSUMING THAT PART GOT THROUGH...
(2) Ignore the c.r. message, the numbers are wrong.
(3,4,5) I don't feel like retyping all that stuff, if you have
any questions, ask me.
-------
∂13-APR-76 1658 JMC
I suppose the next thing is a draft section for the Monitor Manual.
[REM -- Well, as for a general explanation of the ideas involved (data
compression, user-level re-allocating the disk), I think there is
a reasonably-good set of definitions somewhere in CRUNCH.PLN, probably
on page 2. As for command syntax and usage, the commands aren't yet
in their final form so I believe that is premature. I think at least
two persons besides myself should use a program before declaring it
suitable for documentatin in the monitor manual.]
∂14-APR-76 0105 REM via AMET Bug I found in CRU3
If you cause history or Huffman tree to be loaded, any attempt to
later spindle or crunch-and-spindle in the same core image will crash
with FATAL ERROR IN FORMAT OF SPINDLE DIRECTORY due to a couple pointers
nt being zeroed so that they point into random places during the second operation.
I've fixed it (I think) in the source, will wait until next
recompilation to put up a corrected version.
I noticed it; send me a message when it's fixed. Also, when one
wants to crunch-and-spindle several files with the same trees,
it should be possible to do this in a straightforward way.
Also including the trees in the crunched file should be optional.
Otherwise the overhead may be too great in some cases. Naturally,
only the most standard files will be guaranteed to be preserved
for uncrunching.
unless the user saves them himself.
∂14-APR-76 0108 REM via AMET Proposed preliminary default for trees
If you have crunch-and-spindled at least one file, then default is
@<sub-file> for the last crunched-sub-file in the directory which has
not been eleted. Otherwise default is NOTSEV.HIS and SEVEN1.HUF on
[UP,REM]. Of course the "last c.s.f. in dir not deleted" default won't
have a reasonable effect right after you have used the Alphabetize-the-directory
command, but that's a minor inconvenience. Feedback on my improved default idea?
∂14-APR-76 0115 JMC
I noticed it; send me a message when it's fixed. Also, when one
wants to crunch-and-spindle several files with the same trees,
it should be possible to do this in a straightforward way.
Also including the trees in the crunched file should be optional.
Otherwise the overhead may be too great in some cases. Naturally,
only the most standard files will be guaranteed to be preserved
for uncrunching.
unless the user saves them himself.
[REM -- I have patched the .DMP file to set those values to zero,
and it can now crunch four in a row without apparant trouble. I'll
rename the patched version to C3.DMP[1,REM] for the moment. -- I am
already planning to allow a whole list of files to be crunched in one
pass, typing them on one line, and not reloading the trees between files.
-- Since having a tree on a separate file takes up an extra disk block
(actually two since there are two trees) whereas keeping them inside the
spindle adds zero blocks most of the time as well as guaranteeing the
ability to uncrunch, I am firmly of the opinion that having one copy
of each different tree in the spindle is a win.]
∂14-APR-76 0038 JMC
crunch-and-spindle
I think I succeeded in using the program to produce foo.spi[s76,jmc]. The
computation of crunch ratio in crunch-and-spindle seems to compare the
total spindle file with the file just added to the spindle, so that
crunch ratios larger than 1 are obtained. I suggest that we offer the
program for English text first, and that a matched pair of history and
huffman files good for this purpose be offered as a default option signalled
by <cr>.
[REM -- Yes, in the middle of your runs I ran cru3 myself, using your
spindle file (readonly, I just did a DIrectory command then an Exit
command) and saw your four crunched files... -- Yes, it does seem to add up
everything you've spindled to date rather than starting at zero for
each file. After I get a new listing of the program I'll fix that bug and
decide what the official format will be for crunched-spindled files.
Also several other minor bug-fixes etc. at the same time. Note that
the current version is still EXPERIMENTAL ONLY so don't expect foreward
compatibility to be guaranteed.]
∂14-APR-76 1501 RLD TIME CHANGE FOR THURSDAY TALK
To: GROUP.LST[VCG,DCL]:;
SCOT DRYSDALE'S TALK ON QUICKSORT VERIFICATION WILL BE CHANGED TO 1:00 THURSDAY
TO AVOID CONFLICTING WITH RICHARD'S TALK.
∂14-APR-76 1619 FTP:PHW at MIT-AI
Date: 14 APR 1976 1918-EST
From: PHW at MIT-AI
To: jmc at SU-AI
John,
Someone named Lilly Chertov (phonetic spelling)
from the Union of Concerned Scientists asked me
to write a letter to Chavchanidze about some
Russian named Bralovski (also phonetic). I told
her I would think about it, but did not want to
get him in trouble since he seemed like a good
guy. Moreover, I told her I was interested in
your posture. Do you have one on the subject?
Patrick
-------
I think that it isn't reasonable to write Chavchanidze about
something Brailovsky reports that Chavchanidze said to him
in private conversation. So far as is reported, Chavchanidze
took no official action concerning Brailovsky who even lives
in a different city. If I were having a conversation with
C and had heard about the matter from B, I might ask C whether
and why he thought B had only political motivation for wanting
to attend the symposium. In short, there are enough solid
injustices to beef about, so that we should avoid something as
tenuous as this.
∂14-APR-76 1656 FTP:PHW at MIT-AI
Date: 14 APR 1976 1954-EST
From: PHW at MIT-AI
To: jmc at SU-AI
I agree. I will not write.
Patrick
-------
∂14-APR-76 1757 DCL
To: GROUP.LST[VCG,DCL]:;
I would like to keep the old time if possible-David
∂14-APR-76 2347 DCO THURSDAY SEMINAR
To: GROUP.LST[VCG,DCL]:;
DAVID IS TALKING ELSEWHERE AT 115 PM THURSDAY SO WE WILL HAVE THE
SEMINAR AS USUAL AT 330. SORRY FOR THE CONFUSION.
∂15-APR-76 1028 TOB arpa image understanding
There was an image understanding meeting
at USC Mon and Tues, April 12-13. These
will be on six month basis in the future.
There are several new organizations in
the program, including Maryland, Purdue,
CDC,Honeywell,Westinghouse,Hughes.
Heilmeier apparently now supports image
understanding, and there will be a slight
budget increase for the program next year.
See me for any other info.
Tom
∂15-APR-76 1427 FTP:PHW at MIT-AI
Date: 15 APR 1976 1725-EST
From: PHW at MIT-AI
To: jmc at SU-AI
A Russian named Butakov wants to visit us for severl
months. Have you ever heard of him?
Patrick
-------
No.
∂15-APR-76 1627 REM via AMET
I have completely rewritten the introduction to crunching and spindling, CRUNCH.PLN[1,REM](2)
∂15-APR-76 1822 REM via AMET
CRUNCH.PLN[1,REM](3) contains a summary of existing crunch/spindle programs
∂15-APR-76 2321 REM via AMET That size of crunched file bug (total size of spindle)
I had MOVE E followed by SUBI D instead of SUB D, so it took
the top boundary of the file minus 4 rather than minus the lower boundary.
∂16-APR-76 0123 REM via AMET New CRU3.DMP[1,REM]
Contains everything you requested except default trees.
∂16-APR-76 0211 REM via AMET
A default is now in CRU3.DMP
∂16-APR-76 1016 FTP:gjs@mit-ai at MIT-AI
Date: 16 APR 1976 1315-EST
From: gjs@mit-ai at MIT-AI
Sent-by: DVM at MIT-AI
To: jmc at SU-AI
I just read ur review of Joe W.'s book and his rebuttal.
I agree with ur position but I feel that u made a serious tactical
error. The "new left" issue is a phony one which gives JW a target
to shoot down. I suggets that u find some other gensym
to reference the appropriate properties.
-------
Well you are right about the "new left" property. Something along
that line is correct and relevant, but if I can't think how to put
it precisely, I had better skip it.
∂17-APR-76 0712 RWG
as you have probably noticed, weizenbaum is now joseph%mc .
Can you send me a copy of your draft paper on computer message
services? You may be interested in a paper on home terminals
called hoter[w76,jmc] on Home Computer Terminals that I gave at
the AAAS meeting. Message services fit into a larger picture, and
some proposals for regulating message services will preclude other
information applications. I regard the New York Times editorial to
turn these services over to Postal Service monopoly in order to
subsidize ordinary mail as particularly poisonous.
John McCarthy, Artificial Intelligence Laboratory, Stanford CA 94305
∂17-APR-76 2013 BPM
[John: Thought you might be interested in this. Panko is at SRI
Telecommunications Center (?).]
[OFFICE-1]<PANKO>SUMMARY.NLS;2 12-APR-76 07:17
Computer Message Services--summary of a draft paper (longish)
The following is the summary from a draft paper entitled "The Outlook
for Computer Message Services: a Preliminary Assessment." Please let
me know if you would like a draft copy to review (or just peruse).
Since the paper is still somewhat drafty, please don't quote any of
these numbers without checking back with me.
The Non-Medical Use of Drugs Directorate in Canada began using a
computer for its internal communication in late 1974. During its
first six months on the system, NMUD's 92 staff members sent 20,800
messages, at an average price of about $1.50 per message. The
Directorate is very happy with this instant communication medium,
whose terminals are so inexpensive that they can be carried around by
staff members on the road..F="Draft for Comment .Split; Summary";
The Directorate is not the only user of "computer message service."
Other users of computer-based communication systems include the U.S.
Geological Survey, the Office of Telecommunications of the Department
of Commerce, Gulf Oil, the National Security Agency, Stanford
Research Institute, the Business Planning Group of Bell Canada, and a
host of other commercial and governmental groups. Some
organizations, in fact, have been using computer message services
(CMS) for more than a decade.
Today, almost any organization that wishes to use CMS can choose
among a number of software packages for their in-house computers, or
among a number of message services on national and international
computer networks. Even at current prices -- $0.75 to $1.75 per
message on most systems -- computer preparation and delivery may be
cost-competitive with inter-office and postal delivery. Furthermore,
CMS prices could fall to between $0.25 and $0.65 within a year; by
and 1985, conservative cost projection indicates that prices will
fall to between 15 and 30 cents per message. At those prices, most
organizational mail is likely to flow by computer rather than by
mail-carrier. Although our cost projections are preliminary, it is
hard to escape the feeling that a communication revolution of
unparalleled scale is about to unfold.
Proliferation of computer message services could raise a number of
sticky policy issues. For example, the Federal Communications
Commission has already asserted authority over computer "message
switching." But the extent of its authority is unclear, and its aegis
over "electronic mail" may be challenged by the U.S. Postal Service.
It even seems to be an open question whether any existing agency
should be allowed to control computer message services, or even
whether extensive regulation would promote or stymie the emergence of
a healthy national communication.
In other nations, the British Post Office has already taken official
action against computer message services, specifically against users
of the TYMNET and the ARPANET; the Canadian Department of
Communication is now conducting policy-related research on computer
message services. The effect that policy decisions of other nations
could have on the development of computer message services in the
U.S. is unknown.
Record communication already constitutes a large fraction of
international telecommunications traffic. For example, the Canadian
international carrier, TELEGLOBE, obtains 20 percent of its revenues
from Telex traffic. Other carriers are believed to obtain similar
percentages of their total traffic from teletype services. If
computer message services could displace current Telex services, the
development of international agreements on computer message service
should be given high priority, but may become difficult to
interconnect the message services of different nations if radically
different courses of action are taken in the regulation of computer
message services.
A major question that should guide policy makers is whether a
national system of computer-based human communication --
incorporating not only computer mail and computer conferencing but
also electronic funds transfer, document and correspondence control
and other tools to support the diverse forms of communication that
take place in organizations -- is to be created, or whether a
fragmented message service industry -- offering computer mail,
computer teleconferencing and other services as separate and
fragmented tools -- is to be allowed or encouraged to exist.
Particularly if office automation is to be a key to the future growth
of productivity in the United States, whether or not regulators take
an integrated view of computer message services may have
repercussions far beyond affected communication industries.
In the past, CMS has been an invisible industry. Usage has generally
gone unreported, often because of concerns over uncertain FCC
responses. Regulatory agencies, in turn, have generally been
ignorant of computer message services or have been unwilling to
become involved until a clearer picture of CMS emerges. While benign
neglect has probably been beneficial in the past, it remains to be
seen whether a potentially key communication medium can mature into a
coherent national communication system without oversight or whether
the continuing lack of information flows caused by past regulatory
actions is in the public interest.
But the theoretical question of whether or not government SHOULD get
involved is rapidly becomming moot. Walter Hinchman, chief of the
FCC's Common Carrier Bureau, has recently announced that the
Commission is preparing to re-open the Computer Inquiry. By late
1976, computer message services are likely to lose most of their
invisibility.
This paper is a limited incursion into the area of computer message
services. It merely surveys costs, market potential and the general
regulatory picture, in order to show that a large and potentially
policy-laden medium appears to be emerging. It does not probe these
areas deeply, much less consider the broader questions of what
impacts computer message services will have on organizations and the
general public if they are or are not allowed to develop rationally.
∂18-APR-76 0742 FTP:JOSEPH at MIT-MC
Date: 18 APR 1976 1035-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
John:
Happy Easter (or good Yon-Tiv, as the case may be)!
I agree with you that the exchange of responses is tending to
approximate the combinatorial explosion and that we should find
better ways to communicate our views to the community at large.
I must say, though, that your magazine is a fine idea. We
might consider writing something for it along the lines of our
discussions after we have reached some sort of terminal node.
Meanwhile, your idea of a debate sounds fine. I do fear that
I am talking past you - perhaps you have the symmetric fear
as well. But that should not stop us from talking. Also, I
detect a consiliatory note in your last statement. I will, in
my turn, try to tone down the stridency to which, I am
afraid, I am much addicted.
I agree also that certain subjects are best left out of
all this, e.g., whether DOD is "bad" or not and whether there
is a "pillow fight" at M.I.T.
I cannot say now when such a debate as you suggest
(demand?) can take place. The semester here is drawing to a close.
Perhaps a carefully scheduled and thought through - e.g., with
respect to format and all that - meeting can besst be mounted
in the fall. Perhaps we should even think of getting PBS involved.
(I agree with you, by the way, about their "commercials" and
their generally self-congratulatory view of themselves. Still,
I watch a lot of it and think it much better than most ordinary
TV.)
This medium is, however, a not bad way to pass messages
of about the length of this one. I will look for yours.
Joe W.
-------
∂18-APR-76 2348 KRD outlandish idea
The following idea occurred to me while stopping at the 25th stop
sign in a row the other day, and I've been wondering about it. Since
you seem to know something about automotive technology, perhaps you
can tell me what its plausibility is:
it seems silly to spend all sorts of energy accelerating a
car, and then dissipate it in heat slowing it down again using
the brakes. So why not do the following: add to the car a
flywheel, and a transmission that connects the flywheel with
the car wheels. The brake pedal is replaced with a clutch
activating the new transmission. You depress this clutch when
you want to slow down, and because the new transmission is set
to make the car wheels attempt to accelerate the flywheel (no
matter how fast it is already going), the car decelerates.
(The new transmission changes gears as the car slows, to keep
car wheels doing work, slowing the car, and accelerating the
flywheel.)
You thus slow down, but some fraction of the energy you had is
now stored in the flywheel, and can be used to re-accelerate
you or keep you going once you're moving.
Is `flywheel technology' anywhere near up to making this workable?
Flywheels
This proposal has been made many times for cars. Buses
with flywheels that are electrically spun up have been used in
Switzerland for a long time. It is also proposed to charge batteries
while slowing down, and I believe that some electric trains use their
motors as generators while slowing down and pump energy back into the
lines. I suppose this works best in DC powered systems.
There are two problems:
1. The flywheel system increases the overall weight of the
car and thus reduces the gas milege when the car is not braking.
Therefore, a net energy saving depends on certain assumptions
about the fraction of time the car is used in stop-and-go driving.
I have never seen an analysis of this.
2. Cost. At present prices of gasoline it probably wouldn't
pay someone to buy it and keep it in repair. Increased gasoline
prices may improve this picture.
There have been recent advances in materials technology
that make energy storage in flywheels more plausible, and the
idea has its enthusiasts, one of whom, I believe, is at SRI. The
enthusiasts propose a smaller engine and keeping the flywheel spun
up and using it for acceleration. Usually the proposal inv∨lves
a continuously variable gear-ratio transmission, and the technology
of this has also improved recently. I think there have been
articles about it in Popular Science or Popular Mechanics
within the last year and one in Scientific American within the
last few years. The extremists propose spinning up the flywheel
at home, but it seems very unlikely that the required amount of
energy can be stored.
∂19-APR-76 0741 FTP:PANKO at OFFICE-1 (Response to message)
Date: 19 APR 1976 0732-PST
From: PANKO at OFFICE-1
Subject: (Response to message)
To: JMC at SU-AI
cc: PANKO
In response to your message sent 18 APR 1976 0429-PST
John, I'll send you a copy of my paper ASAP, and I would very much like to read your AAAS paper. Yes, terminals in the home certainly represents a potentially very important area. We in the Telecommunication Sciences Center are now doing a technology assessment of communication/travel interactions. In one of our scenarios, we are assuming work at home and extensive terminal support. Perhaps we should get together to chat about that project some time. Ra3y
-------
∂20-APR-76 0514 REM via AMET BUG IN SYSTEM MAKING CRU3 BREAK!!!!
WARNING -- CRU3.DMP[1,REM] and all later versions of CRU3 attempt to do
a RENAME before a CLOSE to set file date-written etc to be the
same as the original input file was. Due to a bug in the system,
the RENAME causes the USET pointer to be zeroed, causing the CLOSE to
write the final buffer of data on top of the first record of the file
instead of at the end of the file where it belongs, thereby garbaging
your output file. The offending instructin is at B328E+6 in the core
image of CRU3 in case you have a spindled file you REALLY want to unspindle
rather than restore from backup tape.
∂20-APR-76 0731 FTP:REM at MIT-ML
Date: 20 APR 1976 0704-EST
From: REM at MIT-ML
To: JMC at SU-AI
The following message was all sent to you, but system died just
before I hit ↑Z at the end...
.MAIL JMC Subject: CRU3 modified to avoid system bug Type message followed by <CONTROL>Z RU C3[1,REM] for RENAME after CLOSE to avoid system bug. It seems to work ok, but haven't tried it extensively.
-------
∂21-APR-76 1140 DCO THURSDAY VERIFICATION SEMINAR
To: DCL, DCO, RSC, RAK, JFS, JJM, WLS, RLD, JP, TL, PMF, RWW, JMC, CM
CORKY CARTWRIGHT WILL CONTINUE DISCUSSING HIS VERIFIER FOR
TYPED LISP. THURSDAY AT 3:30 .
∂21-APR-76 1544 FTP:REM at MIT-MC
Date: 21 APR 1976 1837-EST
From: REM at MIT-MC
To: JMC at SU-AI
This morning, after the system was back up, I typed in the code I
wrote the day before (CRU4.SAI) to keep frequent words in a hash table
in core as it parses words in text using your algorithm for breaking
up words that have mixes of upper-case and lower-case (McCarthy
breaks up as MC CARTHY, jKIhYJKGJhhfyjkjjKJ breaks up as J KI H YJKGJ
HHJYJKJJ KJ). It has the additional feature of displaying the hashing
graphically on any crt terminal with cursor-control characters and tabs,
such as my Beehive and probably the Datamedias that everybody else has.
(Need to do TTY NO ARROW and TTY TAB to avoid system conversion)
The graphical presentation on my Beehive already showed me a badness in
the first hashing algorithm, all single-letter words were hashing to the
same cell, which I fixed (now the hashing is excellent).
This program is a front-end to SSORT which will be followed by
a collator to collect all batches of the same word into one (adding the
count) and another SSORT to put words in sequence by count (they would
already be in alphabetical order after the collator, one item per word).
This will allow us, or a program, to make calculations of crunching by
various word-oriented methods like you proposed.
-------
∂21-APR-76 2136 KRD distinction
To: RWW, JMC
There are two standard ways to define a set: by listing all the members, or
by giving a predicate that indicates the membership criteria. It seems as
though there ought to be a widely-known, single-word term for each of these.
Is there? (I keep thinking of intensional vs extensional, explicit vs
implicit, but neither seems correct. I need the distinction for something
I'm writing.)
Any ideas?
Thanks
There are yet other ways. For example, one may have a word whose ones
indicate which elements of a larger set are in a smaller set. The
predicate is usually called a property, and in Zermelo-Frankel
set theory, not every property defines a set, e.g. the property
¬(xεx) doesn't define a set. However, for any set A and any property π,
there is a set {xεA|π(x)}. I don't know terms for the case of
finite sets presented as lists, but intensional and extensional
mean something quite different.
∂22-APR-76 0723 FTP:BEN at MIT-AI Book recommendation
Date: 22 APR 1976 1055-EST
From: BEN at MIT-AI
Subject: Book recommendation
To: jmc at SU-AI
I would like to recommend that you read
"Small is Beautiful" by E. F. Schumacher.
It is an "alternate economics" book, and
you will probably find that much of it is
"New Left" ideology, but there are also many
very good ideas in it which are worth knowing
if you are criticizing the modern world. In
particular, he criticizes the lack of expressive
power in traditional economic theory, and demonstrates
quite convincingly how much of the foolishness of
governmental policy results from failing to make
important economic distinctions.
Speaking of "New Left" ideology, it was such off-the-cuff
attacks that made your review and reply to JW somewhat
difficult to deal with. Particularly since he is vulnerable
to criticism for being hysterically polemical, you should
be very careful to avoid the same sin.
-------
I have read "Smll is Beautiful". I have removed the
"new left" phrase from the current version. Schumacher
is not "new left"; he is perhaps "eco-left". Since the
book is a collection of essays written at different times,
it is inconsistent. In one part he takes nuclear energy
for granted; later he turns against it - for rather weak
safety reason, but I suppose really because it doesn't
fit his "small is beautiful" theme. However, you guess
right; I don't like it either.
∂22-APR-76 1322 FTP:BEN at MIT-AI
Date: 22 APR 1976 1620-EST
From: BEN at MIT-AI
To: JMC at SU-AI
Inconsistency has never struck me as being a serious problem
when someone has good ideas: you take the good ideas and
work out the inconsistencies in the applications. The only
use of inconsistency is to reveal when an attractive idea
is so badly flawed as to have no use, which seems rare.
Small is Beautiful has a couple of good ideas which are,
I think, important and worth considering:
1. Traditional economists describe value with a one-dimensional
measure, dollar (or market) value, which is inadequate for
expressing the more global aspects of its value, like eventual
depletion. The market value usually takes a shorter view of such
things. Thus, by using an inexpressive measure at the heart of
the theory, economists come up with bizarre conclusions.
2. Traditional methods of economic aid to underdeveloped
countries overlook some important facts about economic development,
and so fail, even by their own standards. Those facts are fairly
simple ones about needing to develop a relatively large base of
medium-cost workplaces (i.e. jobs) before trying to create a small
number of highly capital-intensive workplaces. Doing only the
latter will fail to create the desired economic development.
This is NOT an argument against exporting technology! It could
easily be that cheap personal computers, or terminals, will make
possible medium-cost workplaces of great productivity, for example.
It does argue against helping underdeveloped countries produce
high-technology (i.e. high capital investment per workplace)
objects, because the conditions are not right for translating that
kind of industry into long-term economic growth for the country.
Nuclear power (probably fusion in the long run) is almost certainly
the answer to our energy problems, but in the meantime, I don't want
the landscape cluttered up with decaying radioactives that are
dangerous over geological time. Answer: spend more money on the
space program and put those things where they are safe.
Ben
-------
∂22-APR-76 1507 REG Benchmarks
To: JMC, LES
I think that benchmarks ought to be run under controlled conditions.
To this end, I suggest that a group of benchmarks be collected that can
be run on a bare system. Then, some time will be assigned for the
purpose of running this set of tests on both systems. Anyhow, it seems
unproductive to run benchmarks (for the sake of benchmarks) on our
already crowded system, especially in view of the known inconsistency
of results.
Benchmarks should be short (< 1 minute) and should be run under both
time-sharing and bare machine. Under time-sharing, say 3 times.
∂22-APR-76 1639 REM Beehive programs running on Datamedia in ME's office easy!
A few minutes of trivial changes and both my SETTAB and CRU4 programs
work on the datamedia just like on my Beehive.
∂23-APR-76 1252 LES Benchmarks
To: REG
CC: JMC
Your proposed bare machine tests would yield bare machine performance
comparisons, but those are fairly easy to come by and not particularly
relevant. The question is: how much faster is the KL under load?
∂23-APR-76 1951 DCL
To: GROUP.LST[VCG,DCL]:;
***********************************************************************
NOTE : TIME CHANGE--WEDNESDAY THIS TIME ONLY (SO CERTAIN PEOPLE CAN ATTEND)
VERIFICATION GROUP MEETING WEDNESDAY 27th APRIL
TIME 3.30 pm.
JORGE PHILLIPS "Verification of a mini computer operating system"
This is work Jorge has been doing with Tom Bredt and is the subject of
a recent paper. It should provide us with some alternatives to Brinch
Hansen's way of doing things.
**************************************************************************
∂23-APR-76 2320 REM via AMET
Statistics for HOTER.ESS[W76,JMC] are in HOTER.CR6[1,REM] --
for example, 56.67 % of the different words are single-occurance words,
accounting for 14.415 % of the total count.
Now, is there a particular scheme for crunching you'd like me
to investigate, based on these new word-counting programs,
or shall I investigate whatever simpleminded method I come up with?
I would like also to have the count of double and triple occurence
words and their cumulative frequecy, because they may also need
special treatment. Otherwise, do the simpleminded.
∂24-APR-76 0036 REM via AMET Double and Triple occurrance words
If you look at HOTER.CR6[1,REM] you will see that it contains that
info. Here's a verbose transliteration of the first few lines of
that file:
892 one-each words, .56 by different-word, .14 by count
259 two-each words, one+two add up to .73 by dif.w .22 by count
131 three-each, one+two+three add up to .81 by dif.w .29 by count
78 four-each, runnning total .86 by dif.w .34 by count
etc. ... at the end, 1.00 of the different words and 1.00 of the total
frequency-of-occurrance (count) have been accumulated, of course.
Subtracting any two adjacent items tells the amount in that
item, subtracting any two nonadjacent running totals tells the total for
all items between them. Let me know if yo still have any trouble...
Ok, fine
∂24-APR-76 0147 FTP:REM at MIT-DMS
DATE: 24 APR 76 0531-EDT
FROM: REM at MIT-DMS
ACTION-TO: JMC at SU-AI
MESSAGE-ID: [MIT-DMS].31522
Right now, I'm optimizing the output from CRU5.SAI so that SSORT won't
thrash so much as it does now. Unfortunately SU-AI went down...
∂24-APR-76 0323 REM via AMET Success at making CRU5.SAI optimize output so that SSORT doesn't thrash.
CRU5.SAI outputs 7 files of data to sort and feed into CRU6.SAI
count=1, first letter ' 0-9 A-C
count=1, first letter D-L
count=1, first letter M-Z
count=2
count=3,4
count=5,6,7,8,9
count=10 up
These seven files are then concatinated and sorted by SSORT.
The paging algorithm of SSORT in effect causes separate sorting of
each of the 7 blocks of data, followed by concatination of the result,
so no thrashing occurs with data accumulated from NOTICE[UP,DOC].
Next to try it with a big one, WUTHER or GRIMM !!
∂24-APR-76 1029 FTP:SIGART(X180LE03) at CMUB Weizenbaum's review, etc.
From: SIGART(X180LE03) at CMUB
Date: 24 Apr 1976 1329 EST
Subject: Weizenbaum's review, etc.
To: JMC at SU-AI
cc: SIGART, FIKES at SRI-AI
- - - -
John,
I have just heard, via Minsky, about your review of Weizenbaum's
book, and would very much like you to submit it for the SIGART
Newsletter. (Our next issue goes to press about May 25.)
At Rich Fikes' suggestion, I am seriously considering printing
Dreyfus' review of the books "Artificial Intelligence" (Minsky and
Papert) and "The Psychology of Computer Vision" (Winston). (This
review also apparently appeared in "Creative Computing" in the
March-April 76 issue.) Minsky does not want to respond to it (and
feels I should not print it). Might you be interested in responding?
(If you have not seen the review, I can send you a copy.) Perhaps you
might want to combine your response in some way with the Weizenbaum
one.
Also, please put me on your "subscription" list -- ERMAN@CMU-10B.
thanx,
Lee
-------
I am trying for a larger audience for my review, but if I don't have
it by your deadline, SIGART will be fine. I would like to see the
Dreyfus review, but I am somewhat surprised that he considers
SIGART Newsletter a good place to publish it. However, it shows
a desire to interact with the AI community, which, incidentally,
Weizenbaum also shows, so if you print my review you will also
have to print a reply.
Afterthought
I may want to respond to Dreyfus. The review is WEIZEN.REV[PUB,JMC]
and I suppose you can put it through PUB at CMU.
I don't want to respond to Dreyfus. His criticism is aimed at
the philosophical advertisement of Minsky's and Winston's work,
and I am dubious about that, but I don't consider Dreyfus's
philosophical citations to be relevant either, and I don't want
to do the reading.
∂24-APR-76 1529 DCL
To: GROUP.LST[VCG,DCL]:;
***********************************************************************
NOTE : TIME CHANGE--WEDNESDAY THIS TIME ONLY (SO CERTAIN PEOPLE CAN ATTEND)
VERIFICATION GROUP MEETING WEDNESDAY 27th APRIL
TIME 3.30 pm.
CORRECTED TITLE
JORGE PHILLIPS "Design and Verification of a class of Real Time Systems"
This is work Jorge has been doing with Tom Bredt and is the subject of
a recent paper. It should provide us with some alternatives to Brinch
Hansen's way of doing things.
**************************************************************************
∂24-APR-76 1623 REM via AMET New SSORT desirable
The optimized list of words with counts created by CRU4.SAI
when GRIMM[LIB,DOC] is the input file is too large for SSORT to
efficiently sort. There are over 48,000 records (words) in the
file, causing SSORT to run in 208 pages of core (104 k) which is
too large for this system to support effectively. Thus even though
the output from CRU4 is optimized to make SSORT run efficiently, in the context
∂24-APR-76 1638 FTP:REM at MIT-MC
Date: 24 APR 1976 1930-EST
From: REM at MIT-MC
To: JMC at SU-AI
(system too busy to complete message at SU-AI)
... , in the context of`a timesharing system without paging, SSORT
isn't able to handle more than 25,000 records (items to be sorted)
efficiently, so a new version of SSORT that can effectively handle large
numbers of small records is needed if our research into word-oriented
is to progress very far. (Ie if we ever expect to make up a dictionary
for sorting a large collecdion of files or a large fle like GRIMM or
WUTHER, we NEED a new SSORT!!)
[ERRATA after "word-oriented" insert "crunching"]
-------
∂24-APR-76 1640 FTP:REM at MIT-MC
Date: 24 APR 1976 1930-EST
From: REM at MIT-MC
To: JMC at SU-AI
(system too busy to complete message at SU-AI)
... , in the context of`a timesharing system without paging, SSORT
isn't able to handle more than 25,000 records (items to be sorted)
efficiently, so a new version of SSORT that can effectively handle large
numbers of small records is needed if our research into word-oriented
is to progress very far. (Ie if we ever expect to make up a dictionary
for sorting a large collecdion of files or a large fle like GRIMM or
WUTHER, we NEED a new SSORT!!)
[ERRATA after "word-oriented" insert "crunching"]
-------
I suppose the most reasonable sort for large files would sort as much
as reasonable within core, write a file, sort another block, etc., and
finally merge the resulting files. I suppose also that some sorting
might be necessary withing the cruncher, but since only one copy
of each word need by retained, it looks like we could get by with
much less. I suggest you consider whether you can get by without
a new SSORT for now; if not write it; if yes consider what the
situation will be when the KL-10 comes up and when paging is
more effectively used in the system.
∂24-APR-76 1737 REM "ONE COPY OF EACH WORD"
Until the sort is sufficiently eell al{n$, there is n{ ay t{ $et
rid {f duplicates {f {rds and have {ne {f each. The purp{se {f the su{rt
is preciselty t{ brin$ separate batches {f the same {rd t{$ether and
add their c{unts t{$ether t{ make {ne master c{py ith t{tal c{unt f{r
the {rd. IN tRIMM THER ARE PROBABLY ABOUT 30,000 DIFFERENT WORDS,
SO ONE COPY OF EACH EVEN AT THE END OF THE SORT WOULD MAKE THE
PROGRAM TOO BIG EVEN WITH PAGING (THERE ARE ONLY 18 BIT ADDRESSES,
WITH ACTUALLY 17 BITS FOR THE LOWER SEGMENT). IT IS PREFE$$$$
IT IS PREFERABLE TO HAVE ONLY ABOUT 20K PDP-10 WORDS OF DATA IN CORE
AT ANY ONE TIME SO THAT THE DATA AND OTHER PROGRAMS CAN SHARE
CORE WITHOUT SWAPPING OR PAGING MUCH. AFTER PAGING PERHAPS MORE CAN
BE KEPT AROUND BUT THE SORTING ALGORITHM MUST STILL BE DESIGNED
INTELLIGENTLY TO AVOID THRASHING. SINCE REG SAYS THERE IS NO SCHEDULE
OR PLAN CURRENTLY TO EVER IMPLEMENT TRUE PAGING, THIS IS PROBABLY A
RED HERRING. I.E. T$ S$RT THE DATA IN GRIMM, IT IS NECESSARY TO MAKE
A NEW SSORT THAT RUNS IN 20K EVEN WITH A LARGE FILE, RATHER THAN 104K.
(ACTUALLY 30K. 10K PROGRAM PLUS 20K DATA IN CORE.)
HOWEVER, IF YOU AR SATISFIED TO NEVER EVER CREATE A LARGE DICTIONARY,
THEN SSORT AS IT NOW STANDS, WITH THE OPTIMIZED OUTPUT
FROM CRU4 AND CRU5, SHOULD SUFFICE, FOR EXAMPLE NOTICE[UP,DOC] WAS
ADEQUATELY DICTIONARIZED WITH THE CURRENT PROGRAMS. (SEE NOTICE.CR6[1,REM])
∂24-APR-76 1932 REM CRU4.WRU
I have retyped most of the 3{ line text that as l{st hen the ames-tip
dropped carrier and SU-AI killed my j{b. The file is CRU4.$RU and
expal xxx explains h{ t{ use my ne pr{$rams t{ c{mpile a dicti{nary
(a {rd and c{unt list f{r a file). D{nrt d{ it ith any file lar$er
than ab{ut 50k because that is the lar$est file that can be effectively
na xxx handles xxx handled ith the present SS$RT.
With all the ames-tip numbers killin$ mu xxx my c{nnecti{n
rand{mly {r insertin$ ↑C randomly in the middle {f hat I type in,
and with TTY11 d{in$ its usual $arba$in$ $arba$in$ tARBAGING GARBAGING
of whatever I type, there is alm{st n{ ay t{ $et much {rk d{ne effectively.
p.s. I am not the only one able to make TTY11 go int{ its $arba$in$ m{de.
There is a random pers{n h{ asked me t{ link t{ him a fe days a${, and
hile ere ere xxx hilexxx $HILE $E XXX
WHILE WE WERE TALKING WITH HIM ON TTY11 AND ME ON AMES-TIP,
I NOTICED THAT WHAT HE WAS TYPING WAS GETTING GARBAGED IN
THE SAME AS WHAT I TYPE DOES. I ASKED HIM WHAT TYPE OF TERMINAL AND
MODEM HE HAD, AND IT TURNED OUT HIS TERMINAL IS A DIFFERENT
BRAND FROM MINE AND HIS MODEM IS A DIRECT-COUPLE HOMEBREW THING. THUS
IT IS VIRTUALLY IMPOSSIBLE THAT THE TROUBLE WITH TTY11 IS ALL IN MY
TERMINAL, OR MOSTLY IN MY TERMINAL. $$MEDAY I'LL ET AROUND TO MAKING
CONNECTION WITH TED AND GOING THROUGH THE HASSLE OF DISCONNECTING
EVERYTHING HERE, CARTING MY TERMINAL AND MODEM UP TO THE LAB, AND LETTING
HIM PROBE INTO TTY11 ETC. AS WE MAKE IT ACT UP.
∂24-APR-76 2147 REG
∂21-APR-76 1903 100 : BILLSCHELL via RTGT GUEST ACCOUNTS
YOUR ELIMINATION OF A GUEST ACCOUNT HAS NOT ONLY STOPPED MISUSE BUT
HAS ALSO GREATLY INCONVIENCED LEGITEMATE PROGRAMMERS WHO WOULD LIKE
TO USE YOUR MACHINE. PLEASE REINSTATE THESE ACCOUNTS, EVEN IF THEY
ARE OF A RESTRICTED NATURE. IF YOU DESIRE TO CONTACT ME, YOU MAY
SEND MAIL TO USER BILL/S ON THE MIT-ML MACHINE (LOCATION 198).
THANK YOU FOR YOUR TIME.
WILLIAM SCHELL
RUTGERS UNIVERSITY
CENTER FOR COMPUTER AND INFORMATION SERVICES
Our machine is about the most crowded on the network, but if there
is some facility here that you think would be of benefit to our
work to have you use, please send me details - John McCarthy, Director
∂26-APR-76 0834 REM via AMET Working on word-dictionary crunching scheme...
A simple-minded method is to separate delimiters from words,
using "A" in delimiters info to indicate co-return to insert a word,
and using " " in words info to indicate co-return to insert some delimiters.
Each word is prefixed with a mode indicator (UPPER-CASE, Capitalized,
lower-case) and then follows either a 0 bit followed by the crunched
word, or a 1 bit followed by the Huffman code for the entry in the
dictionary. The delimiter stream would be crunched in the CRU1/CRU2
way exactly.
A smarter method is to use "U" or "C" or "L" in the delimiter
stream to indicate the mode of the word to be inserted, thus the first
word after ". " is usually capitalized and others aren't, so the compression
ratio ought to be better not only because of context but because the
separate code for the prefix in the simple-minded scheme is wasteful
even standing alone. I am now making a complete analysis of this
crunching scheme, partly by hand, partly aided by existing programs,
for POX.LOG[1,REM] (because it is small enough to be doable by hand).
∂27-APR-76 0910 REG
∂27-APR-76 0027 JMC slowness
Does the system spend excessive time garbage collecting when there is
very little disk left?
The system doesn't garbage collect the disk ever. However, when there's
little space left, the system will spend more time looking for free
tracks (i.e., rejecting more claimed blocks between free ones), but I
wouldn't expect that to account for much time.
The fault lies in the load. Do WHO and type R to see what's really
going on; you'll find we're RUNNING at least twice as much virtual
core as physical core.
∂27-APR-76 1424 DML Byte magazine
Please bring in your copies of Byte magazine. Thanks. -- David Levy
∂29-APR-76 0011 DCL
To: GROUP.LST[VCG,DCL]:;
More examples have been added to EXMPLS.VER[VCG,DCL]
∂29-APR-76 0539 RWG at TTY41 0539
macsyma has unlimited floating point. bfloat(<exp>); gives
<exp> to the number of digits specified by fpprec (which keeps changing from
being a function to being a variable. if the variable is undefined, use
the function).
∂29-APR-76 0645 REM via AMET First results for word-dictionary crunching algorithm
Input text POX.LOG[1,REM]
CNTLIM=3 (i.e. any word occuring 3 or more times is entered in dictionary)
INPUT 38412 BITS, DICTIONARY+OUTPUT 27385 BITS, C.R. .713
(Figures for alphanumerics only, assuming input is 6-bit code,
with a special end-of-word character after each word)
Now to survey HOTER.ESS and NOTICE to get realistic results.
∂29-APR-76 0734 REM via AMET
HOTER.ESS INPUT, CNTLIM=7, ALPHA C.R. .679
∂29-APR-76 0738 REM via AMET
HOTER.ESS INPUT, CNTLIM=3, ALPHA C.R. .576
∂29-APR-76 0914 MLM AUTOLOGOUT
The usual situation, though, is that the non-using user has disappeared
from his office and doesn't respond to messages. Do you mean he/she
should be paged for this purpose. Do we return to the old "fair game"
rules of thumb -- too long without running and you are subject to
a keyboard map and k/f operation?
I guess a major shortcoming of autologout was that people would
do something trivial every twenty minutes or so just to maintain
the channel, so I am not arguing against disabling of autologout,
just saying the problem is not so simple.
∂29-APR-76 0923 FTP:REM at MIT-MC
Date: 29 APR 1976 1200-EST
From: REM at MIT-MC
To: JMC at SU-AI
Complete crunching prediction for input fle HOTER.ESS[W76,JMC]
Original file, 8448*36 = 304128 bits
Separated delimiters (ascii) 103236 crunches to 24494, c.r.=.24
Separated upperized words (sixbit) 221322 crunches to 127633, c.r.=.58
Combined outputs, 152127 bits
c.r. from original file to combined outputs c.r.=.47
since this is about the same as we are already getting with the
simple left-context scheme already implemented, the scheme looks
infeasible so-far, but it might do better with NOTICE[UP,DOC] input...
-------
∂29-APR-76 1018 BPM ~autologout(x)
The problem is that one can't find an available channel to tell who has too many
channels or who hasn't used them for 30 minutes or longer.
∂29-APR-76 1910 RDA AUTOLOGOUT
It seems to me that eliminating autologout would greatly inconvenience
a lot of people trying to login without significantly benefiting anyone.
∂30-APR-76 0723 REM via AMET Finishing touch on word-oriented crunching estimates for HOTER.ESS
I have now including crunching the text of the words using the
CRU1 scheme, i.e. the words that are quoted in the dictionary and the
rare words that are quoted in the text. (Recall, the system already
includes crunching the delimiters using the CRU1 scheme and crunching
all words with count at least 3 using dictionary lookup Huffman code.)
Previously, input was 221322 bits and output 127633 bits of which
17400 bits were quoted text in dictionary and 68622 bits were
quoted text in text, total 86022 bits quoted.
The crunching of quoted text reduces 86022 to 60732, reducing the
grand total to 102343 output,
reducing the overall c.r. less than one percentage point,
to c.r.=.462
Thus the new more-complicated
crunching scheme isn't worth implementing, at least for HOTER.ESS
as input. Next to try it on NOTICE[UP,DOC]...
[EXCUSE THE SHORT LINES, I AM ON 961-9650 WHICH IS INSERTING
↑C AT AN AVERAGE OF ONCE PER LINE
(THE SHORTER A LINE IS, THE LESS I HAVE TO RETYPE).]
We will have to talk. My intuition is still not readjusted to agree
with the results you have.
∂30-APR-76 1551 RAK via AMET Autologou
To: JMC, LES
To-JMC,LES
Your systems for finding a propietor of an old line and requesting
that he logout doesn't work too well in those cases where you dial
in or come over the net.
Furhtermore, if you ask anyone, they respond
invariably they think "I may need it eventually, so no."
Autologutt ought to come SOONER!!
Dick Karp
∂30-APR-76 1949 ZM referees
I suggest Hoare, Burstall, Pnueli, Ashcroft, Waldinger, Dijkstra,
and Igarashi. The first five names appeared in my previous list.
Please feel free to eliminate or add names to the list.
I shall return from my trip on June 17. Zohar
∂30-APR-76 1140 JMC
We will have to talk. My intuition is still not readjusted to agree
with the results you have.
[REM - Yes, as soon as I finish trying the full algorithm on NOTICE[UP,DOC]
I'll contact you to arrange a talk...]
∂01-MAY-76 0958 REM Improvement in CRU4.SAI caching algorithm
The purge algorithm for the hash table in CRU4.SAI has been changed
from random to Least-Recently-Used, and the effectiveness (number of items
processed / number of records output) has nearly doubled when processing
NOTICE[UP,DOC]. Any word with count greater than 40 or so now remains in
the hash table throughout the entire run of CRU4.SAI, resulting in a single
record containing the full count. Sorting the output with SSORT now
generates no thrashing messages at all.
I should be finished with the full crunch-analysis for NOTICE
later today...
P.S. The effectiveness is now about 8.7, i.e. 8.7 words are
processed for each output record, which implies the average word-frequency
(total instances / total different.words) is at least 8.7 (and because
the caching is nearly optimal, it isn't much more than 8.7 either).
∂01-MAY-76 1301 REM Final results for word-dictionary crunch on NOTICE[UP,DOC]
Input (incl. nulls etc.) 2,396,160
Delimiters 769,545 crunch to 204,875
Words 1,433,316 crunch to 393,164 of code plus 207,402 of six-bit quoted text.
Quoted text crunches to 136,060
Grand total output = 204,875 + 136,060 + 393,164 = 734,099
c.r. = .306 (i.e. 31%)
Note that the c.r. includes the crunching obtained by deleting
all the nulls at the end of each page of NOTICE[UP,DOC] (assuming I
remember correctly that it is in E format), thus the c.r. isn't mucz
better than obtained by the much simpler scheme already implemented
in CRU1/CRU2/CRU3.
Ready to get together with you when you are ready...
∂01-MAY-76 1316 REM DIALOG
P.S. I sighed up for Lockheed's DIALOG info-retr system. My account
arrived Thursday, and I have used half of my free-time to scan SCI
and NTIS for data-compression things. SCI turned up 26 items which
I will read offline, and NTIS turned up lots of junk which I'm not
listing as of yet plus two interesting items: a Navy document claiming
to be a complete bibliography of data-compression methods, and a
Ph.D thesis claiming to have a method of compressing English text
to 1.5 bits per character (c.r. = 21% in my notation).
In a week I should have list$ngs of these citations in the mail...
∂02-MAY-76 0428 REM
1/3 DONE, THE CRUNCHED WORD-STREAM IS REPORTED IN WUTHEH.LST[1,REM]
∂02-MAY-76 0833 FTP:JOSEPH at MIT-MC
Date: 2 MAY 1976 1133-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
Dear John:
I read the current version of the review in the PUB,JMC
directory. Since it postdates my response to the original version, I
must assume that you read and considered my response. You may, of
course, write anything you like. However, I did give you citations that
make it perfectly plain that I do not believe that the DOD sponsores
speech recognition work "in order to" improve snooping on the part of the
government. I do not believe that there is a committe of scientists and
government people who are "ogres" and who gleefully rub their hands over
the prospect of doing great harm. Still, you persist in writing that I
maliciously, with bias, and baselessly conjecture exactly what I have
told you I do not conjecture and what I did not write.
I suppose you intend to publish (at least a version of) your
review in important journals. I would welcome seeing it widely
distrubted and discussed. However, I urge you to delete the blatant
falsehood with respect to my position on speech recognition. I fear my
response to it would never catch up to your assertion.
Yours,
Joe W.
-------
I still think that you have attacked the Defense Department in an anything
goes spirit. On the other hand, what my review currently says, replies in
too similar a spirit. It will be changed before a version is published;
I am not at all satisfied with what I have said so far. My current opinion
is that all worries are genuine including yours, but right now I think that
to respond to the genuine concerns that motivated your book, because the
concerns themselves are not well expressed, and I would have to speculate
as to what they are. Therefore it is more straightforward to reply to
what you said, i.e. play the game of displaying extremist citations.
I am beginning to think that my time is better spent in promoting
positive applications of science and technology rather than in responding
to unjustified attacks on it. I have a book in the works on new
inventions, and I think I erred in suspending work on it to defend jAI
against Lighthill and you and to take part in the campaign to defend
nuclear energy. Can I not persuade you to undertake something constructive?
∂02-MAY-76 1228 REG CSD Comprehensive Exam
To: TED, JBR, ME, REG
CC: LES, JMC
The programming problem for the Comprehensive Exam will be given during
the week of May 17. We should endeavor to keep the system up, and not
schedule downtime, during that period (which, I believe, ends at 1700
on Friday May 21). Hopefully, KL10 system development should not be
impacted by this.
∂02-MAY-76 1336 FTP:JOSEPH at MIT-MC
Date: 2 MAY 1976 1636-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
John:
I too feel that I am involved in a time sink and that, as you
say, I should be doing something more constructive. I also believe that,
while you and I are certainly far apart on important issues, there is
some quite fundamental ground on which we could agree. I detect in
myself an impulse to stridency whenever I submit my very real
disagreements with technological optimists for judgment to third parties,
e.g., the public. I have a hunch that you and I could talk profitably
(that is, to each other's profit - or at least to mine) were we talking
only to each other. On the other hand, I must confess, I have have long
had an image of you as a person with whom close and non-sporatic
conversation was not possible. Perhaps -- and I say this with no intent
of flattery whatever -- this is due entirely to the awe with which I have
long experienced your intelligence, in other words, to my fear of
appearing stupid in the eyes of someone who obviously does not suffer
fools gladly.
One thing is clear to me: I don't want to retreat from the
responibility for what I have written, but I do want to put the book
behind me and go to work.
Yours, Joe
-------
∂02-MAY-76 1636 REM via AMET Results for WUTHER
input (after deleting nulls) = 4,913,676
output bits = 1,059,224
quoted-alpha crunched = 344,774
delimiters crunched = 413,662
grand total output = 1,817,660
c.r. = .370
See WUTHER.CR6[1,REM] for histigram of frequencies-of-occurrance,
see *.CR6[1,REM] for similar data for other files previously surveyed.
I have looked at and printed wuther.cr6. The only thing that seems
suboptimal is the size of the delimiter output. Since wuther is mostly
English sentences,there shouldn't be so much. It occurs to me that
distinguishing two kinds of occurences of capital letters might help.
Proper names should be capitalized in the dictionary, and sentence
transitions should go with the period.
Would you print the delimiter tree for WUTHER?
∂02-MAY-76 1742 REM via AMET Addenda
Note that the list of left-contexts used in the WUTHER delimiter-stream
crunching was the one created from NOTICE[UP,DOC] on the assumption
that it would save time without losing many important contexts...
The listing of the delimiter code for WUTHER is in WUTHED.LST[1,REM].
If you like I can spool it from here and tell you when it is done.
∂02-MAY-76 1916 REM via AMET
CRU3A.WRU[1,REM] is a brief how-to for CRU3.DMP[1,REM]
∂03-MAY-76 0420 100 : DON Auto-logout
It's all well and good to say "to get a channel, find the proprietor of an old
job and request that he log out", but in most cases the reason an old job is
laying about is that the proprietor has gone away completely. Admittedly it
isn't as though we're short on terminals at this hour of the morning, but it
seems ridiculous to leave terminals signed on when their owners went home five
hours ago! Before I make too firm a judgment, however, what were your reasons
for getting auto-logout removed?
∂02-MAY-76 2220 JMC spindle and crunch
Spindling works ok, but the message you get when you try to crunch is
useless when combined with the writeup. It should suggest some good
existing history and Huffman trees or else offer to make them on the
spot.
[REM - Earlier today I fixed the bug that made Bubble crash under
some circumstances (the INPUT uuo would error-return because of an
attempt to read past the last word of the spindle) - I have now changed
the message you get from "α in the crunch-and-spindle mode
so that it tells you that just <cr> gives you the default. Perhaps
it would help if the writeup at least told the user there were
such a thing as two files needed to perform a crunch, instead of your
"don't tell him any theory, just the commands" format of writeup.]
∂02-MAY-76 2210 JMC
cru3 shouldnt halt when it meets illegal chars in file name.
[REM - Ok, I've rewritten GETSIX and SCNFIL to set the flag FAILED to
-1 if an error occurs, rather than halting, and have modified all
places where either is called to test the flag. There may be some new
bugs introduced, so the program is C3.DMP[1,REM] until it has
ben adequately tested.]
∂03-MAY-76 0550 REM via AMET Timing for CRU3 (actually C3, the new version)
To: JMC, REM
Spindle -- 3/4 second per thousand words
Unspindle -- 1/2 second per thousand words
Crunch-by-pages-and-spindle -- 4 seconds to load default tree, then 2 sec per k
Uncrunch-by-pages-and-spindle -- 3.3 seconds to load tree, then 1 sec per k
Test files were CRUNCH.PLN (8039 words) for maximum time, and SNE.DO
(24 words) to get a second point on the linear function assumed for
Crunch and Uncrunch.
∂03-MAY-76 0911 REG
[I don't (think you) want (me) to think about SPINDL this month.
If you disagree, you could ask again.]
∂03-MAY-76 0917 REG
[Well, I for one, would rather wait for bugs to be fixed]
Bubble -- copies the entire spindle, deleting all wasted space such
as deleted files -- don't do this until I fix a bug in it...
∂04-MAY-76 0628 FTP:JM at MIT-MC
Date: 4 MAY 1976 0929-EST
From: JM at MIT-MC
To: JMC at SU-AI
If you wish you could send the comments to me(jm @ mit-mc).
all other varients(moses @ mit-ml) tend to reach me
sooner or later. I believe Mike's login name is MLDertouzos,
but I could be wrong. In any case whatever you send me
will be transmitted to Dottie .
Regards,
Joel
-------
∂05-MAY-76 0932 FTP:Dehall at SRI-AI Abstracts and Papers
Date: 5 MAY 1976 0934-PDT
From: Dehall at SRI-AI
Subject: Abstracts and Papers
To: Workshop Speakers:
We have not received an abstract for at least one of your
scheduled talks for the Berkeley Workshop on Distributed Data
Management and Computer Networks. The April 26th deadline
for abstracts is past, and the May 10th deadline for complete
papers is approaching rapidly.
We feel strongly that a pre-published proceedings will greatly
enhance the exchange of information at the workshop. Could
you please let us hear from you as soon as possible?
Yours sincerely,
Dennis Hall DEHall@Multics
Don Austin DAustin@Multics
-------
∂05-MAY-76 1702 RCM dinner invitation
My wife and I are very pleased to accept your
invitation to dinner on Thursday May 13.
Since there was some confusion about the
date, please let me know if there is any
change.
∂06-MAY-76 0339 REM via AMET New features in CRU3
Default trees for crunching now are @<last crunched file in spindle directory>
unless there are no crunched files at all in which case you get system default.
Various things cleaned up.
On exit, it predicts what a bubble will do, and if a block will be saved
it tells you. I haven't yet done the "type Y to bubble before exit".
Nor have I timed the program to get an estimate of bubble compute-time.
New code for spindle command allows as many files as you want in one batch.
List of new files is reversed before adding to directory to compensate
for the fact that they were set up in reverse (i.e. L ← (CONS NEW L) each
time a new file spindled or crunched) -- i.e. they will now appear
in the directory in the correct order, previously Crunch command
multifile resulted in backwards additions, and upon adding new code for
Spindle multifile the same thing hapened there.
Let me know if you find any bugs after I rename all the versions
of CRU3. If ok, let me know if there is anything else very important to
do soon. Otherwise I'll default to background tasks such as survey of
.LSP and .FAI files (making a system-default set of trees for each),
trying the word-oriented system on .FAI etc. files, improving POX, etc.
∂06-MAY-76 0352 REM via AMET New version of CRU3.DMP[1,REM] new spindle program.
To: JMC, RF, LES
CRU3.DMP renamed to C3A.DMP
C3.DMP renamed to C3B.DMP
brand-new version CRU3.DMP, tell REM if you find any bugs in it, and roll
back to C3B.DMP if you need to.
Remember not to delete the DSK:<original> of any file until it has
been Pdumped at least once. I am willing to mount dart tapes and
restore up to 5 files if a bug in the spindler destroys the spindle file,
but that's the best I can offer since CRU3 is so new.
I have made a few more changes in SPINDL.REM. Please check it
out again. As soon as you have put in the bubble question,
put it up as SPINDL.DMP[1,3], put a note in NEWS, print some copies
of the writeup, and MAIL * announcing the facility. Suggest that it
be tried on old files to start with.
∂06-MAY-76 2233 100 : REM via AMET
Ok, top priority is modularizing the piece of code for bubble so it can be called from exit routine, then the etc. you listed...
∂07-MAY-76 0007 REM via AMET Bug in CRU3.DMP from 76.5.06 0338 to 76.5.07 0005
To: JMC, RF
WARNING! -- Use of the Spindle command during that period, i.e.
using the version of CRU3 written 76.MAY.06 0338 resulted in subfiles
with the short-byte-count field incorrrectly-written, resulting in
0 to 4 extra random characters at the end of every page. I have
rolled back to the previous version using the old one-file-at-a-time
code until I fix the bug in the new version. I.e. CRU3.DMP has been
renamed C3BADC.DMP and should not be used, C3B.DMP has been
renamed CRU3.DMP temporarily.
∂07-MAY-76 0134 REM via AMET New CRU3, hopefully with all bugs fixed...
To: JMC, RF
If neither of you can find a bug in the new version, I'll put it
up experimentally (with much warning) for users on [1,3] later today
or so. Backup C3B.DMP if there is a bug. After that nasty bug I
found tonight, I'm reluctant to put it up right away for others.
∂07-MAY-76 1324 REM via AMET
I've purged and crunched nearly to rock bottom, and am still 50k over allocation
Considering all the programs I am working on (POX, CRU1-7),
an increase of disk allocation to 200k would almost enable me to
squeeze in.
∂07-MAY-76 1343 WD lock
I started on the door, but met a new difficulty. I will continue either
tomorrow or Monday.
∂08-MAY-76 0142 JMC bug
Did you get my message last night about crunch bug, and is it fixed?
[REM - YES TO BOTH QUESTIONS]
∂07-MAY-76 0402 JMC crunching is broken
An attempt to crunch gallag.le1,gallag.le2 while aliased to let,jmc
causes the message
Crunching DSK:? Error in job 6
Illegal UUO at user 15563
You should have comprehensive test cases that you run before putting up new version.
[REM - I was unable to find the two files mentionned above,
however the bug bites no matter what files are crunched. I have
fixed the bug now, and put up a new CRU3, which
you have used, and which I tested by doing all 5 basic
operations (S U C U B) successfully. - Re the note you SENDed me
while I was RCVing this batch of mail, ok, will do.]
∂06-MAY-76 1322 JMC
I have made a few more changes in SPINDL.REM. Please check it
out again. As soon as you have put in the bubble question,
put it up as SPINDL.DMP[1,3], put a note in NEWS, print some copies
of the writeup, and MAIL * announcing the facility. Suggest that it
be tried on old files to start with.
[REM - Waiting for a few more minutes to be sure it works adequately.
Except for the prompt regarding default trees, is it okπ
Note that default is the last set of trees you used for crunching,
but if you don't have any crunched-spindled files in the spindle at
all it defaults to the English trees. I am not sure how to word
that, perhaps just "cr for default" with more details in πinfo.]
∂08-MAY-76 1055 REM via AMET
Note that the bare linefeed in the file makes it incompatible with
both SOS and ETV, i.e. if you ever edit the file the file will be "mumged"
by having the linefeed replaced with a cr-lf or flushed entirely.
Therefore by definition it is not a text file. However it is a mostly-text
file so I'll try to support it somehow (PTYJOB saved TTY i/o are also
mostly-text files, which ETV barfes at before flushing all the rubouts etc.).
Meanwhile I'll make the error set FAILED and gracefully skip that one file
rather than halt or whatever. (I'll be looking at it later tonight)
I have no requirement that the line feed be saved, so my suggestion
is that if your program thinks it may be that kind of file, it
ask permission to convert it to our form as it goes, saying something
like "bare line feeds, may be SNDMSG file, Convert or Skip?" expecting
C or S as an answer. Need Spindle have any problem at all though?
∂08-MAY-76 2322 FTP:REM at MIT-ML
Date: 9 MAY 1976 0222-EST
From: REM at MIT-ML
To: JMC at SU-AI
Because almost every <cr> or <lf> in text files at su-ai occur inside
<cr><lf> pairs, an early decision in CRU1 et al was to collapse such
pairs to a single token. Ascii 012 was used for this purpose, thus
<cr><lf> in a file gets converted to <lf> inside crunched files and
spindles normally. Bare <lf>'s NEVER occur in output from SOS or E/REFORMAT,
but bare <cr>'s occur just before each page mark in SOS files that have
been copied without linenumbers, so '015 for <cr> is retained. The result
is that text files created or edited at SU-AI, even if copied/nolinenumbers
can be handled ok. However, there is no room left in 7-bit character
representation to hold a 129th character, namely bare <lf>, so that your
file and other nonstandard files won't work. There is a BARELF flag in
CRU2 which can be manually set (by RAID) if one desires to crunch such
a file -- the result of the flag being set is that <cr><lf> is no longer
compressed into a single token, but that <lf> by itself can be represented
as <lf> now. This BARELF flag has not yet been implemented in CRU3 (SPINDL),
although it could be implemented if desired. An alternative I am
thinking of is to continue to represent <cr><lf> as <lf> but to recycle
one of the rare characters like <null> or <rubout> to be some escape
character that can represent <lf> or itself depending on a second
character. Another alternative is to simply reformat such a file to
SU-AI standards, as would happen if it were ever edited with E or SOS.
(You suggested this alternative, the user could abort the spindling of
the file or allow it to be reformatted as it gets spindled.)
Well, there are the major options. The reformat/abort choice
is fairly easy, replacing any bare linefeeds by <cr><lf> tokens,
which upon unspindling become a <cr> and <lf> pair.
-------
OK, do the reformat-abort one.
∂09-MAY-76 0145 REM via AMET
I've checked over SPINDL.REM, making lots of little changes. Looks reasonable now, proceeding with your checklist...
∂09-MAY-76 0147 REM via AMET Proposed entry in NEWS = NOTICE[UP,DOC]
*9 MAY 1976 SPINDL REM
Basic manual for spindling and optional crunching of files is SPINDL.REM[UP,DOC].
Explanation of things in more detail is CRUNCH.PLN[1,REM], including how
the SPINDL program (CRU3) fits in the overall project "CRUNCH".
(I'm not sure at all what belongs in NEWS. Perhaps nothing yet, only
in SPINDL[3,2].)
∂09-MAY-76 0204 REM via AMET Enhancements to SPINDL proposed by JFR
To: JMC
CC: JFR
SPINDL.TXT[1,JFR] contains some specifications which I believe are too
elaborate for the present project (in particular his use of "must" is
inappropriate in most places), however some of his ideas are worth
considering eventually. In particular the bare-linefeed reformating
for almost-text-files, and a binary-mode spindle for non-text-files,
are already planned (perhaps) within a few weeks.
∂08-MAY-76 1230 JMC
I have no requirement that the line feed be saved, so my suggestion
is that if your program thinks it may be that kind of file, it
ask permission to convert it to our form as it goes, saying something
like "bare line feeds, may be SNDMSG file, Convert or Skip?" expecting
C or S as an answer. Need Spindle have any problem at all though?
[REM - I netmailed a detailed reply to this a while ago from ML.
Let me know if you have any further questions or suggestions.
"bare line feeds, may be SNDMSG or PTYJOB file, Reformat or Abort?
>" Although "Reformat" may sould too much like what ETV does when
creating a directory, it should indicate to the user approximately what
degree of munging will occur whereas "Convert" is slightly too gentle
and might cause the user to convert a file he didn't want converted.
Similarily "Skip" might mislead the user into thinking the linefeed
rather than the entire file is to be skipped, whereas "Abort" will
probably cause less confusion.]
∂09-MAY-76 1412 FTP:JOSEPH at MIT-MC
Date: 9 MAY 1976 1713-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
Dear John:
I have viewed the BBC tape Nils Nilsson kindly sent me. The
conclusion I have reached is that the footnote that places the comment I
attribute to you on the BBC tape is wrong. I have taken steps to see to
it that it is deleted from subsequent printings of the book. I do,
however, maintain that the quote is (at least substantively) correct.
Let me explain:
You are heard to say on the tape "...the general formulation of
what the world is like has not been accomplished ... part of this is due
to a defect in our current systems of mathematical logic ..." One may
infer from that that you believe a general formulation of what the world
is like to be possible, that, in the mathematicians sense, there exists a
system of mathematical logic free of the relevant "defects" of currently
known systems and sufficiently powerful to realize your belief. If the
most ambitious goals of AI are to realized with the aid of such a system,
it must be a system with the aid of which one could formalize every
aspect of the real world. I would say that this invloves a contradiction
in terms - for a formalization of any real world phenomenon must
necessarily be an abstraction and must therefore leave out some aspects
of what it is a formalization of. I believe this to be close to the
heart of what divides us with respect to the question of how close AI can
come to capturing every aspect of what it means to be and to behave like
a whole human person.
I believe that at one of the showings of the BBC tape at the
IJCAI-73 I questioned you on the remark you made at the Lighthill debate
and which I here reproduced. I must then and there have recorded your
response on the card from which I months later copied the quote to my
manuscript. It is possible that I misheard you say "simulating" when you
said "formulating" or even that you misspoke. Some months after the
conference I erred in citing your remark as having come from the BBC tape
as opposed to from a discussion ancillary to the showing of the tape.
That is a journalistic error for which I apologize. I am, however,
convinced that the remark I copied onto my note-card and later
transcribed to my manuscript reflects what I thought I heard you say.
You agreed elsewhere that CP&HR raises important and genuine
issues. I trust you will not permit my honest and basically
inconsequential error to overshadow the issues we ought to be discussing.
Yours,
Joe
-------
Datamation has agreed to delete "I fear he made it up."
I hope your withdrawal of the quote can be included in your response there.
Let me suggest that you go easy on claiming that the quote substantially
accurately represents my position. There are context problems even with
literal quotes, and it would be better to devote any space the media may
offer to a statement of my position on the extent that machines can
"understand" human affairs and your criticism of it.
In a fast argument involving several people I might confirm a statement
that I believe the world can be formalized, but if given time to qualify
the statement, I would have, at any time since 1958, have said that
it can eventually be formalized to an extent that will give machines
an "understanding" as good or better than humans have. With regard
to machines "understanding" human affairs, I suppose I would claim
only that machines can understand humans only as well as men can
understand women - admitting that I can't make the claim precise
at present.
∂09-MAY-76 1441 FTP:JOSEPH at MIT-MC
Date: 9 MAY 1976 1742-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
John:
Datamation pressed me very hard to get my response to
your letter in by last Monday. I there say that the tape in question
is on its way to me - thus leaving the question open. I doubt that
they have time to modify that letter now. Would it satisfy you if I
were to send them the letter I just sent you with, of course, a short
covering note? Another tack would be to have them delete the whole
matter from both our letters. That, however, would be asking you for
more generosity than I have the right to ask.
Joe
-------
I can't let the misquote go, because it was a substantial part of
the content of McCracken's review. However, the editor in charge
of the letters column agreed on Wednesday to make my change when
proofs come back, so if you phone him Monday, you may catch it.
∂09-MAY-76 1633 FTP:JOSEPH at MIT-MC
Date: 9 MAY 1976 1934-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
John:
My letter to you did not "withdraw" the quote - it explained
the circumstances surrounding its aquisition. However, I agree with
your point about things said in "fast argument". I am sincerely open
to your suggestion as to how you would like me to modify my letter
to Datamation and, if we can agree (and I hope we can) I will
certainly call Datamation first thing Monday.
Joe
-------
I will be quite happy with any statement that the words
put in quotation marks by you and quoted by McCracken don't appear
on the tape. You may make any inference you like from words I did
say to what I must have meant, because I think my letter
reasonably expresses my position about what "generality" means.
What I said and what appeared in quotes have quite different meanings
to me even if they say almost the same thing to you.
∂09-MAY-76 1705 FTP:JOSEPH at MIT-MC
Date: 9 MAY 1976 2006-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
John:
What do you think of the following as a paragraph in my
Datamation letter - if it can be squeezed in on time.?
Joe
I must confess to an error with respect to the quote John
McCarthy could not find on the BBC tape on which my footnote alleged I
found it. On that tape Prof. McCarthy says only "...the general
formulation of what the world is like has not been accomplished ... part
of this is due to a defect in our current systems of mathematical logic
..." He said what I attribute to him in the book in a discussion
following the showing of the BBC tape. I apologize for the error and for
what is a more serious mistake on my part, namely that, by permanently
recording what Prof. McCarthy said in the course of a fast-paced
argument I have lent solidity to a position which on more careful thought
he would surely have qualified.
-------
∂09-MAY-76 1749 FTP:JOSEPH at MIT-MC
Date: 9 MAY 1976 2050-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
Cannot tell whether your "I will be quite happy..." is in
response to my suggested paragraph or not. Please clarify.
-------
I hadn't seen the paragraph, but I am happy with it.
Well, reasonably happy. I dislike listening to or watching tapes
of my speech, and I don't think I watched the tape of debate at
3IJCAI, and I don't remember a subsequent discussion. However,
I won't pursue the matter further.
∂09-MAY-76 1756 FTP:JOSEPH at MIT-MC
Date: 9 MAY 1976 2057-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
Fine! Intend to substitute "under more tranquil circumstances"
for "on more careful thought".
If it cannot be gotten in the current version, I'll
insist on a subsequent letter. But I will do my best to get
the agreed on version in.
Joe
-------
∂09-MAY-76 1844 FTP:JOSEPH at MIT-MC
Date: 9 MAY 1976 2102-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
its settled then. From now on I'll quote (if at all) only from
generally available printed documents. Good luck with your
proposed book. Good night.
-------
∂09-MAY-76 2151 WD spindl
The spindl does not interface very well with dart. Files hidden inside
spindles cannot be found by locate, even though the spindl has been dumped.
Aside: I think that the efficience of backup(files wanted dumped/files
dumped) would be improved by increasing the temporary dump buffer to about
three to six months and lengthening the time required for a file to get
permanently dumped to between two and four weeks.
∂09-MAY-76 2312 REM via AMET
To: JMC, WD
New SPINDL.DMP[1,3] with the bugs you reported fixed, backup SPINDL.508
∂09-MAY-76 2354 100 : REM via AMET @LICKLI.LE6[LET,JMC]
It is telling you the default, namely to use the @<file> specification
for trees using the last file you crunched. Perhaps the telling-the-default
should be reworded a little...
P.S. WD says CTSS uses "archiv" for what "spindl" does, and a new word
shouldn't be invented. (I wonder what CTSS is? I've heard of it
somewhere before.) But TENEX uses "archiv" for what "dart" does,
so probably best to use "spindl" if we plan to export it to TENEX.
CTSS is long dead, so stick with SPINDL. It was the first major
time-sharing system - on the 7090 at M.I.T.
∂10-MAY-76 1034 FTP:JOSEPH at MIT-MC
Date: 10 MAY 1976 1333-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
John:
Datamation connection made in good time.
Joe
-------
∂10-MAY-76 1234 REM via AMET
The old SPINDL program wrote a QQSPIN.RPG (?) file containing the names
and typed "DELETE @QQSPIN.RPG" into your input buffer upon exit.
Perhaps the new SPINDL (CRU3) should do the same, although there are
some difficulties making the list accurate if the user immediately deleted
some of the spindled subfiles before exiting, he wouldn't want their
originals deleted, the algorithm for "what files were spindled in this
session and weren't deleted from spindle" will take some work.
I guess I should make up a priority list for all things remaining to
clean up in CRU3.
I assume it leaves the delete command in the input buffer so that CR
is required to perform the deletion.
∂10-MAY-76 1243 JMC
I assume it leaves the delete command in the input buffer so that CR
is required to perform the deletion.
[REM - Yes, the old spindler did PTOSTR(0,"DEL @SPINDL.RPG"); and you
had to type the <cr>. Probably better for the new spindler would be
just to tell the user of the existance of the SPINDL.RPG file, such
as "use DEL @SPINDL.RPG to delete files you have just spindled",
alternately have another exit option, after the bubble question (or
lack thereof) ask "shall I now delete the files you have spindled
in this sessionπ
>" and have CRU3 delete them rather than create a delete list to pass
to COPY. - For now I'll just make a note in SPINDL.REM[UP,DOC]
clarifying that the user must delete them himself.]
∂09-MAY-76 2331 JMC
I think the directory of a spindle should allow a comment line for each file.
[REM - I think this falls in the realm of bells and whistles, unless you
mean that if a PRUNE.DAT exists then the data in it should be copied
into the spindle when spindling and PRUNE.DAT should be
updated whenever a file is unspindled. - Perhaps this should be on our
low-priority list, to be done after PRUNE gets some things fixed itself.]
∂09-MAY-76 1209 JMC
OK, do the reformat-abort one.
[REM - Ok, done in the version I put up earlier today (or last night, I forget).
If you Omit the file, it undoes the partially-written file and
proceeds with the next one in the original list of files to process.
The code is rather ad-hoc flakey so let me know if you find a bug.]
I don't really have time to look for bugs. Few people use PRUNE
these days, but some kind of comment is more necessary for crunched
files, because they can't be inspected without uncrunching them.
I think putting the DELETE command in the input buffer is the best
solution.
∂11-MAY-76 0937 FTP: Dertouzos at MIT-Multics
From: Dertouzos at MIT-Multics
Date: 05/11/76 1235-edt
To: JMC at SAIL
Please send me your critiques as they issue.
Thanks M.D.
OK, I can send about 5 today.
∂14-MAY-76 0420 FTP:GARY GOODMAN(A610GG01) at CMUB thesis update
From: GARY GOODMAN(A610GG01) at CMUB
Date: 14 May 1976 718 EDT
Subject: thesis update
To: MCCARTHY at SAIL
- - - -
John,
Due to an oversight, updates for section 8.1.6 "The Tasks"
(page 73 of Chapter 8) were not included in the copy you received.
This section now reads:
8.1.6 The Tasks
The thesis presented many results of analysis for several
tasks. The following is a brief summary of what was learned about
those tasks.
The Chess task[Reddy, et al.,1972; Baker, 1975; Lowerre,
1976] has an effective search space size of 23.31. Its equivalent
vocabulary size of 1.46 is the lowest of all the tasks studied. The
effective branching factor for this task is 1.09; also the lowest and
the same as for the IBM task.
The Lincoln Labs "extended" task [Forgie, et al., 1974] has
the largest search space size, 38.79. It is the most difficult task
by all measures except effective branching factor; Lizard and VPNS
having larger effective branching factors. The "basic" task, even
though its vocabulary contains 236 words, is of roughly the same
difficulty as the voice programming task when considering syntactic
and effective branching factors.
The IBM task [Tappert, 1975; Baker and Bahl, 1975] has an
effective search space size of 23.23. Its effective branching factor
is 1.09, the same as for the chess task. The syntactic branching
factor for this task is 7.32, lowest of all the tasks.
For the Lizard task[Lowerre, 1976], the search space size,
19.56, is the smallest of all the tasks. Its effective branching
factor of 1.46, however, is the largest of the languages having
syntactic constraints.
The voice programming task [Erman, 1974; Baker, 1975;
Lowerre, 1976] VP, has an effective search space size of 28.24. This
task has the largest syntactic branching factor of the medium sized
languages. VP with no syntax has the highest syntactic branching
factor.
The important contribution of this thesis is that it provides
a way to characterize the relative difficulties and accomplishments
of different speech understanding systems. Vocabulary size is not a
good measure of lexical complexity; some other measure of vocabulary
size, normalized for relative ambiguity would be better. The number
of production rules is not a useful measure of grammatical
complexity. In fact, quite the opposite may be true; more rules imply
more constraint. Some other measure, such as the average number of
alternatives at each choice point would be better. Investigators in
the area of speech understanding should reference their results to
some standard. This thesis presents some useful measures.
Gary
-------
∂15-MAY-76 1305 REM ARCHIVES AT MIT-DM
To: JMC, WD
According to TT@DM, their archives (devices AR0: etc.) are a software
simulation of a DSK: device in which the data resides on the disk but is
packed end-to-end to fill up gaps from the end of a file to the end of
a record. Thus it is almost exactly what we envision for the spindle
eventually. That does not mean, however that I favor changing our name
to "archive", because TENEX not ITS is the major export market for spindle.
It does mean that consulting with documentation on their implementation
might be useful in designing some aspects of our system support when we
get around to that phase of implementation. Since their system doesn't
involve crunching as far as I know, ours may end up superior when we
get it fully implemented.
I will return Wednesday night. I will be in Portland, Oregon till
Monday and in Boston at Fredkin's house thereafter.
∂16-MAY-76 0043 BZM Identification
To: JMC
CC: LES, BPM
∂16-MAY-76 0031 JMC
Why are you an unknown to our system?
Because Les flushed my name from whatever file it should be in. This
action is quite mysterious to me as well, for I was in it as recently
as last week. My project was supposedly for two and not three quarters,
however, so perhaps there was some auto-flush action. I am working
with Cordell's automatic programming group as my primary function,
though, so I seem self-legitimate enough! This mode will continue
through the middle of June, although I hope to convince the powers-
that-be to give me a guest account so that I can occassionally
access the system from Carnegie next year. As well as keeping a
finger in the PSI-pie, there are clearly some non-crunching things
for which this machine is better than the very DEC-like stuff at CMU.
.. Bruce Nelson
∂16-MAY-76 0339 REM SPINDL.REM
I have agreed with LES that it was a bad idea to put SPINDL.REM
in PUB format. It took forever to compile, there were many difficulties
in making it work, the input text became unreadable after all the
.SKIP etc. were inserted to make the output look like the input,
and worst of all your header caused it to compile 10 fonts (9 specified
by you and one extra by me) whereas only 3 were used (2 of yours and mine).
It now takes only 10-11 seconds to compile (with the simulator running).
I'll put it on [UP,DOC] after I see the output to be sure it works.
∂16-MAY-76 0349 REM P.S.
4 1/2 SECONDS COMPILE TIME WITHOUT SIMULATOR, SPINDL.REM + POX
∂17-MAY-76 1041 QIB New Telephone Number
I apologize for the inconvenience you had in getting me this weekend, at home, due
to my telephone number change. I had some problem callers and had it changed to an
unlisted number. So that it does not happen again, my new number is (408) 257-8115.
∂18-MAY-76 0622 REM SSORT (String-Sort)
To: TVR, LES, JMC
New SSORT.DMP[1,REM] doesn't have the crock of typing at itself,
and as of this morning has some new bugs fixed. Of course, report
any bugs you may find. Hope you like it...
The only new feature is the /Q switch (same as in COPY).
∂18-MAY-76 0810 FTP:SIGART(X180LE03) at CMUB[Lee] Weizenbaum review.
From: SIGART(X180LE03) at CMUB[Lee]
Date: 18 May 1976 1108 EDT
Subject: Weizenbaum review.
To: JMC at SU-AI
cc: SIGART
- - - -
As promised, this is a poke at you for a pointer to the "final"
version of the Weizenbaum review (the PUB source would be preferred).
(Or if you prefer, just MAIL it to SIGART@CMU-10B.)
Also, do you have Weizenbaum's response online? If so, that might
speed up getting it from him. In any case, I'll send him a copy of
your review and solicit his response.
thanx for your cooperation.
-------
∂18-MAY-76 1210 EJG via AMET System Downtime
To: TED
CC: JMC, LES, REG
I had been given the impression that people were going to
try hard not to schedule system downtime before 1700 Friday,
because of the Comprehensive programming problem this week.
So howcum TED has grabbed 1700-1859 on Tuesday, Wednesday,
and Thursday?
∂18-MAY-76 1427 EJG via AMET System downtime (revised)
To: TED
CC: JMC, LES, REG
LES tells me that TED needs the machine for fixing
a dead Ampex disc drive. I was unaware that there
were serious hardware troubles (more than usual).
Please consider my last question cancelled.
∂18-MAY-76 2104 FTP:CARL at MIT-AI
Date: 18 MAY 1976 2337-EST
From: CARL at MIT-AI
To: CARL at MIT-AI, jmc at SU-AI
John,
Please send me the artivle on relevance logic that you promised
to find for me.
Thanks,
Carl
-------
∂19-MAY-76 0807 PJ
VIDEO.CST[VID,PJ] has my current estimate for the cost of the prototype
video display. These are not final because the design is not finalized
but they are good estimates.
∂20-MAY-76 0653 REM
Yeah, I guess that bubble-question-on-exit is a winner, makes things so much more convenient...
∂20-MAY-76 1345 FTP:GLS at MIT-AI
Date: 20 MAY 1976 1109-EST
From: GLS at MIT-AI
Sent-by: GLS1 at MIT-AI
To: jmc at SU-AI
Please send ur small lisp compilers to GJS@MIT-AI. thanx.
-------
∂20-MAY-76 1430 REG
To: JBR, TED, LES, JMC, REG
For what it's worth, Fred Wilhelm of DEC called today and said that Per Hjerppe
is the person who knows what's going on about the microcode, etc. Fred also
said that the microcode should be "shipped within two weeks". Ho hum.
I need "Formal Philosophy" by Richard Montague - 1st from library and
then to buy a copy.
∂21-MAY-76 0725 100 : David Roode via AMET DEC-20 [LOTS]
Have you seen the Pres. Advisory Council report. Basically, they said
"Yes" enthusiastically. But although they recommend no ties to
SCIP, the action taking place [getting the 20 further "approved" by
the SCIP executive committee] indicates that SCIP will still be
between the 20 and the University, despite what was recommended.
∂22-MAY-76 0910 JRA lisp
hi:
it was nice to learn a bit more about lisp's history thursday night.
i too have been acosted by stoyan; he wants a copy of my lisp book.
do you think he'd be a good reviewer?
about your comments on multi-valued lisp: the previous week i'd been
working on just that! indeed i have a very neat approach to such
business. it's better than pop-2 or the scheme proposed recently
by friedman and wise. i'll write it up and send it to you.
your comment about parallelism in lisp implementations intrigued
me. Is someone else pursuing that avenue? my description of lisp
for such business is coming along very well. we'll be simulating
it this summer, and its true colors will become apparent.
john
I know of no other parallel project. I think Stoyan is reasonably
worthy of co-operation.
∂23-MAY-76 1216 WD Lock
I've thought about the lock, and come to the following conclusions.
Putting it on without the key is simple, and I can do that Tuesday. It can
be placed so as to leave room for a deadbolt should you wish to add one of those
in addition. The simplex lock will be stronger thatn the present arrangement,
because, although both are latches, it is at the back of the door where it is
harder to reach. It is not as resistant to force as a deadbolt, but,being a
latch, it is more likely not to be left undone.
OK, please do it.
∂24-MAY-76 1230 FTP:SIGART(X180LE03) at CMUB Weizenbaum review.
From: SIGART(X180LE03) at CMUB
Date: 24 May 1976 1528 EDT
Subject: Weizenbaum review.
To: JMC at SU-AI
cc: SIGART
- - - -
I plan on getting out the SIGART Newsletter by this Sunday. Any
chance of getting your review in the next day or so so that I can get
a response from Weizenbaum?
thanx, Lee
-------
The review is weizen.rev[pub,jmc]. Please tell Weizenbaum to
answer the latest version.
∂24-MAY-76 2230 FTP:SIGART(X180LE03) at CMUB Acknowledgment of receipt of Weizenbaum review.
From: SIGART(X180LE03) at CMUB
Date: 25 May 1976 129 EDT
Subject: Acknowledgment of receipt of Weizenbaum review.
To: JMC at SU-AI
cc: SIGART
- - - -
I have FTP'd a copy of WEIZEN.REV[JMC,PUB]@SU-AI (which is listed in
your directory with a creation date of 24-Apr-76). I have also taken
the liberty of getting a copy of WEIZEN.ANS (of 16-Apr). I will try
to get permission from Weizenbaum to use that (or some updated
version, if he prefers).
thanx,
Lee
-------
∂24-MAY-76 2306 RSC Cartwright's Thesis Orals
To: JMC, DCO
David Luckham would like to have my orals postponed until the week of July 5.
Is there any afternoon during that week which would be a convenient time for
you? -- Corky
I have no schedule.
∂24-MAY-76 2353 LES
∂18-MAY-76 1154 BPM
To: LES, REG
Resume-- Kenneth L. Harrenstien
22 Magazine St. #6 work: Rm 938, 545 Tech Sq.
Cambridge, MA 02139 Cambridge, MA 02139
(617) 547-0265 (617) 253-6765
single 6'0" 170 lbs. exc. health
Professional Computer applications, particularly in the areas of
Interest man-machine interaction, artificial intelligence, and aerospace.
Education MASSACHUSETTS INSTITUTE OF TECHNOLOGY CAMBRIDGE, MA
Candidate for Bachelor of Science degree in Electrical Engineering,
May, 1976. Concentration in computer science (systems, structure)
and psychology (linguistics, cognition).
Tau Beta Pi, Eta Kappa Nu, Sigma Xi.
Experience MASSACHUSETTS INSTITUTE OF TECHNOLOGY CAMBRIDGE, MA
1973-Present Artificial Intelligence Laboratory. Work on the A.I. PDP-6/10
installation doing various system programming and software
maintenance tasks. Some emphasis on real-time graphics; more
on programs interacting with other installations on the ARPA
computer network. Last two summers spent here.
Summer 1974 Designed and implemented general purpose inter-site user message
facility.
Summer 1973 Developed system user data base with programs to access and update.
COLORADO STATE UNIVERSITY FT COLLINS, CO
Summer 1972 Computer Center. Worked with staff systems programmer on routines to
test magnetic tape hardware/software, for export to other CDC
installations.
UNIVERSITY OF HAWAII HONOLULU, HI
Summer 1971 Developed a Gomoku-playing program and monitor for
use by a professor's programming classes.
Summer 1970 Library researcher for Engineering Dept. on alternative energy
sources for Hawaii.
Background Born in Kansas; raised in Ames, Iowa, Tucson, Arizona, and
& Interests Honolulu, Hawaii. Lost hearing at age 5, but speechread adequately.
Like outdoors; hockey player, archer, S/F fan, computer gamester;
long romance with Aero/Astro fields.
Short run-through:
My language background covers primarily Fortran, PL/1, and Lisp,
with one or two minors. I've done assembler programming for the
IBM 1130 (2 yrs), 360 (enough), and PDP-10 (past 2.5).
At MIT my relevant courses covered matters of operating systems,
computation structures, systems programming, modelling, etc. with
some digital hardware lab experience.
Current state:
For the past two years I have worked for variously
fun, money, or academic credit at the M.I.T. Artificial
Intelligence Lab; my main supervised project has been
the development of a sophisticated mail system which handles
local and ARPA network messages. In the course of
pusrsuing my interests I have become quite familiar
with the network, its user/server programs, and their associated
protocols. Apart from that, I have played a fair amount with
graphics, particularly real-time.
References
(professional)
Joan Tyson Comp. Center, CSU, Ft. Collins, CO (303)-491-5133
Thomas Knight MIT AI Lab, 545 Tech Sq, Cambridge, MA (617)-253-6765
Berthold Horn MIT AI Lab, 545 Tech Sq, Cambridge, MA (617)-253-6218
Brian Harvey Stanford AI Lab, S.U., Stanford,CA (415)-497-2462
(personal)
E.P. Goldenberg MIT LOGO Lab, 545 Tech Sq, Camb, MA (617)-253-6213
E. Feinler Aug. Res. Cen., SRI, Menlo Park, CA (415)-326-6200
∂25-MAY-76 1313 FTP:SIGART(X180LE03) at CMUB Yet more on the JW review.
From: SIGART(X180LE03) at CMUB
Date: 25 May 1976 1610 EDT
Subject: Yet more on the JW review.
To: JMC at SU-AI
cc: SIGART
- - - -
1) I assume that in the sentence ending "it is meaningless to ask
whether a machine can be more intelligent than a machine?", that the
last "machine" should be "human".
2) Do you want to include a reference for the White quote?
3) Weizenbaum has agreed to have his response printed. But he does
want to make some changes in it, mostly in the direction of sortening
some of the more "bare-fisted" aspects of it, and of incorporating
some of the later exchanges that he had with you. Do you want to
make any changes to your 24-Apr version along similar lines? If so,
we can accept them through Sunday (May 30).
thanx
-------
∂25-MAY-76 1319 FTP:Omalley at SRI-AI MEETING
Date: 25 MAY 1976 1317-PDT
From: Omalley at SRI-AI
Subject: MEETING
To: BAAIC-People:
GREETINGS
ANOTHER EXCITING MEETING OF THE BAY AREA CIRCLE WILL BE THIS
WEDNESDAY AFTERNOON. PANEL DISCUSSION ON PROBLEMS AND PROSPECTS OF AI.
TIME IS 4:00 PM (1600 FOR ARPA PEOPLE) IN ROOM 10 EVANS HALL
AT U. C. BERKELEY. ]EETING WILL BE FOLLOWED BY DINNER AT KING TSIN.
HIGH POINT OF THE DAY WILL BE A PARTY AT MY HOUSE -- 2910 ASHBY,
BERKELEY -- AFTER THE DINNER, WHICH MEANS AT ABOUT 8:00 PM.
EVEN THOSE WHO HATE KING TSIN AND ARE BORED WITH ALL PANELS
ARE WELCOME TO THE PARTY.
RSVP
MIKE 0'MALLEY
OMALLEY AT SRI-AI
-------
∂25-MAY-76 1344 100 : DREW HI THERE
PROFF. MCCARTHY
I WOULD LIKE TO TALK TO YOU ABOUT ALL THIS LOTS
MATERIAL AT THE NEXT AVAILABLE OPPORTUNITY.
EVERYTHING SEEMS TO BE FINE, BUT I WOULD VERY MUCH
APPRECIATE IT IF YOU COULD HELP ME KEEP AN EYE ON FRANKLIN....
DREW LANZA
∂25-MAY-76 1501 FTP:Ray Reiter Theorem Proving Meeting
Date: 25 May 1976 1753-EDT
Sender: WEBBER at BBN-TENEXD
Subject: Theorem Proving Meeting
From: Ray Reiter
To: jmc at SU-AI
Message-ID: <[BBN-TENEXD]25-May-76 17:53:20-EDT.WEBBER>
Message received -- your name has been added to the mailing list.
-------
∂26-MAY-76 0625 100 : REM via AMET What I have been up to
After making preliminary default codes for .FAI and .LSP, I listed the
two codes (the .LSP one wasn't in the spooling box when I came up to
pick it up) and spent several days hand-purging nodes (left-contexts)
which didn't seem to contribute much to the effectiveness of the code,
and at great difficulty found enough compute time to perform a simple
edit of the list of left-contexts. Then I ran the new list of left
contexts and got a code that is only one percentage point worse but
uses only about 60% to 70% as much core (rough guess). That pretty much
burned me out, combined with not getting a paycheck after driving all
the way up there to get it. The hassle of poor system response due
to nonpaging system with no swapping disk destroys continuity of
thought when attempting to get any work done. I think we had better
forget about any bells and whistles in SPINDL, and any further research
on this system, finish up any essentials, and try to phase over to
TENEX as soon as we find a customer. Note that IMSSS doesn't count
because of the lack of FTP and NETMAIL. Perhaps SRI or SUMEX or ISI
or whereever... Unless you can think of aything, the essentials are:
(1) Binary-mode SPINDL, (2) Type portion of file on TTY, (3) Publish
a paper on methods and results in Comm ACM or similar journal, (4) Fix
the bug about incorrect c.r. reported when crunching. Only (4) is
truly essential.
Don't get discouraged. The KL-10 without cache will be up in a few
days (1.5 times the crunchies) and with cache (5 times) in a week
or so.
∂26-MAY-76 2205 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
Date: 26 MAY 1976 2040-EST
From: REM at MIT-MC
Subject: Preliminary draft of abstract etc. of article on crunching
To: REM at SU-AI, JMC at SU-AI
History-Tree Text Compression
by Robert Elton Maas
Stanford University Artificial Intelligence Project
ABSTRACT
A method is described for using left-context information
to compress text (ASCII) data. Text is modeled as a Markov process
in which the state (a function of the 3 or fewer preceeding characters)
is used to assign propabilities to the next character to appear in the file.
When this method is applied to typical free-format English-language
documents and essays existing as disk files on the PDP-10 based system
at Stanford A.I., a 2-to-1 reduction of storage is achieved.
KEYWORDS: DATA-COMPRESSION, ASCII, TEXT, ENGLISH, ENTROPY, MARKOV-PROCESS,
PDP-10.
-------
∂26-MAY-76 2218 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
Date: 26 MAY 1976 2040-EST
From: REM at MIT-MC
Subject: Preliminary draft of abstract etc. of article on crunching
To: REM at SU-AI, JMC at SU-AI
History-Tree Text Compression
by Robert Elton Maas
Stanford University Artificial Intelligence Project
ABSTRACT
A method is described for using left-context information
to compress text (ASCII) data. Text is modeled as a Markov process
in which the state (a function of the 3 or fewer preceeding characters)
is used to assign propabilities to the next character to appear in the file.
When this method is applied to typical free-format English-language
documents and essays existing as disk files on the PDP-10 based system
at Stanford A.I., a 2-to-1 reduction of storage is achieved.
KEYWORDS: DATA-COMPRESSION, ASCII, TEXT, ENGLISH, ENTROPY, MARKOV-PROCESS,
PDP-10.
-------
∂26-MAY-76 2242 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
Date: 26 MAY 1976 2040-EST
From: REM at MIT-MC
Subject: Preliminary draft of abstract etc. of article on crunching
To: REM at SU-AI, JMC at SU-AI
History-Tree Text Compression
by Robert Elton Maas
Stanford University Artificial Intelligence Project
ABSTRACT
A method is described for using left-context information
to compress text (ASCII) data. Text is modeled as a Markov process
in which the state (a function of the 3 or fewer preceeding characters)
is used to assign propabilities to the next character to appear in the file.
When this method is applied to typical free-format English-language
documents and essays existing as disk files on the PDP-10 based system
at Stanford A.I., a 2-to-1 reduction of storage is achieved.
KEYWORDS: DATA-COMPRESSION, ASCII, TEXT, ENGLISH, ENTROPY, MARKOV-PROCESS,
PDP-10.
-------
∂26-MAY-76 2230 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
Date: 26 MAY 1976 2040-EST
From: REM at MIT-MC
Subject: Preliminary draft of abstract etc. of article on crunching
To: REM at SU-AI, JMC at SU-AI
History-Tree Text Compression
by Robert Elton Maas
Stanford University Artificial Intelligence Project
ABSTRACT
A method is described for using left-context information
to compress text (ASCII) data. Text is modeled as a Markov process
in which the state (a function of the 3 or fewer preceeding characters)
is used to assign propabilities to the next character to appear in the file.
When this method is applied to typical free-format English-language
documents and essays existing as disk files on the PDP-10 based system
at Stanford A.I., a 2-to-1 reduction of storage is achieved.
KEYWORDS: DATA-COMPRESSION, ASCII, TEXT, ENGLISH, ENTROPY, MARKOV-PROCESS,
PDP-10.
-------
∂26-MAY-76 2252 FTP:REM at MIT-MC Preliminary draft of abstract etc. of article on crunching
Date: 26 MAY 1976 2040-EST
From: REM at MIT-MC
Subject: Preliminary draft of abstract etc. of article on crunching
To: REM at SU-AI, JMC at SU-AI
History-Tree Text Compression
by Robert Elton Maas
Stanford University Artificial Intelligence Project
ABSTRACT
A method is described for using left-context information
to compress text (ASCII) data. Text is modeled as a Markov process
in which the state (a function of the 3 or fewer preceeding characters)
is used to assign propabilities to the next character to appear in the file.
When this method is applied to typical free-format English-language
documents and essays existing as disk files on the PDP-10 based system
at Stanford A.I., a 2-to-1 reduction of storage is achieved.
KEYWORDS: DATA-COMPRESSION, ASCII, TEXT, ENGLISH, ENTROPY, MARKOV-PROCESS,
PDP-10.
-------
∂26-MAY-76 2311 FTP:SIGART(X180LE03) at CMUB[Lee Erman] Weizenbaum's Response.
From: SIGART(X180LE03) at CMUB[Lee Erman]
Date: 27 May 1976 203 EDT
Subject: Weizenbaum's Response.
To: JMC at SU-AI
- - - -
Enclosed below is Weizenbaum's response. I am hoping to wrap up this
issue of the Newsletter by noon on Friday, so if you have any changes
to yours, please let me know as soon as possible.
thank you for your help and participation,
Lee
encl:
The reader should know that Prof. McCarthy wrote and published (In
'John McCarthy's Electronic Magazine', i.e. in su-ai's directory
[pub,jmc]) his first review of my book on April 18, 1976. The
response reproduced below is to that review. On April 24th Prof.
McCarthy published a revision of his original review and that, as far
as I can tell, is the one reproduced here. Finally, Prof. McCarthy
also responded to my response. That last communication was sent to
me and to a few other people at MIT and, I suppose, elsewhere. I was
notified of SIGART's intention to publish McCarthy's review (i.e.,
the second version) only a few days before SIGART's deadline and have
therefore not had an opportunity to revise my response to fit
McCarthy's revision. Prof. McCarthy and I have, however, had some
correspondence on our various points of contact. A few remarks on
that correspondence are in order. I append them to the end of the
original version of my response.
A RESPONSE TO JOHN McCARTHY by JOSEPH WEIZENBAUM.
Whatever the merit of John McCarthy's review of Computer Power and
Human Reason may be, it is all but undone by his repeated assertion
that the positions taken in the book are derived from a "new left"
political ideology. Not long ago the terms "pinko" or "commie"
served the function McCarthy assigns to "new left" in his review. I
would have thought that people might have learned something from the
events of the tragic decades the United States has just passed
through and that all participants in scholarly debates had by now
renounced argument by irrelevant political association.
McCarthy's warning to "the outside observer" that the book is
motivated by a struggle over academic appointments, tenure decisions,
etc. going on within M.I.T. is bizarre and absurd.
I am disturbed by John McCarthy's misreading of my book's "main
points." The books actual main point, to the extent that there is one
main point, is that no single way of seeing the world, whether it be
that of the computer metaphor, of science, of religion, of some
political dogma, or of whatever, is sufficient to yield an
understanding of the world worthy of the human potential to
understand.
McCarthy sees the book making the following main points:
1) COMPUTERS CANNOT BE MADE TO REASON AS POWERFULLY AS HUMANS
I wrote: "...I see no way to put a bound on the degree of
intelligence [a computer] could, at least in principle, attain." (p.
210) I then go on to argue that a computer's socialization, that is,
its aquisition of knowledge from its experience with the world, must
necessarily be different from the socialization of human beings. A
computer's intelligence must therefore be always alien to human
intelligence with respect to a certain range of human affairs. (p.
213) Nowhere do I limit the computer's "reasoning power." The whole
book is, however, an attack on the dogmatic coupling of reason to
power. This coupling is so much part of the Zeitgeist that
single-mindedly committed technological enthusiasts simply cannot
conceive of a discussion of reason -- whether by computers or not --
that is not at the same time centered on questions of power.
McCarthy's gratuitous projection of his own preoccupations unto me in
the form of his attribution to me of this "main point" is further
evidence for that.
2) THERE ARE TASKS THAT COMPUTERS SHOULD NOT BE PROGRAMMED TO DO.
Yes, that is genuinely a main point of the book. And McCarthy is
right in observing that task that should not be done at all should
not be done by computers either. McCarthy and I agree that
psychotherapy should under some circumstances be practiced. I am
opposed to machine administered psychotherapy and McCarthy cannot see
what objections there might be to it (other than those that arise
from "new left" motivations) if it were to "cure" people. Prefontal
lobotomy "cures" certain mental disorders. But at what price to the
patient and, I would add, to the surgeon as well? I believe that
machine administered psychotherapy would induce an image of what it
means to be human that would be prohibitively costly to human
culture. One may disagree with this belief. But one would first
have to understand it and to take it into account.
Elsewhere I say that an individual is dehumanized whenever he is
treated as less than a whole person. The relevance of that to the
present discussion can be seen if one recalls how inhumanely many
surgeons treat their patients and people generally. They have, after
many years of seeing their patients mainly as objects to be cut and
sewed, come to see them as nothing more than objects. Many surgeons
eventually see everyone, most importantly themselves, in this narrow
way. Similar remarks apply to other professions. It is of course
necessary for all of us to adobt an effectively clinical attitude
toward people we deal with in a large variety of situations. The
surgeon could not actually cut into living flesh were he not able to
impose a psychological distance between himself and his patient while
actually wielding the scalpel. But somewhere in his inner being he
should hold on to his perception of his patient as a whole person.
Even more importantly, the patient must never be led into a situation
in which he is forced, or even merely encouraged, to regard himself
as a mere object. My fear is that computer administered
psychotherapy necessarily induces just this kind of self-image in the
patients who would be subject to it. That, basically, is my
objection to it. I cannot see how such a system could "cure" people
in any reasonable sense of the word "cure", that is, in a
sufficiently encompassing interpretation of that word.
3) SCIENCE HAS LED PEOPLE TO A WRONG VIEW OF THE WORLD AND LIFE.
There is no "correct" or "wrong" view of life to which science or
anything else can lead. The point McCarthy here misconstrues is that
science, or any other system of thought, leads to an impoverished
view of the world and of life when it or any system is taken to be
the only legitimate perspective on the world and on life.
4) SCIENCE IS NOT THE SOLE OR EVEN THE MAIN SOURCE OF KNOWLEDGE.
How reliable would McCarthy say is his knowledge that his children
are biologically his children or that the person he knows as his
father is his biological father? Is science his source of such
knowledge? What proportion of the truly important actions McCarthy
has taken in the course of his adult life were predicated solely or
even mainly on knowledge he validated by appeals to science? Did the
ancients have reliable knowledge? Or have we had reliable knowledge
only since the founding of the British Royal Society -- or that of
the Stanford AI Lab -- or not yet at all?
5) CERTAIN PEOPLE AND INSTITUTIONS ARE BAD.
I know of very few people I would call "bad" -- Hitler and Himmler
are examples. I think some of the people McCarthy lists are often
wrong and sometimes behave irresponsibly especially when they speak
to and write for lay audiences. There appears to be wide agreement
on that within the AI community itself. I think the views expressed
by some of the people mentioned are dangerous. These views should be
discussed, not suppressed. My book contributes to the required
discussion. Some people will surely find my views wrong and perhaps
even dangerous. If they think my views worthy of more than contempt,
they should discuss them. Does McCarthy think I am "bad?" I don't
believe so. Why then should he believe I think the people he
mentions are bad?
The quotation marks on Department of "Defense" are, I would guess,
entirely appropriate in the eyes of most of the people of the world
-- especially in the eyes of many who have read Orwell. If the
emperor wears no clothes, we should say he wears no clothes.
Other remarks:
I do not say and I do not believe that "if the problem hasn't been
solved in twenty years, we should give up." I say (p. l98) "... it
would be wrong ... to make impossibility arguments about what
computers can do entirely on the grounds of our present ignorance."
That is quite the opposite of what McCarthy charges me with saying.
I do not say or imply that "the Defense Department supports speech
recognition research in order to be able to snoop on telephone
conversations." Attributing this view to me is, in McCarthy's words,
"biased, baseless, false, and [seemingly] motivated by malice." I
wrote: "This project then represents, in the eyes of its chief
sponsor, a long step toward a fully automated battlefield." I then
state my opinion that, should we get speech recognition, large
organizations such as the government would use it for snooping, etc.
I believe that. My belief is buttressed by the revelations (N.Y.T.,
August 3l l975) that the "NSA eavesdrops on virtually all cable,
Telex and other non-telephone communications leaving the U.S. and
uses computers to sort and obtain intelligence from the contents ..."
(emphasis mine). Clearly the exclusion of telephone communications
from this operation is a consequence of only technical limitations
that would be removed if we had automatic speech recognition systems.
The reference cited in note 9, page 286, is anonymous because the
person in question granted me permission to quote from his internal
memorandum on the condition that I not cite his name. It is not nice
of John McCarthy to press me to violate my word.
I do not "idealize the life of primitive man." It is a cheap shot
often practiced by technological enthusiasts to charge anyone who
mentions a loss entailed by man's commitment to technology (and there
surely have been and are losses) with advocating a return to
pretechnological times. Every modern writer I know of knows that
there cannot have been a pretechnological time in the history of what
we would call man, and that history cannot be reversed. But it is
important that we recognize and understand the costs associated with
our current commitment to technology and that we seek for ways to
reduce the costs we deem too high. There is nothing
anti-technological, anti-scientific, or anti-intellectual in that.
McCarthy suggests that my statement "Those who know who and what they
are do not need to ask what they should do" is "menacing" in that he
believes it to require a priesthood to apply it to a particular case.
The statement appears in a context (p. 273) in which I had just
alluded to the fact that people are constantly asking experts what
they should do. I don't believe people need "expert" guidance on
moral questions. The statement, on its very face, argues that no
priesthood is ever necessary to tell people what they must do. I
find McCarthy's exact opposite analysis extremely puzzling.
My assertion that "An individual is dehumanized whenever he is
treated as less than a whole person" is simply a statement of fact.
It is incomprehensible to me that shame or guilt fall on me because
McCarthy believes similar statements to be part of the catechism of
the "encounter group movement," about which, by the way, I know next
to nothing.
Does John McCarthy have a logical calculus within which he has proved
that any idea held by the new left or the encounter group movement or
by Mumford, Roszak, or Ellul is wrong, a menace, and certain to be
used as part of the arsenal of "priests [who] quickly crystallize
around any potential center of power?" (Here again we see evidence of
McCarthy's preoccupation with power.)
Finally, McCarthy asserts "Philosophical and moral thinking has never
found a model of man that relates human beliefs to the physical world
in a plausible way." Only someone who has mastered the entire
philosophical and moral literature could have the authority to say
that. What truly God-like humility! The distance that separates
John McCarthy from Joseph Weizenbaum is truly measured by the
challenges these two hurl at one another: McCarthy defies Weizenbaum
to "Show me a way to knowledge besides science!" And Weizenbaum
responds: "Can there be a way toward an authentic model of man that
does not include and ultimately rest on philosophical and moral
thinking?"
No wonder we talk past one another.
!.subsec(COMMENTS OCCASIONED BY McCARTHY'S REVISION);
By the time McCarthy undertook to revise his first draft, he had in
hand my response. He saw, in particular, that I had not said that
the DOD sponsored speech recognition work "in order to" do snooping.
Indeed, I corresponded with him on that point and obtained his
agreement that he would delete that wrong and harmful assertion from
any published version of the review. That he has chose to let it
stand, having agreed that it is false, distresses me greatly. One
should not knowingly publish falsehoods. His error is especially
grievous in that he characterizes the false statement which he
attributes to me as being "biased, baseless, false and [seemingly]
motivated by political malice." I don't know how to put a charitable
interpretation on this behavior.
I admit to an error with respect to the quote John McCarthy could not
find on the BBC tape from which I had said I aquired it. On that
tape Prof. McCarthy is heard only to say "... the general
formulation of what the world is like has not been accomplished ...
part of this is due to a defect in our current systems of
mathematical logic." My attribution should have been to a discussion
following a showing of the BBC tape. I apologize for the error.
Even though my original citation made it clear that the statement in
question was made in the context of a debate, I regret that I may
have lent undue wheight to what was in fact spoken in the course of a
fast paced argument and which, given more tranquil circumstances,
Prof. McCarthy would surely have qualified.
I confess that I am unable to give reasons that explain why I find
the idea of disembodying the brain and the eyes of a cat, keeping
them artificially "alive" and using them as visual receptors of a
computing system obscene. If simple human decency allows nothing to
be taken for granted, then I retreat before the onslought of my
critics. There are some things I, however, do take for granted. One
of these is that a machine of the sort here in question (and that my
critics can apparently view with equanimity) would appear to me to be
a monster from which I would turn with disgust and revulsion. I have
in mind also the teaching urged on us by such leaders of the AI
community as H.A.Simon that there is nothing unique about the human
species, that, in fact, the embrace of the "illusion" of human
uniqueness amounts to a kind of "species prejudice" and is unworthy
of enlightened intellectuals. If we find nothing abhorrent in the
use of artificially sustained disembodied animal brains as computing
components and if there is nothing that uniquely distinguishes the
human species from animal species, then ....!?
With respect to the question "what do judges know that we cannot tell
computers," let me make the following analogy: Consider the question
"what has art contributed to human welfare?" Were that question asked
in order to define a theme for a curriculum of study, then I would
applaud its asking. If, however, the question is asked rhetorically,
i.e., were it really an assertion that art had contributed nothing to
human welfare, and if, moreover, that assertion were made as part of
an argument that artists ought to be declared social parasites, then
I would find the question obscene. I agree, certainly, that there is
a form of the question "what do judges know that we cannot tell
computers" that raises substantive questions about the limits of
computation. These are the very questions to which the bulk of my
book is addressed. But the way John McCarthy asked me that question
at the Stanford AI conference in 1973 clearly indicated that the
question was not a question at all but an assertion that all human
knowledge is reducible to computable form. This diminishing of the
human being I find offensive. I apologize for not having made this
point sufficiently clear in the book.
I stand by the sentence "Scientists who continue to prattle on about
'knowledge for its own sake' in order to exploit that slogan for
their self serving ends have detached science and knowledge from any
contact with the real world". It has the ominous meaning McCarthy
assigns to it only if the qualifying phrase "in order to exploit it
for their self serving ends" is ignored. I suspect that exactly
those "scientists" whom the shoe fits will ignore that phrase and be
angered by the sentence.
Prof. McCarthy hears me saying that industrial results are necessary
to validate science. But the sentence he quotes is the last sentence
of a paragraph that, taken as a whole, lends an entirely different
meaning to that sentence than that given it by McCarthy. The
paragraph speaks of the results AI promised for problem solving and
cognitive psychology but which it has not yet delivered and of the
results -- also not yet attained -- promised for the "visible
future". It then concludes with the remark that few results have
found their way into industry. This, clearly, has nothing to do with
"validating" science.
McCarthy's deepest misunderstanding is revealed by his insistent and
petulant complaints that the big boys in linguistics are not watching
the games played by the computer linguists, e.g. Schank and Winograd,
in their playpens. He insists that Chomsky's and Schank's goals are
"similar" but that Schank is closer to success. But that is totally
and absolutely wrong. Chomsky's goal is to discover the constraints
that apply to the structure of every human language and then to infer
from those constraints what a mind that can aquire a human language
might be like. He is like an explorer who, having come accross a
LISP machine, attempts to discover its logical structure, its innate
properties, by examining the constraints on the linguistic utterances
it is able to accept and produce. The computer linguists may believe
themselves to be attending to the same problem "ultimatly". But
their research program, which consists mainly of getting the machine
to "understand" more and more diffcult linguistic utterances, cannot
lead them to the ultimate goal -- however impressive their
intermediate "results" -- since it simply does not ask questions
related to that goal. In the sense here itended, they are like
programming language designers who continue to add new "features" to
their languages, truly enhancing the expressive powers of their
languages, to be sure, and mistakenly believe themselves to be coming
closer to a theory of computation. Notice that McCarthy commends
"the performance of [Schank's] student's programs [as] compared with
reality", not any principles arrived at by Schank that are in any way
competitive with, let alone stronger in explanatory power than,
Chomsky's. That someone like Prof. McCarthy, who has made
impressive contributions to the theory of computation, should not
understand this is puzzling.
-------
∂27-MAY-76 0153 KRD
To: LES, JMC
If running the Lab's PUB to produce xgp-output is possible on SUMEX
I'll be happy to do it that way.
∂29-MAY-76 0722 WD
Your knife is in your center desk drawer.
∂31-MAY-76 1409 FTP:SIGART(X180LE03) at CMUB[Lee Erman] A question about the Weizenbaum review (with candidate correction).
From: SIGART(X180LE03) at CMUB[Lee Erman]
Date: 31 May 1976 1707 EDT
Subject: A question about the Weizenbaum review (with candidate correction).
To: JMC at SU-AI
- - - -
There is a sentence in your review which we are unclear about. In
the section called "Will artificial intellegence be good or bad?":
"I wouldn't like that, but once man is confronted by the
actuality of full AI, they may find our opinion as relevant
to them as we would find the opinion of Pithecanthropus
about whether subsequent evolution took the right course."
My interpretation is "... [he] may find our opinion as relevant
to [him] as ...". Is that right? Or would you like to
rewrite it differently?
We are in the last throes and expect to mail it out tomorrow for
printing.
Sorry to bother you again and thanks for your help.
-------
Your interpretation is correct.
∂01-JUN-76 0345 RSC at TTY72 0345
My oral examination has definitely been cancelled. I will try to arrange a new
time within the next week. -- Corky Cartwright
∂02-JUN-76 1049 FTP:JOSEPH at MIT-MC
Date: 2 JUN 1976 1343-EST
From: JOSEPH at MIT-MC
To: jmc at SU-AI
Have not seen NATURE review. Will look up review and
send letter to the editor making the same clarification that I
sent to DATAMATION and that is incorporated in my response to
your review that is to appear (or is now extant) in the
SIGART bulletin. That, by the way, is now on-line at CMU. You
may wish to look at it and comment. My respinse has a new
addendum.
I will wait a day or so before writing to NATURE.
-------