COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	.require "memo.pub[let,jmc]" source
C00008 ENDMK
C⊗;
.require "memo.pub[let,jmc]" source;
.cb COMMENTS ON DENICOFF PAPER


	Marvin Denicoff does a real service in attempting a history of
computing, and his topics are well chosen.  However, it is one of the
first attempts and naturally encounters disagreement from other
participants and observers.  Here are mine:
.item←0

	#. On the whole the fifties were more optimistic years than
the seventies and not only in computer science.  However, the picture
drawn by Denicoff doesn't do justice to the varied opinions extant
at that time about the probabilities of success of various endeavors
in given times.  When asked for a prediction of how soon a research
or development goal will be realized, a scientist is bound to give
an optimistic answer if he takes into account only difficulties that
he recognizes.  Therefore, he should cautiously answer that if the
difficulties he sees are all there are, it will take ⊗n years.

	#. Denicoff's history of time-sharing assigns to little
credit to Corbato (who initiated CTSS) and me and too much to Strachey
(read his 1959 paper), Teager, and Fano.  Institutionally, it neglects the
IBM and NSF support of early time-sharing at M.I.T. and N.I.H. support of
time-sharing at BBN.  Administratively, Philip Morse and J.C.R. Licklider
also deserve credit.
Fredkin deserves credit for recognizing that time-sharing was feasible on
a computer as small as the PDP-1 and devising most of the hardware
modifications required to make it feasible.
The importance of MULTICS is exaggerated; it never produced cost-effective
time-sharing and neither did the IBM TSS effort.  Denicoff inexplicably
totally ignores the PDP-6 and PDP-10 computers which initiated cost-effective
time-sharing and whose descendants are still the most commonly used
time-sharing machines.  Their success compared to MULTICS and TSS is
due to concentrating on real problems rather than the pseudo-problems
such as levels of access and extreme elaboration of paging and segmentation.

	#. Artificial intelligence.  The Dartmouth conference was in
1956.  The history of artificial intelligence is even harder to write
than that of time-sharing.  Time-sharing is successful, and careful
interviewing and reading will trace the successful ideas to their
sources.  Artificial intelligence has not yet achieved enough of
its goals to make it uncontroversial which activities are contributions
and which are blind alleys.  Moreover, the theoretical foundation
of the subject is still so undeveloped that it can't even be
definitively said what the basic ideas are.  E.g. is natural language
understanding fundamental or a side issue?

	#. Software development.  I find this discussion
mixing issues of management, language and data
structure design, and proofs of correctness rather murky.

	#. Large file systems.  One suspects the author of
overemphasizing activities sponsored by ONR.  The pacing
element in large on-line data systems has been the cost
of disk file.  Now that disk files are cheap, there can
be movement away from monthly batch processing of keypunched
data to immediate processing of data entered at terminals
as soon as it is received with immediate feedback to the
person who enters the data.  This change is important for
social reasons, because much of the rigidity and mystery
of computer systems has been caused by the infeasibility
of interrogating and changing data relevant to an individual
in the middle of a large tape library.  The problem may
be important enough to almost justify outlawing batch
processing.