perm filename GOALS.LIC[CUR,JMC] blob
sn#134752 filedate 1974-12-09 generic text, type C, neo UTF8
COMMENT ā VALID 00002 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00002 00002 \\M0BASL30\M1BASI30\M2BASB30\M3NGR40\F3
C00022 ENDMK
Cā;
\\M0BASL30;\M1BASI30;\M2BASB30;\M3NGR40;\F3
\CGOALS OF THE STANFORD ARTIFICIAL INTELLIGENCE LABORATORY
\F0\J The Stanford Artificial Intelligence Laboratory has existed since
July 1963 and has had the general goal of pursuing research in artificial
intellgence. It has always pursued some related fields, and in 1970
mathematical theory of computation as aimed at replacing debugging by
formal proofs of program correctness was added to the work statement.
During its existence, the laboratory has worked on the following
topics:
1. Representation of information for AI. This has been the main
work of J. McCarthy and has involved one or two other people from time to
time.
2. Heuristics. This has included theorem proving and game playing.
At present there is no game playing work except for one student thinking
about chess pattern recognition. Luckham and Allen are continuing their
resolution based theorem proving.
3. Vision. This has been a major activity all through the history
of the Lab. The current leaders are Binford, Quam and Baumgart. Baumgart's
approach to vision is through graphics, and this has led to considerable
achievement in this area.
4. Speech. This work started in 1963 and was the first work on speech
that used a computer equipped only with and A-D converter and direct analysis
of the sound wave. This work is currently dormant for lack of support.
5. Manipulation. Connecting vision with manipulation was originally
seen as the best way to see the real-life vision problems. We have been
working on automatic assembly for a long time, and recently we have gotten
some NSF support for it.
6. Natural language. At different times, different people have
adopted different approaches. Roger Schank and Yorick Wilks took their
respective approaches away with them, and now Terry Winograd holds this
fort.
7. Automatic programming. Cordell Green and David Luckham are
pursuing distinct but complementary approaches. Green's approach emphasizes
connections with natural language and Luckham's connections with mathematical
logic and theorem proving.
8. Mathematical theory of computation. This work contains a theoretical
part and a practical part. The theoretical part consists of devising formalisms
and means of inference whereby facts about computer programs can be proved.
In the past, Luckham, McCarthy, Manna, Milner, Igarashi, London and numerous
students have worked and published papers in this area at the Stanford AI Lab.
At present Gordon, McCarthy, Luckham, and Weyhrauch are active in the theoretical
area. The practical side includes proof-checkers and proofs that actual
large programs meet their specifications. Weyhrauch is in charge of this work
now with help from Glassmire, Poole, Thomas, and Arnold.
Besides the main research areas of the Lab, we have made contributions
in time-sharing system design, in computer design, and in programs
using displays. This was probably the first laboratory to equip all offices
with display terminals, and a particularly economical way of installing
displays was required. Further improvement of the computer facilities
is necessary for our work and will require further hardware and software
developments in the system area.
The subsequent sections of the briefing document describe the plans
for the next period as laid out by the group leaders in charge of the various
areas. It is not clear that there will be resources to carry out all these
plans, because they have not yet been compared and priorities assigned. Our
formal proposal will embody much greater reconciliation of the plans of the
subgroups.
At present our computer facility is seriously overloaded and both
Green and Winograd have been getting time for their groups from other
organizations on the ARPA net. In part this has come from their desire to
use Interlisp which is not available here, but mainly it is because the
large amount of compute time that the use of Interlisp requires is simply
not available. Fortunately, in recognition of our contributions to the
design of KL-10 computer, Digital Equipment Corporation has promised to
give us a KL-10 processor which will hopefully be five times as fast as
our present machine. However, getting the advantage of the speed of the
KL-10 will require other software and hardware improvements and will cost
more money. Details of our requirements for this will be included in
our forthcoming proposal.\.