: Defense Science Board

Task Force on Supercomputer Applications

Meeting Minutes

Date/Place

9:00 a.m. - 6:00 p.m., 138 May 1983
Stanford University, California

Attendees

Task Force Members: DSB Members:

Dr. Joshua Lederberg, Chairman Dr. Ivan Sutherland
CDR Ronald B. Ohlander, Executive Secretary

Dr. Dr. William J. Perry DSB Staff:

Lt. Gen. Philip D. Shutler, USMC (ret) LCDR Ralph Chatham

invited Guest Participants:
Dr. Robert E. Kahn

Dr. Eliiott C. Levinthal
Gordon Bell

Business
The second meeting of the DSB Task Force on Supercomputer Applications was opened by the
Chairman, Dr. Joshua Lederberg. The primary objectives of this meeting were to further refine task

force requirements and gain familiarity with military defense systems requirements and artificia!
intelligence (Al) technology.

Dr. Kahn of DARPA then presented another overview of the Supercomputer program for the benefit of
those who had not attended the initial DSB Task Force meeting. During the presentation, the issue of
training sufficient numbers of people with the needed skills was raised. It was anticipated that the
Supercomputer Program would place an unprecedented demand on skills that are in relatively short
supply, i.e., expertise in Al and design of specialized machine architecture and software. The
consensus of opinion was that universities would play the key role in generating this pool of talent.
Universities would also be critical to transferring existing technology to industry and developing new
milestones in technology that would be necessary for the program to succeed. The DARPA program
overview also emphasized that the focus was on the development of parallel symbolic computational
machines, rather than the traditional numeric processors. This approach differs significantly from the
main thrust of the major computer manufacturers who are pursuing more conventional technology
(although account must be taken of efforts like CDC’s AFP).

The task force then received a series of three briefings on military applications and Al technology.
The briefings were scheduled to inform members of the task force on the application needs of DoD
and the state-of-the-art in technology that has the potential of contributing to some of the more
pressing application needs. Copies of briefing charts are appended to these minutes.

The first briefing was presented by Hughes Corporation and addressed a research project to apply
the latest image understanding technology to a port monitoring task. The Hughes vision system was
based on the ACRONYM image understanding system developed by Stanford University under
DARPA sponsorship. ACRONYM was a useful starting point but had to be modified substantially to
meet the needs of a real photointerpretation task with its attendant poor imagery. The entire
interpretation evolution took about 10 hours. In order to obtain real-time performance a speedup of 3
to 4 orders of magnitude in computationa! power would be required. It was postulated that, although
computer vision tasks are exceedingly difficult, image understanding technology is ripe for
exploitation. It is not clear, however, whether ACRONYM is subtle enough to be of real value in
exploitations without some expansion of its rule base.

The second presentation was on simulation and wargaming: important DoD application areas with
demanding computational! requirements. Currently, the work in these areas is mainly software limited.
As algorithms are developed to address such problems as man-machine interfaces, greater
complexity in models, etc., and applications justify large investment in rule knowledge bases, the
need for much greater computational power will manifest itself. This is expected to happen in the
next couple of years. The estimated additional computational power that would be needed for future
applications ranged from 3 to 6 orders of magnitude over a Cray-1, depending on the specific
application. ~

The third presentation was in the area of knowledge-based systems. The briefing gave a resume of
the state-of-the-art of this technology and forecast requirements for future, complex systems.
Estimates of needed computational resources were in the range of an increase of 3 orders of
magnitude.

On the basis of the day’s discussions, Dr. Perry proposed 3 potential areas of application:
autonomous weapons, signal understanding, and simulation. Autonomous weapons would involve
real-time vision and intelligence for planning and mission execution. Signal understanding
applications would be oriented around intelligence exploitation and would have less severe real-time
requirements. Simulation would include both wargaming and airplane modeling.

The last part of the meeting involved a discussion of future meetings and topics to be addressed.
Suggested topics for future briefings included signal understanding, guided weapons, and
information search and retrieval.

 

Certified: -
Hoshua Lederberg,
/ Chairman