SUMEX

STANFORD UNIVERSITY
MEDICAL EXPERIMENTAL COMPUTER RESOURCE

RR - 00785

COMPETING RENEWAL APPLICATION

Submitted to

BIOTECHNOLOGY RESOURCES PROGRAM
NATIONAL INSTITUTES OF HEALTH

arith Ban, Ruan O47.
u

June 1, 1980

STANFORD UNIVERSITY SCHOOL OF MEDICINE
Edward A. Feigenbaum, Principal Investigator
Prelude: An Overview and Personal Statement

by Edward A. Feigenbaum, Principal Investigator

This prelude is unabashedly a statement of advocacy. AS we prepare
this proposal, gathering up the threads of our past achievement and weaving
them into a coherent picture of our future, there is in the SUMEX Project a
sense of pride and accomplishment, and a feeling of exhilaration and
momentum regarding the future.

SUMEX was established with three main goals:

1. to provide computing resources and human assistance to those
scientists working on applications of artificial intelligence
research in medicine and biology;

2. to test the idea that it was feasible to provide resources and
assistance to the nation from a single site, with time-shared
operating systems, national computer communication networks, and a
staff oriented toward the special problems of remote users;

3. to grow, from seed to plant, the community of scientists interested
in working on applications of AI to the biomedical sciences;
facilitating the growth, health, and vigor of-the community by the
use of electronic communications linking its members. One question
we were asking was, "Is there a new style of science that will
emerge in a communications-enhanced setting of national, rather than
institutional, scope?”

These goats were and are unique to SUMEX, and their pursuit has given
rise to a "spirit of SUMEX"--a spirit that unfortunately does not come
across well in the dry recitations of a proposal document; hence, this
personal prelude.

SUMEX's success as a national research resource

 

The SUMEX Project has demonstrated that it is possible to operate a
computing research resource with a national charter--that the services
providable over networks were those that were facilitative of the growth of
Al-in-Medicine. Previous NIH computer RR's were mostly institutional in
scope, occasionally regional (like the UCLA resource).

Some of the most notable projects in the history of Artificial
Intelligence were done with terminal-and-network, without a computer on
site. In human terms, this means,of course, without the headaches and
energy drains of proposing a machine, installing it, maintaining it and its
software, hiring its system programmers and operators, dealing with
communication vendors, etc. The famous INTERNIST program was developed
from Pittsburgh in this way. And the ACT computer model was begun at

Privileged Communication i £. A. Feigenbaum
Michigan, continued at Yale, and later at Carnegqie-Mellon, all without
moving the program or losing a day's work because of machine transition
problems.

The projects SUMEX supports have generally required substantial
computing resources with excellent interaction. This is hard to obtain in
all but a few universities. SUMEX is, in a sense, a “great equalizer". A
scientist gains access by virtue of the quality of his/her research ideas,
not by the accident of where s/he happens to be situated--in other words,
the ethic of the scientific journal.

SUMEX has demonstrated that a computer resource is a useful "Vinking
mechanism" for bringing together and holding together teams of experts from
different disciplines who share a common problem focus. For example,
computer scientists have been collaborating fruitfully with physical
chemists, molecular biochemists, geneticists, crystallographers,
internists, ophthalmologists, infectious disease specialists, intensive
care specialists, oncologists, psychologists, biomedical engineers, and
other expert practitioners. And in some of these cases, the
interdisciplinary collaboration, usually so difficult to achieve in the
best of circumstances, was achieved in spite of geographical distance
between the participants, using the computer networks.

SUMEX has achieved successes as a community builder. AI concepts and
software are among the most complex products of computer science.
Historically it has not been easy for scientists in other fields to gain
access to and mastery of them. Yet the collaborative outreach of SUMEX has
been able to bridge the gap in a number of cases. For example, Dr John
Osborn (Pacific Medical Center, San Francisco) and I found common
scientific interests in the application of AI to intensive care, and
initiated a SUMEX-based collaboration. That project resulted in a system
of potential significance to intensive care medicine; in two Stanford
computer science Ph.D. dissertations, hence two new doctoral-level recruits
to the ranks of computers-in-medicine specialists; in one computer
science/physiology Special Ph.D. Program for one of Dr. Osborn's biomedical
engineers; and an award to Dr. Osborn's team in 1979 from the Association
for the Advancement of Medical Instrumentation.

I wish to contrast this success story with the traditional
difficulties I have encountered outside the health research field in trying
to bridge the gap to engineering-oriented industrial firms. The human
resource and motivation was present. The SUMEX base of easily available
shared software technology was not. The resulting problems have generally
raised too high a threshold to overcome.

The SUMEX mission has been able to capture the contributions of some
of the finest computers-in-medicine specialists and computer scientists in
the country. For example, Professor Joshua Lederberg (SUMEX's first PI,
now President of The Rockefeller University) is Chairman of SUMEX's
Executive Committee: and Professor Donald Lindberg, M.D., Director of the
University of Missouri's Health Care Technology Center, is Chairman of the
AIM Advisory Group. Professor Herbert Simon of Carnegie-Mellon University,
Professor Marvin Minsky of MIT, and many other distinguished scientists

E. A. Feigenbaum WW Privileged Communication
serve on that peer review committee. These people are active participants
in SUMEX. Lederberg and Lindberg are continuing collaborators in the
research itself. And Simon, for exampte, was the person who prompted our
collaboration with psychologists at the University of Colorado.

SUMEX now has the reputation of a model national resource, pulling
together the best available interactive computing technology, software, and
computer communications in the service of a national scientific community.
Planning groups for national facilities in cognitive science, computer
science, and biomathematical modeling have discussed and studied the SUMEX
model.

SUMEX and Artificial Intelligence Research

 

The SUMEX Project is a relative latecomer to AI research. Yet its
scope has given strong impetus to this historic development in computer
application. AI research is that part of computer science that
investigates symbolic reasoning processes, and the representation of
symbolic knowledge for use in inference. It views heuristic knowledge to
be of equal importance with "factual" knowledge, indeed to be the essence
of what we call "expertise". In its "Expert Systems" work, it seeks to
Capture the expertise of a field, and translate it into programs that will
offer intelligent assistance to a practitioner in that field.

For computer applications in medicine and biology, this research patn
is crucial, indeed ineluctable. Medicine and biology are not presently
mathematically-based sciences; not like physics and engineering capable of
exploiting the mathematical characteristics of computation. They are
essentially inferential, not calculational, sciences. If the computer
revolution is to affect biomedical scientists, computers will be used as
inferential aids.

Perhaps the larger impact on medicine and biology will be the
exposure and refinement of the hitherto largely private heuristic knowledge
of the experts of the various fields studied. The ethic of science that
calls for the public exposure and criticism of knowledge has traditionally
been flawed for want of a methodology to evoke and give form to the
heuristic knowledge of scientists. The AI methodology is beginning to fill
that need. Heuristic knowledge can be elicited, studied, critiqued by
peers, and taught to students.

The tide of AI research and application is rising. AI is one of the
fronts along which university computer science groups are expanding. The
NSF's program in Intelligent Systems is vigorous and growing. The pressure
from student career-line choices is great: to cite an admittedly special
case, approximately one-third of the students applying to Stanford's
computer science Ph.D. program cite AI as a possible field of
specialization. In industry, new groups have been forming regularly: Texas
Instruments two years ago formed a substantial AI group: so did the oi1-
industry-service firm, Schlumberger, Inc.; IBM has reinitiated its AI work:
and the new genetic engineering firms are becoming interested.

Privileged Communication Vii E. A. Feigenbaum
The tide is rising largely because of the development in the 1970's
of methods and tools for the application of AI concepts to difficult
professional-level problem solving; and the demonstration in various areas
of medicine and other life sciences that these methods and tools really
work. Here SUMEX has played a key role, so much so that it is regarded as
"the home of applied AI."

SUMEX has been the nursery, as well as the home, of such well-known
AI systems as DENDRAL (chemical structure elucidation), MYCIN (infectious
disease diagnosis and therapy), INTERNIST (differential diagnosis), and ACT
(human memory organization). These, and other programs developed at SUMEX,
have played a seminal role in structuring modern AL paradigms and
methodology. First among these has been a shift of AI's focus from
inference procedures to knowledge representation and use. There is now a
recognition that the power of problem solvers derives primarily from the
knowledge that they contain--of the elements of the problem domain, of the
strategies for solving problems in that domain, and of the forms in which
the knowledge is to be acquired. In 1977, Goldstein and Papert of MIT,
writing in the journal Cognitive Science, described the change of focus as
a "paradigm shift” in AI. This shift was induced largely (though of course
not exclusively) by the work at SUMEX, beginning with the DENDRAL
development in 1965. ,

Toward the mid-'80s: the Future of SUMEX

Success breeds its problems. The revolution in computer technology
and costs adds complexity to their solution.

At the beginning, the SUMEX community was small, and idea-limited.
The SUMEX computer facility was an ideal vehicle for the research. Now the
community is large, and the momentum of the science is such that its
progress is now limited by computing power. The size and scientific
maturity of the SUMEX community has fully consumed the resource in every
critical dimension: CPU power, main memory size, and file space.

The limitation that AI researchers agree most critically limits their
scientific imagination, and adds inordinately to program development time,
is the 256K word main memory space, brought about by the 18 bit address of
the PDP-10's and 20's. Economically, main memory size need not be much of
a limitation any more, but it is essential to move to a machine with more
addressing bits.

But which machine? In the turmoil of the computer developments of
today, this is not easy to answer, Computers will come in many different
sizes and prices and each will fit a particular class of needs. Our
planning axiom for the period 1981-86 has been: the need to accommodate a
HETEROGENELTY of computers and peripheral devices. We must maintain a
flexible posture with respect to the introduction of new capabilities and
changing costs during this continuing revolution. Yet we must choose.

Our plan, sketched below, is conservative in maintaining and
extending SUMEX's current service level; yet is forward-looking enough to

E. A. Feigenbaum iv Privileged Communication
position SUMEX properly for mid-course corrections and for the computing
world of the late 1980's. Here it is, briefly sketched.

The existing DEC KI-10 duplex, with its superb software, will be
"fiiied out"--stretched to the point of diminishing returns from hardware
addition; then frozen. It is an amiable workhorse. We can not (indeed
dare not) do without it during this period of turbulence. But it has seen
better days, and will be ineffective by the end of the grant period.

A DEC VAX 11/780 will be acquired in the first year. Based on more
modern technology and a more competitive price, it has the extra address
bits that are required. On VAX we get the same kind of low-cost ride on
the software work of others that we got when we adopted TENEX and INTERLISP
for the KI-10's, The UNIX operating system is available, and is being
further developed under ARPA support. ARPA is also supporting the
reprogramming of INTERLISP for VAX. For integrated circuit design
research, ARPA has already placed two VAX computers at our Computer Science
Department, so we are building experience rapidly in VAX use. And, de
Facto, the VAX has become the “computer science machine” of the early ‘80s,
so that nationally its software development is moving rapidly. A family of
VAX's, both more and less powerful, at (hopefully) appropriate prices, is
in the wings.

The "technology transfer" machine to which we will move the heavy
national use of SUMEX's mature AI applications (such as DENDRAL, SECS,
MOLGEN, VM) will be another DEC VAX, acquired in the middle of the period.
This machine's role is intended to be entirely analogous to the role
currently played by the DEC 2020 at SUMEX vis a vis the KI-10 duplex. It
will be the VAX-era prototype of the "spinoff" machine, loosely tethered to
SUMEX by networks. In the last DENDRAL Project renewal, the NIH Study
Section denied such a machine to DENDRAL, suggesting that the required
resource would better be provided by SUMEX. We seek, and plan, to assume
this obligation.

And what about the single-user professional scientific workstation--
the powerful, small, cheap officemate that will serve most of the
researcher’s computing needs? Much of the present turbulence in the
computing world swirls around this question. Yes, we believe it is coming,
and will probably be an economically viable concept in the late '80s. No,
we do not believe it will be powerful enough or cheap enough for most
routine research needs in the planning period. Yet we must begin to
explore the space of possibilities opened up by these machines, eschewing
articles of faith for real experience. We must learn to build systems of
these machines and to build and manage graceful software for these systems.
If decentralization is in our future, we must learn its technical
characteristics. Consequently, we have planned the acquisition of a number
of such single-user workstations over the course of the coming period, some
to be placed at Stanford, some in the national community, at the decision
of the Executive Committee.

These machines will be tethered to the SUMEX central facility and

staff by local digital network at Stanford and by national network to the
non-Stanford community. With DEC 10's, 20, VAX's, and workstations

Privileged Communication Vv E. A. Feigenbaum
coexisting to serve community needs, it is economical and convenient to
continue the centralization of file storage, and the networks make it
possible for most applications at Stanford and many applications
nationally. Computer scientists are in general agreement that economies of
scale will continue to dominate in secondary storage for some time. We
have planned, therefore, to alleviate the present file space shortage not
by adding discs to machines in an ad hoc fashion but by adding a common
file server to the resource. To facilitate the transfer of software and
access to valuahle common facilities, the SUMEX complement of equipment
will be linked by focal digital networks to other major centers of
computing at Stanford, most important of which is the Computer Science
Department.

The success of SUMEX is the success of its dedicated and
extraordinarily competent staff, headed by Tom Rindfleisch. This human
resource of SUMEX should not, and will not, be decentralized. In the world
of computer systems talent and user-assistance expertise, there are indeed
continuing large “economies of scale".

The smoothly operating management structure of SUMEX is one of its
joys and victories. We do not plan to fix something that is not broken.
We plan that the Executive Committee and the AIM Advisory Committee will
continue to function as they now do.

So this is it in a nutshell:
Run the present configuration with more main memory; acquire two VAX
large-memory systems (years 1 and 3) for new research and for maturing
project communities; cautiously add some single-user professional
workstations; acquire a common file server; link everything in a
transparent digital networking scheme; continue the central staff and
management structure, essentially unchanged in size and function.

As we add up the budget (flinchingly, I hasten to say), we note that
the cost will not be cheap, despite the much-touted fall in the cost of
computing. But we believe we have been conservative; that the scientific
community we serve needs these resources; and that by its science and its
applications orientation, it has earned them.

I look at the widely acclaimed NSF report calling for the
refurbishing of computer equipment for experimental computer science (the
so-called "Feldman Report") and note that it calls for “refurbishing”
expenditures for just a single department greater than that budgeted in
this proposal, with a "refresh" cycle of five years to accommodate
advancing technology. The scientific work of the SUMEX-AIM community is
the quintessence of experimental computer science. It is advancing, and
gaining acceptance, beyond expectations. SUMEX serves the nation, not one
university or department. I believe that its budget accords well with the
national interest and with the scientific interest.

E. A. Feigenbaum vi Privileged Communication
Conclusion: the "Spirit of SUMEX"

 

I would like to conclude not with my own words but with the words of
Professor Douglas Brutlag, a Stanford Biochemist who collaborates with my
group on the MOLGEN project and who sent me, unsolicited, the letter quoted
below in its entirety. Nothing I could say could more accurately portray
the "spirit of SUMEX" mentioned earlier.

"My original role in the Molgen project was that of
a biochemist advisor to those developing a knowledge base of
molecular biological information and techniques. I rapidly
found that SUMEX could be very useful to my own work in ways
that I had never expected. First, MOLGEN was a success very
early and I now routinely use the artificial intelligence
methods incorporated within the frame oriented knowledge base
in my everyday work in the laboratory. I use the knowledge
base not only to store our results from experiments and to
analyze them, but I can readily interact with the knowledge
base to examine the data from several different viewpoints and
display it in different ways,

In addition to the interactive nature of knowledge
base work, I have found computer networks and file transfer
protocols to be exceptionally useful. The nation wide
commercial networks have permitted many of my colleagues across
the country to try out the software we have developed at
Stanford in pilot projects. This together with message sending
capabilities has resulted in instantaneous feed back about the
work we have done and allowed us to develop our program and to
incorporate ideas from a much larger base of expertise.

Several collaborative arrangements have been set up and some
have even become involved in our programming efforts.

Moreover, our software has had such general utility that
subsequently many of the other workers have obtained accounts
on their local computers and we have sent them the software by
file transfer protocols. Electronic information transfers have
Saved both time and energy in preparing hard copy versions’ as
well as facilitated the update programs at many distant
locations.

I think that one of the major reasons that SUMEX
works so well is that it is designed with the naive user in
mind, Because it is so interactive and user oriented, the
activation energy to learn how to use the system is very Tow.
Of all of the interactive systems with which I have worked
(five in all), SUMEX was not only the easiest, but was indeed a
real pleasure. I felt more like the system was working for me
from the very beginning, rather than me fighting the system.
Hence, my productivity on SUMEX has increased immeasurably. In
addition, I have no hesitation encouraging others at remote
sites to use SUMEX in the collaborative efforts mentioned
above."

Privileged Communication vii E. A. Feigenbaum
Table of Contents

 

Section Page

Prelude: An Overview and Personal Statement . . . . . . . . .7

List of Figures Be ee aN

1. Biographical Sketches See kk 2
2. Budget ra
2.1 First Year Budget Detail (8/1/81 - 7/31/82) eee le le SS
2.1.1 Total First Year Budget . . . . . . UL 8

2.1.2 First Year Personnel Detail . . . . . . . . . .4

2.2 5-year Budget Summary (8/81 - 7/86) . . . . . . . LS

2.3 Budget Explanation and Justification. . . . . . . . .6

3. Introduction and Aims a 13
3.1 Overview of Objectives and Rationale . . . . ... . 14
3.1.1 Definitions of Artificial Intelligence . . . . . 14

3.1.2 Resource Sharing ee 16

3.2 SUMEX-AIM Background . . . . we 16

3.3 Specific Aims a 18
3.3.1 Resource Operations a 19

3.3.2 Training and Education Se eee 20

3.3.3 Core Research a 20

4, Significance Be kk 22
5. Progress Se ee ee ke ew 80

E. A. Feigenbaum viii Privileged Communication
5.1 Brief Statement of Prior Goals . . . . . . .
5.1.1 Resource Operations
5.1.2 Training and Education a
5.1.3 Core Research ee ee
§.2 Summary of Progress: 11/77 - 4/80
5.3 Detailed Progress Highlights
5.3.1 Resource Operations
5.3.1.1 System Hardware
6.3.1.2 System Software oo.

5.3.1.3 Network Communication Facilities

5.3.1.4 Resource Management...
5.3.2 Core Research eee
5.3.3 SUMEX Staff Publications . . . . . . .
6. Methods of Procedure -
6.1 Resource Operations Plans

6.1.1 Resource Hardware
6.1.1.1 Rationale for Future Plans
6.1.1.2 Summary of Proposed Hardware Acquisitions
6.1.1.3 Existing Hardware Operation
6.1.1.4 Large Address Space Machines
6.1.1.5 Single-User Professional Workstations
6.1.1.6 File Server .
6.1.2 Communication Networks
6.1.2.1 Long-Distance Connections
6.1.2.2 Local Intermachine Connections
6.1.3 Resource Software

6.1.4 Community Management

Privileged Communication ix E. A.

30
30
30
31
32
34
34
34
39
46
AT
49

51

52
52
52
52
55
56
57
57
59
62
62
62
64

66

Feigenbaum
6.2 Training and Education Plans re

6.3 Core Research Plans

6.3.

6.3.
6.3.
6.3.

7.

8.

9.

1

2

3

4

Knowledge Representation

3.1.1 RLL -- The Representation Language

Language

3.1.2 Research on Planning

.3.1.3 Causal Models

Knowledge Utilization and Tools for Building
Expert Systems

3.2.1 Attempt to Generalize (AGE)
.3.2.2 AI Handbook. . . . 1 2 whe ee

3.2.3 Research in Automated Consultation about

Expert Systems,

3.2.4 EMYCIN see

Knowledge Acquisition

Explanation . .,

Available Facilities

Literature Cited

Collaborative Project Reports

9.1 Stanford Projects.

9.1.

9.1.

E. A. Feigenbaum

1

2

AGE - Attempt to Generalize .
AI Handbook Project

DENDRAL Project .

MOLGEN Project

MYCIN Project

Protein Structure Project

RX Project

68
69
71

71
73

77

78
78

78

79
80
82
85

89

90

135
136
137
145
149
171
186
205

211

X Privileged Communication
9.2 National AIM Projects

9.2.

9.2.

1

2

Acquisition of Cognitive Procedures (ACT)

SECS - Simulation and Evaluation of Chemical
Synthesis

Hierarchical Models of Human Cognition
HMF - Higher Mental Functions
INTERNIST Project

PUFF/VM Project

Simulation of Cognitive Processes

Rutgers Computers in Biomedicine Project
fRutgers-AIM]

Decision Models in Clinical Diagnosis [Rutgers-
AIM] re .

Heuristic Decisions in Metabolic Modeling
[Rutgers-AIM ]

Stanford Projects
Ultrasonic Imaging Project

AIM Projects

9.4.1 Coagulation Expert Project
9.4.2 Communication Enhancement Project
9.4.3 A Computerized Psychopharmacology Advisor
9.4.4 Computer-Aided Refinement of Medical Knowledge
9.4.5 Interactive Statistical Package Advisor
9.4.6 Conceptual Structures for Medical Diagnosis
[Rutgers-AIM] .
Privileged Communication x7 E. A,

220

. 221

226
234
240
244
250

261

267

282

286
289
290
297
298
302
309
317

321

323

Feigenbaum
E.

A,

Appendix A

Community Growth and Project Synopses

Appendix B

Resource Operations and Usage Statistics

Appendix C

Local Network Integration...

Appendix D

Remote Network Communication Facilities

Appendix E

Resource Management Structure

Appendix F

LISP Address Space Limitations

Appendix G

AI Handbook Outline

Appendix H

MAINSAIL System Demonstration

Appendix I

AIM Management Committee Membership

Feigenbaum xii

soe « « 331

355

374

376

383

390

392

398

399

Privileged Communication
List of Figures

 

Figure

1. Current SUMEX-AIM KI-10 Computer Configuration

2. Current SUMEX-AIM 2020 Computer Configuration

3. Intermachine Connections via ETHERNET . .

4, Proposed VAX configuration

5. Planned Ethernet System to integrate System Hardware
6. SUMEX-AIM Growth by Community .

7. Total CPU Time Consumed by Month see et
8. Peak Number of Jobs by Month . . . . . . .
9. Peak Load Average by Month

10. Monthly CPU Usage by Community

11. Monthly File Space Usage by Community . . .

12. Monthly Terminal Connect Time by Community

13. Average Diurnal Loading (4/80): Number of Jobs

14. Average Diurnal Loading (4/80): Load Average

15. Average Diurnal Loading (4/80): Percent Time Used
16. TYMNET Terminal Connect Time

Privileged Communication xiii

Page

36

37

38

. 60

61

331

356

. 357

357

. 359

360

361

369

370

370

371

Feigenbaum
17.

18.

19.

20,

21.

E.

A.

ARPANET Terminal Connect Time

TYMNET Network Node List

ARPANET Geographical Network Map

ARPANET Logical Network Map

TELENET Geographical Network Map

Feigenbaum Xiv

‘372

379

380

381

382

Privileged Communication