AIList-REQUEST@SRI-AI.ARPA (AIList Moderator Kenneth Laws) (01/08/86)
AIList Digest Wednesday, 8 Jan 1986 Volume 4 : Issue 2
Today's Topics:
Corrections - Feigenbaum's Comments & Xerox Reader Count,
Query - AI Paradigm,
Review - Stanford SDI Debate (12/19)
----------------------------------------------------------------------
Date: Sun 29 Dec 85 22:18:33-PST
From: Edward Feigenbaum <FEIGENBAUM@SUMEX-AIM.ARPA>
Subject: re the news report on my speech in the Netherlands
Saw my name in the Fri 27 Dec. 1985 AIList Digest V3 #192.
Since it's best not to let silly things propagate, let me say
here what I said (I actually said many many things; I don't understand
why those few things were picked out).
I said that among the most commercially important applications of
expert systems in the next ten years would be factory management
applications and financial service applications. (I didn't even
mention factory automation).
I said that speech understanding applications would become economically
very important. (I never mentioned speech generation.)
Best wishes for a journalistically accurate New Year (fat chance),
Ed Feigenbaum
------------------------------
Date: 6 Jan 86 16:22 PST
From: Newman.pasa@Xerox.ARPA
Subject: Reader Count
I don't know if you want to post this to the net or not, but in the
interest of accuracy, Xerox has approximately 248 readers of the AIList.
>>Dave
------------------------------
Date: Tuesday, 7 January 1986 02:19:31 EST
From: Duvvuru.Sriram@cive.ri.cmu.edu
Subject: AI Paradigm
I have seen the word "AI Paradigm" in several papers/reports. My dictionary
[Random House] says that a Paradigm is either an example or a model. Is
there any other meaning to it or is it just a better word for "example"?
Sriram
------------------------------
Date: Tue, 31 Dec 85 11:55:37 pst
From: jrisberg@aids-unix (Jeff Risberg)
Subject: Stanford SDI debate (12/19) summary
The following is a somewhat long summary of the technical debate on SDI
entitled "SDI: How Feasible, How Useful, How Robust?" that was held at
Stanford on December 19th. Since this debate was announced on AILIST,
we felt that readers would be interested in this summary.
[I was reluctant to permit the initial announcement and I am reluctant
to permit the summary. I have decided to forward them because SDI
may well involve major funding in the area of AI. Please restrict any
discussion in AIList to the areas of AI, pattern recognition, or the
feasibility of distributed decision making. Political discussions
would be more appropriate on Arms-D@MIT-MC, Risks@SRI-NIC, or
perhaps Space@MIT-MC. -- KIL]
The panelists at the debate were:
Advocates:
Professor Richard Lipton, Professor of Computer Science at Princeton
University, current member of SDIO's Panel on Computing and Support of Battle
Management.
Major Simon Peter Worden, the Special Assistant to the Director of the SDIO
and Technical Advisor to the Nuclear and Space Arms Talk with the USSR
in Geneva.
Opponents:
Dr. Richard L. Garwin, IBM Fellow and Adjunct Professor of Physics at
Columbia University, Physicist and Defense Consultant.
Professor David Parnas, Lansdown Professor of Computer Science at the
University of Victoria, former member of the SDI Organization's
Panel on Computing and Support of Battle Management.
Dr. Goldberger, President of CalTech, served as the moderator of the
discussion. He presented a bit of history relating to the subject of
defensive warfare and then allowed the panelists to speak. There are
certainly historical precedents for defensive systems, in fact, each US
leader since the 1950's has sought a defense. SDI is simply the largest
scale and most visible concept to date.
Because of the complexity of the issue, a question like "can it work?"
can only be answered by determining 'what does "can" mean?', 'what does
"it" mean?', and 'what does "work" mean?'. There have been various
justifications proposed for SDI, and the technical and political
community has raised numerous questions.
The format of the debate consisted of three sections: during the first
section, each speaker was allowed 20 minutes with which to present his
case; following that, each speaker had a 5-minute rebuttal period;
finally, the audience was allowed to ask questions via screened 3 x 5
index cards.
Major Worden spoke first and discussed some positive aspects of SDI. In
his view, the principal justification is for Arms Control. In that
view, a major goal is that the SDI system have a lower marginal cost
than that of building additional offensive systems. Survivability is
another goal.
He said that the object of SDIO is to establish the feasibility of the
system, but not to build it. Similarly, it may not involve space
weapons, although most of the current concepts include a space segment.
We would also like to get the Soviets to admit their own work in such
systems. They regularly deny such work, but when we show them ariel
photos of their high-power lasers, they say, "Oh, we do have laser
research for medical purposes".
The numerical aspects of a Defensive Reliant Deterance are that each
layer of defense drives the number of offensive warheads needed up
further. There are a series of layers, with each layer consisting of
sensors, weapons, and battle management systems. He showed some of the
standard slides of this design. The concept is "Proliferated and
Distributed". They are planning for the late 1990's.
The key issue as he sees it is countermeasures, such as the fast burn
booster. There are three types of threats and five types of
countermeasures. He expected that countermeasures will develop much as
aircraft gained shielding, speed, and proliferation countermeasures
after WW I.
He gave a cost context of the project in comparison to the cost of
insurance in the private sector. The cost of insurance is over $300
billion/year, while the SDI work is currently costing $1.5 billion.
Major Worden admitted that President Reagan caught everyone off guard
with his speech about SDI two years ago.
Dr. Garwin spoke next. He recommended that we think carefully about
just what are the goals, costs, and likely Soviet Response to SDI. He
said that the Scowcroft Commission reported that U.S. security could be
maintained without SDI. His opinion is that while SDI has been proposed
to replace deterrance, it is really simply another form of deterrance.
He is concerned with the layered approach of the system in that there is
catastrophic failure if one layer does not do its job. For example, the
design of each layer assumes that the prior one does its job in reducing
the number of incoming objects.
Examples of potential areas of failure were given: space mines could
easily knock out any space segment units and midcourse intercept could
be overwhelmed by large numbers of decoys. Dr. Garwin feels that the
systems needed for SDI can not be built under the ABM treaty.
There has been a progression of goals, and in effect "replace deterance"
has become "strengthen deterance".
He closed by describing his view of a viable strategic balance, which
would be to limit each side to 1000 warheads, deployed on small
missiles, small subs, and cruise missiles, with no counterforce threats
against strategic targets. Preserve the ADM treaty. BMD research may
be continued, in order to confirm that there is no threat to the system.
Dr. Lipton was the next speaker. He joined the technical panel of SDIO
last summer. (An interesting point is that Dr. Lipton is a former
student of Dr. Parnas.) His major focus was in the importance of the
de-centralization of software. The Fletcher panel design was centralized,
with software in charge of everything. "Is it possible to build a system
without these problems?" Discussion of feasibility must encompass all
design possibilities, and Lipton stressed the merits of a decentralized
design.
He led into this by an analogy with the banking industry. Banking works
because it is a large collection of loosely organized components.
In the SDI example, he refered to large numbers of satellite groups
handling independent battle management functions. Fault tolerance would
provide reliability, like the concept of the strategic weapons triad.
He argued that these seperate groups would be testable, by putting a few
into orbit and shooting missiles at them.
The false alarm problem could be controlled by activating different
numbers of systems. Coordination problems were raised by the Fletcher
panel toward a goal of conserving "bullets". Dr. Lipton's studies
indicated that the shot overhead of low coordination is not that high.
Dr. Parnas spoke last. He had good points in his speech, but had a
problem with becoming quite caustic in his remarks about the SDIO
members. He said that (loosely quoted) "I used to feel that arms
control people are guilty of wishful thinking, but I have now seen a
whole new standard." His major complaint against SDI is that SDI forces
us to trust the system; if SDI need not be perfect, it must at least be
trustworthy, and he feels that this is not possible.
Conditions for validation include: mathematical analysis,
exhaustive case analysis,
or prolonged, realistic testing.
Even after one or more of these conditions have been met, the system
must still be operated under controlled conditions.
The validation of software is inherently different from the standard
engineering problems such as bridge design. Something different about
software. It is made up of discrete, rather than continuous functions.
Thus design principles such as building for twice the weight do not
truly apply, but instead, the number of discrete cases must be
examined, along with thorough testing. Even after a number of years of
use, bugs may still be found. True testing requires thousands of years.
For most software, we can allow unreliable software, as long as we trust
it. For SDI, we cannot.
He doesn't believe that de-centralization provides added trustworthiness
to the system. He stated that he never took the Fletcher design
seriously in the first place, feeling it that was no more than a rough
sizing of the problem. There are a series of myths around
de-centralization.
Dr. Parnas' final point is that SDI is not a limit of computers, but of
human beings.
The rebuttals were then held. Major Worden questioned the meaning of
deterence and then mentioned some possible alternatives to SDI:
automatic launch under attack, preventive attacks, and bombs under U.S.
cities. He indicated that Dr. Garwin had shown only that he could design
a system that SDIO wouldn't buy. His final comment re-iterated the
linkage of SDI to the arms control process.
Dr. Garwin (and the others) kept mentioning the Scowcroft report which
produced possible defensive measures other than SDI. Dr. Garwin pointed
out that Dr. Lipton had only found that the system proposed by Fletcher
might not work, but that Lipton believed others might. In any case, so
long as Soviets can deliver by other means (cruise missiles), we will
continue to need deterance.
Dr. Lipton restated his belief in the need for independent
systems. He recognized that nothing is perfect, that even computers
are not reliable, but they are used on a daily basis. The use of
independent battle stations would stress the sensors, but he argued
that teraflops would alleviate the need for independent views.
Dr. Parnas again made a couple of inappropriate shots..."I've
been to a lot of Mickey Mouse meetings, but the ones sponsored by SDI
had the biggest ears and biggest nose I've ever seen." He thinks that
the idea of separate systems does not remove the size or complexity of
what is needed; dividing 10 million lines of code into small modules of
1000 lines does still not ensure error-free code. People do not write
independent code.
Questions raised from the floor asked about different types of lasers,
the time to phase-in to SDI, and about the non-ICBM threat. Worden
replied that cruise missiles are not strategic weapons because of their
flight time, and that smuggling bombs into the US would not be a realistic
approach for a Soviet leader to taken.
The speakers then made closing comments:
Dr. Garwin said that we currently have a real opportunity for arms
reduction. This would be much more survivable than continued escalation
and research into defensive weaponry. He feels that both sides should
abandon defense efforts. Control of nuclear proliferation is essential.
Major Worden agreed with these points of Dr. Garwin, but said that it is
necessary and vital to carry forward a defensive program within the ABM
treaty to provide a different kind of security.
Dr. Parnas said that in software, the engineering term of "tolerance"
depends on continuity. "Almost right" does not make sense in the
context of SDI. He fears espionage that would result in someone getting
a copy of the software. Reasons for not going ahead with SDI anyway
include the lost opportunity for other projects, low quality of results,
and weakening of the strategic position.
Dr. Lipton said that if the independent segments of the SDI system do
not interact, the code is not vulnerable. He pointed out that there
are simple systems, such as elevators, that we do trust.
Dr. Goldberger then made some closing comments. He said that strategic
defense and arms control must be approached seriously. The laws of
physics are immune to political views and we are currently at a critical
political point. A decision to push forward defensively without a
reduction in offense would be a mistake. SDI has been proposed as part
of current moves toward lowering threat of destruction, yet it is
difficult, with verification problems, and major risks. He hopes that
the human spirit will prevail in the decisions that must be made.
In summary, the debate was quite interesting, although inconclusive if
judged in a strict manner. We were most surprised that software
technical details were hardly mentioned, and that political and non
computer technology issues were the focus of the discussion. Dr. Parnas
and Dr. Lipton made several comments against each other, which detracted
from the technical discussion. It didn't appear that Dr. Lipton was
overly familiar with the SDI problems; he continually talked in
generalities, with few facts with which to back up his statements. Dr.
Garwin and Major Worden were much more prepared in their talks and
didn't take any cheap shots with which to score points with the
audience.
The comments above are strictly our personal opinions and not
representative of any organization.
Jeff Risberg (jrisberg@aids-unix)
Susan Rosenbaum (susan@aids-unix)
------------------------------
End of AIList Digest
********************