Time Magazine -- Computers of the Future
Barry Shein
The cover article this week for Time Magazine is "Computers of the
Future". Mainly what they mean by that are the two paths
supercomputing and artifical intelligence (eg. neural networking.)
I haven't finished it, it seems fun, some errors of course but you
don't expect extreme technical accuracy from such an article anyhow.
-Barry Shein, Boston University
Ken Lee
What do people think of the PRACTICAL future of artificial intelligence?
For a while, it seemed like it was going to take off. All sorts of
expert systems and tools houses started to appear. Most of these are
bankrupt now. Even the biggies like Symbolics, Teknowledge, and
IntelliCorp are having major trouble. The only companies that are
successful are Star Wars contractors, and I'm not sure if that's what
you'd call a practical application.
Is AI just too expensive and too complicated for practical use? I
spent 3 years in the field and I'm beginning to think the answer is
mostly yes. In my opinion, all working AI programs are either toys or
could have been developed much more cheaply using conventional
techniques.
Why is AI expensive? No matter how good our theoretical inference or
representation techniques get, we still have the practical problem of
extracting human knowledge transfering it to the machine. The problem
I see is that every application is sufficiently unique to preclude
automated knowledge engineering. Since you must solve it by hand
anyway, you could probably more effiently program it using conventional
techniques.
Does AI have any advantage over conventional programming? There are
claims about the benefits of learning systems, meta-level algorithms,
abstraction, etc., but for these to be cost effective, they must be
developed in an application-independent fashion. Is this possible?
I don't think so, at least not in the near future.
I'm probably being short-sighted and ignoring the (very) long-term
possibilities. There are also side-effects, like the popularizing of
object-oriented programming and powerful programming environments,
that are very practical. Maybe I'm just looking at AI too broadly and
not at specific application areas.
What do you thing? Thanks for your thoughts. No, I'm not an AI trying
to clone you.
Ken
--
What's the difference between a used car salesman and a computer salesman?
The used car salesman knows when he's lying.
Sean Brunnock
>What do people think of the PRACTICAL future of artificial intelligence?
>
>
Bear with me while I put this into a sociological perspective. The first
great "age" in mankind's history was the agricultural age, followed by the
industrial age, and now we are heading into the information age. The author
of "Megatrends" points out the large rise in the number of clerks as
evidence of this.
The information age will revolutionize agriculture and industry just as
industry revolutionized agriculture one hundred years ago. Industry gave to
the farmer the reaper, cotton gin, and a myriad of other products which
made his job easier. Food production went up an order of magnitude and by
the law of supply and demand, food became less valuable and farming became
less profitable.
The industrial age was characterized by machines that took a lot of
manual labor out of the hands of people. The information age will be
charcterized by machines that will take over mental tasks now accomplished
by people.
For example, give a machine access to knowledge of aerodynamics,
engines, materials, etc. Now tell this machine that you want it to
design a car that can go this fast, use this much fuel per mile, cost
this much to make, etc. The machine thinks about it and out pops a
design for a car that meets these specifications. It would be the
ultimate car with no room for improvement (unless some new scientific
discovery was made) because the machine looks at all of the possibilities.
These are the types of machines that I expect AI to make possible
in the future.
I know this is an amateurish analysis, but it convinces me to study
AI.
As for using AI in conventional programs? Some people wondered what
was the use of opening up a trans-continental railroad when the pony
express could send the same letter or package to where you wanted in just
seven days. AI may be impractical now, but we have to keep making an effort
at it.
Sean Brunnock
University of Lowell
sbru...@eagle.cs.ulowell.edu
Andrew Jennings
at an airport waiting for a plane, and the issue of Time was on the newstands.
I guess you have to allow for a fair bit of hype in these things. Even so the
article on AI seemed fairly over-optimistic to me. Sure there are many "second-
wave" applications of expert systems out there, but there are also a lot of
research issues open.
I'd recommend the articles, and the picture of DeKleer's socks is also worthwile.
(Postmaster:- This mail has been acknowledged.)
John B. Nagle
>industrial age, and now we are heading into the information age. The author
>of "Megatrends" points out the large rise in the number of clerks as
>evidence of this.
The number of office workers in the U.S. peaked in 1985-86 and has
declined somewhat since then. White collar employment by the Fortune 500
is down substantially over the last five years. The commercial real estate
industry has been slow to pick up on this, which is why there are so many
new but empty office buildings. The new trend is back toward manufacturing.
You can't export services, except in a very minor way. (Check the numbers
on this; they've been published in various of the business magazines and
can be obtained from the Department of Commerce.)
> For example, give a machine access to knowledge of aerodynamics,
>engines, materials, etc. Now tell this machine that you want it to
>design a car that can go this fast, use this much fuel per mile, cost
>this much to make, etc. The machine thinks about it and out pops a
>design for a car that meets these specifications. It would be the
>ultimate car with no room for improvement (unless some new scientific
>discovery was made) because the machine looks at all of the possibilities.
Wrong. Study some combinatorics. Exhaustive search on a problem like
that is hopeless. The protons would decay first.
John Nagle
Rick Wojcik
>
>spent 3 years in the field and I'm beginning to think the answer is
>mostly yes. In my opinion, all working AI programs are either toys or
>could have been developed much more cheaply using conventional
>techniques.
>
flames low :-). Please try to remember that AI is a vast subject area.
It is expensive because it requires a great deal of expertise in language,
psychology, philosophy, etc.--not just programming skills. It is also a
very high risk area, as anyone can see. But the payoff can be
tremendous. Moreover, your opinion that conventional techniques can
replace AI is ludicrous. Consider the area of natural language. What
conventional techniques that you know of can extract information from
natural language text or translate a passage from English to French?
Maybe you believe that we should stop all research on robotics. If not,
would you like to explain how conventional programming can be used get
robots to see objects in the real world? But maybe we should give up on
the whole idea. We can replace robots with humans. Would you like to
volunteer for the bomb squad :-)? In the development stage, AI is expensive,
but in the long term it is cost effective. Your pessimism about the field
seems to be based on the failure of expert systems to live up to the hype.
The future of AI is going to be full of unrealistic hype and disappointing
failures. But the demand for AI is so great that we have no choice but to
push on.
--
Rick Wojcik csnet: rwo...@boeing.com
uucp: {uw-june uw-beaver!ssc-vax}!bcsaic!rwojcik
address: P.O. Box 24346, MS 7L-64, Seattle, WA 98124-0346
phone: 206-865-3844
John Snyder
>... But the demand for AI is so great that we have no choice but to
>push on.
We always have the choice not to develop a technology; what may be lacking
are reasons or will.
jsn...@june.cs.washington.edu John R. Snyder
{ihnp4,decvax,ucbvax}!uw-beaver!jsnyder Dept. of Computer Science, FR-35
University of Washington
206/543-7798 Seattle, WA 98195
Simon Brooke
>>What do people think of the PRACTICAL future of artificial intelligence?
>>
>>Is AI just too expensive and too complicated for practical use? I
>>
>>Does AI have any advantage over conventional programming?
>
> Bear with me while I put this into a sociological perspective. The first
>great "age" in mankind's history was the agricultural age, followed by the
>industrial age, and now we are heading into the information age. The author
Oh God! I suppose the advantage of the net is that it allows us to betray
our ignorance in public, now and again. This is 'sociology'? Dear God!
> For example, give a machine access to knowledge of aerodynamics,
>engines, materials, etc. Now tell this machine that you want it to
>design a car that can go this fast, use this much fuel per mile, cost
>this much to make, etc. The machine thinks about it and out pops a
>design for a car that meets these specifications.
And here we really do have God - the General Omnicompetent Device - which
can search an infinite space in finite time. (Remember that Deep Thought
took 7 1/2 million years to calculate the answer to the ultimate question
of life, the universe, and everything - and at the end of that time could
not say what the question was).
Seriously, if this is why you are studying AI, throw it in and study some
philosophy. There *are* good reasons for studying AI: some people do it in
order to 'find out how people work' - I have no idea whether this project
is well directed, but it is certain to raise a lot of interesting
problems. Another is to use it as a tool for exploring our understanding
of such concepts as 'understanding', 'knowledge', 'intelligence' - or, in
my case, 'explanation'. Obviously I believe this project is well directed,
and I know it raises lots of of interesting problems...
And occasionally these interesting problems will spin off technologies
which can be applied to real world tasks. But to see AI research as driven
by the need to produce spin-offs seems to me to be turning the whole
enterprise on its head.
** Simon Brooke *********************************************************
* e-mail : si...@uk.ac.lancs.comp *
* surface: Dept of Computing, University of Lancaster, LA 1 4 YW, UK. *
*************************************************************************
--
** Simon Brooke *********************************************************
* e-mail : si...@uk.ac.lancs.comp *
* surface: Dept of Computing, University of Lancaster, LA 1 4 YW, UK. *
*************************************************************************
S.Saal
something is still only vaguely understood, it is considered a
part of AI. Once we start understanding the `what,' `how,' and
(sometimes) `why' we no longer consider it a part of AI. For
example, all robotics used to be part of AI. Now robotics is a
field unto itself and only the more difficult aspects (certain
manipoulations, object recognition, etc) are within AI anymore.
Similarly so for expert systems. It used to be that ES were
entirely within the purview of AI. That was when the AI folks
had no real idea how to do ESs and were trying all sorts of
methods. Now they understand them and two things have happened:
expert systems are an independant branch of computer science and
people have found that they no longer need to rely on the
(advanced) AI type languages (lisp, etc) to get the job done.
Ironically, this makes AI a field that must make itself obsolete.
As more areas become understood, they will break off and become
their own field. If not for finding new areas, AI would run out
of things for it to address.
Does this mean it isn't worth while to study AI? Certainly not.
If for no other reason than AI is the think tank, problem
_finder_ of computer science. So what if no problem in AI itself
is ever solved? Many problems that used to be in AI have been,
or are well on their way to being, solved. Yes, the costs are
high, and it may not look as though much is actually coming out
of AI research except for more questions, but asking the
questions and lookling for the answers in the way that AI does,
is a valid and useful approach.
--
Sam Saal ..!attunix!saal
Vayiphtach HaShem et Peah HaAtone
The Polymath
My empoloyers just sponsored a week-long in-house series of seminars,
films, vendor presentations and demonstrations of expert systems
technology. I attended all of it, so I think I can reasonably respond to
this.
Apparently, the expert systems/knowledge engineering branch of so called
AI (of which, more later) has made great strides in the last few years.
There are many (some vendors claim thousands) of expert system based
commercial applications running in large and small corporations all over
the country.
In the past week we saw presentations by Gold Hill Computers (GOLDWORKS),
Aion Corp. (ADS), Texas Instruments (Personal Consultant Plus) and Neuron
Data (Nexpert Object). The presentations were impressive, even taking
into account their sales nature. None of the vendors is in any financial
trouble, to say the least. All claimed many delivered, working systems.
A speaker from DEC explained that their Vax configurator systems couldn't
have been developed without an expert system (they tried and failed) and
is now one of the oldest and most famous expert systems running.
It was pointed out by some of the speakers that companies using expert
systems tend to keep a low profile about it. They consider their systems
as company secrets, proprietary information that gives them an edge in
their market.
Personal Impressions:
The single greatest advantage of expert systems seems to be their rapid
prototyping capability. They can produce a working system in days or
weeks that would require months or years, if it could be done at all, with
conventional languages. That system can subsequently be modified very
easily and rapidly to meet changing conditions or include new rules as
they're discovered. Once a given algorithm has stabilized over time, it
can be re-written in a more conventional language, but still accessed by
the expert system. The point being that the algorithm may never have been
determined at all but for the adaptable rapid prototyping environment.
(The DEC Vax configurator, mentioned above, is an example of this. Much of
it, but not all, has been converted to conventional languages).
As for expense, prices of systems vary widely, but are coming down. TI
offers a board with a LISP mainframe-on-a-chip (their term) that will turn
a MAC-II into a LISP machine for as little as $7500. Other systems went
as high as an order of magnitude over that. I personally think these
won't really take off 'til the price drops another order of magnitude to
put them in the hands of the average home hacker.
Over all, I'd have to say that expert systems, at least, are alive and
well with a bright future ahead of them.
About Artificial Intelligence:
I maintain this is a contradiction in terms, and likely to be so for the
forseeable future. If we take "intelligence" to mean more than expert
knowledge of a very narrow domain there's nothing in existence that can
equal the performance of any mammal, let alone a human being. We're just
begining to explore the types of machine architectures whose great^n-
grandchildren might, someday, be able to support something approaching
true AI. I'll be quite amazed to see it in my lifetime (but the world has
amazed me before (-: ).
--
The Polymath (aka: Jerry Hollombe, holl...@TTI.COM) Illegitimati Nil
Citicorp(+)TTI Carborundum
3100 Ocean Park Blvd. (213) 452-9191, x2483
Santa Monica, CA 90405 {csun|philabs|psivax|trwrb}!ttidca!hollombe
Arti Nigam
In article <45...@june.cs.washington.edu> you write:
>
>We always have the choice not to develop a technology; what may be lacking
>are reasons or will.
I heard this from one of the greats in computer-hardware-evolution, only
I don't remember his name. What he said, and I say, is essentially this;
if you are part of an effort towards progress, in whatever field or
domain, you should have some understanding of WHERE you are going and
WHY you want to get there.
Arti Nigam
G Wilson
>replace AI is ludicrous. Consider the area of natural language. What
>conventional techniques that you know of can extract information from
>natural language text or translate a passage from English to French?
Errmmm...show me *any* program which can do these things? To date,
AI has been successful in these areas only when used in toy domains.
>The future of AI is going to be full of unrealistic hype and disappointing
>failures.
Just like its past, and present. Does anyone think AI would be as prominent
as it is today without (a) the unrealistic expectations of Star Wars,
and (b) America's initial nervousness about the Japanese Fifth Generation
project?
> But the demand for AI is so great that we have no choice but to
>push on.
Manifest destiny?? A century ago, one could have justified
continued research in phrenology by its popularity. Judge science
by its results, not its fashionability.
I think AI can be summed up by Terry Winograd's defection. His
SHRDLU program is still quoted in *every* AI textbook (at least all
the ones I've seen), but he is no longer a believer in the AI
research programme (see "Understanding Computers and Cognition",
by Winograd and Flores).
Greg Wilson
Boris N Goldowsky
In article <29...@sfsup.UUCP> sa...@sfsup.UUCP (S.Saal) writes:
Ironically, this makes AI a field that must make itself obsolete.
As more areas become understood, they will break off and become
their own field. If not for finding new areas, AI would run out
of things for it to address.
Isn't that true of all sciences, though? If something is understood,
then you don't need to study it anymore.
I realize this is oversimplifying your point, so let me be more
precise. If you are doing some research and come up with results that
are useful, people will start using those results for their own
purposes. If the results are central to your field, you will also
keep expanding on them and so forth. But if they are not really of
central interest, the only people who will keep them alive are these
others... and if, as in the case of robotics, they are really useful
results they will be very visibly and profitably kept alive. But I
think this can really happen in any field, and in no way makes AI
"obsolete."
Isn't finding new areas what science is all about?
Bng
--
Boris Goldowsky bo...@athena.mit.edu or @adam.pika.mit.edu
%ath...@eddie.UUCP
@69 Chestnut St.Cambridge.MA.02139
@6983.492.(617)
Rick Wojcik
>[re: my reference to natural language programs]
>Errmmm...show me *any* program which can do these things? To date,
>AI has been successful in these areas only when used in toy domains.
>
other companies, LUNAR, the LIFER programs, CLOUT, Q&A, ASK, INTELLECT,
etc. There are plenty. All have flaws. Some are more "toys" than
others. Some are more commercially successful than others. (The goal
of machine translation, at present, is to increase the efficiency of
translators--not to produced polished translations.)
>... Does anyone think AI would be as prominent
>as it is today without (a) the unrealistic expectations of Star Wars,
>and (b) America's initial nervousness about the Japanese Fifth Generation
>project?
>
I do. The Japanese are overly optimistic. But they have shown greater
persistence of vision than Americans in many commercial areas. Maybe
they are attracted by the enormous potential of AI. While it is true
that Star Wars needs AI, AI doesn't need Star Wars. It is difficult to
think of a scientific project that wouldn't benefit by computers that
behave more intelligently.
>Manifest destiny?? A century ago, one could have justified
>continued research in phrenology by its popularity. Judge science
>by its results, not its fashionability.
>
Right. And in the early 1960's a lot of people believed that we
couldn't land people on the moon. When Sputnik I was launched my 5th
grade teacher told the class that they would never orbit a man around
the earth. I don't know if phrenology ever had a respectable following
in the scientific community. AI does, and we ought to pursue it whether
it is popular or not.
>I think AI can be summed up by Terry Winograd's defection. His
>SHRDLU program is still quoted in *every* AI textbook (at least all
>the ones I've seen), but he is no longer a believer in the AI
>research programme (see "Understanding Computers and Cognition",
>by Winograd and Flores).
Weisenbaum's defection is even better known, and his Eliza program is
cited (but not quoted :-) in every AI textbook too. Winograd took us a
quantum leap beyond Weisenbaum. Let's hope that there will be people to take
us a quantum leap beyond Winograd. But if our generation lacks the will
to tackle the problems, you can be sure that the problems will wait
around for some other generation. They won't get solved by pessimists.
Henry Ford had a good way of putting it: "If you believe you can, or if
you believe you can't, you're right."
Gilbert Cockton
> The information age will revolutionize agriculture and industry just as
>industry revolutionized agriculture one hundred years ago
Sociologists study the present, not the future. I presume the "Megatrends" books
cited is Toffler style futurology, and this sort of railway journey light
reading has no connection with rigorous sociology/contemporary anthrolopology.
The only convincing statements about the future which competent sociologists
generally make are related to the likely effects of social policy. Such
statements are firmly rooted in a defendible analysis of the present.
This ignorance of the proper practices of historians, anthropologists,
sociologists etc. reinforces my belief that as long as AI research is
conducted in philistine technical vacuums, the whole research area
will just chase one dead end after another.
Thomas Maddox
>I think AI can be summed up by Terry Winograd's defection. His
>SHRDLU program is still quoted in *every* AI textbook (at least all
>the ones I've seen), but he is no longer a believer in the AI
>research programme (see "Understanding Computers and Cognition",
>by Winograd and Flores).
>
Using this same reasoning, one might given up quantum
mechanics because of Einstein's "defection." Whether a particular
researcher continues his research is an interesting historical
question (and indeed many physicists lamented the loss of Einstein),
but it does not call into question the research program itself, which
must stand or fall on its own merits.
AI will continue to produce results and remain a viable
enterprise, or it won't and will degenerate. However, so long as it
continues to feed powerful ideas and techniques into the various
fields it connects with, to dismiss it seems remarkably premature. If
you are one of the pro- or anti-AI heavyweights, i.e., someone with
power, prestige, or money riding on society's evaluation of AI
research, then you join the polemic with all guns firing.
The rest of us can continue to enjoy both the practical and
intellectual fruits of the research and the debate.
Thomas Maddox
>
>Sociologists study the present, not the future. I presume the "Megatrends" books
>cited is Toffler style futurology, and this sort of railway journey light
>reading has no connection with rigorous sociology/contemporary anthrolopology.
>
>The only convincing statements about the future which competent sociologists
>generally make are related to the likely effects of social policy. Such
>statements are firmly rooted in a defendible analysis of the present.
>
>This ignorance of the proper practices of historians, anthropologists,
>sociologists etc. reinforces my belief that as long as AI research is
>conducted in philistine technical vacuums, the whole research area
>will just chase one dead end after another.
"Rigorous sociology/contemporary anthropology"? Ha ha ha ha
ha ha ha ha, &c. While much work in AI from its inception has
consisted of handwaving and wishful thinking, the field has produced
and continues to produce ideas that are useful. And some of the most
interesting investigations of topics once dominated by the humanities,
such as theory of mind, are taking place in AI labs. By comparison,
sociologists produce a great deal of nonsense, and indeed the social
"sciences" in toto are afflicted by conceptual confusion at every
level. Ideologues, special interest groups, purveyors of outworn
dogma (Marxists, Freudians, et alia) continue to plague the social
sciences in a way that would be almost unimaginable in the sciences,
even in a field as slippery, ill-defined, and protean as AI.
So talk about "philistine technical vacuums" if you wish, but
remember that by and large people know which emperor has no clothes.
Also, if you want to say "one dead end after another," you might
adduce actual dead ends pursued by AI research and contrast them
with non-dead ends so that the innocent who stumbles across your
remark won't be utterly misled by your unsupported assertions.
Simon Brooke
(flaming against an article submitted by Gilbert Cockton)
> "Rigorous sociology/contemporary anthropology"? Ha ha ha ha
>ha ha ha ha, &c.
What do the third and subsequent iterations of the symbol 'ha' add to the
meaning of this statement? Are we to assume the author doubts the rigour
of Sociology, or the contemporary nature of anthropology?
>And some of the most interesting investigations of topics once dominated
>by the humanities, such as theory of mind, are taking place in AI labs.
This is, of course, true - some of it is. Just as some of the most
interesting advances in Artificial Intelligence take place in Philosophy
and Linguistics departments. This is what one would expect, after all; for
what is AI but an experimental branch of Philosophy?
>sociologists produce a great deal of nonsense, and indeed the social
>"sciences" in toto are afflicted by conceptual confusion at every
>level. Ideologues, special interest groups, purveyors of outworn
>dogma (Marxists, Freudians, et alia) continue to plague the social
>sciences in a way that would be almost unimaginable in the sciences,
Gosh! Isn't it nice, now and again, to read the words of someone whose
knowledge of a field is so deep and thorough that they can some it up in
one short paragraph!
It is, of course, true that some embarassingly poor work is published in
Sociology, just as in any other discipline; perhaps indeed there is more
poor sociology, simply because sociology is more difficult to do well than
any other type of study - most of the phenomena of sociology occurs in the
interaction between individuals, and this interaction cannot readily be
accessed by an observer who is not party to the interaction. Yet if you
are part of the interaction, it will not proceed as it would with someone
else...
Again, sociological investigation, because it looks at us in a
rigorous way which we are not used to, often leads to conclusions which
seem counter-intuitive - they cut through our self-deceits and hypocrisies.
So we prefer to abuse the messenger rather than listen to the message.
For the rest:
He who knows not an knows not he knows not......
A dictum which I will conveniently forget next time I feel like shooting
my mouth off.
** Simon Brooke *********************************************************
* e-mail : si...@uk.ac.lancs.comp *
* surface: Dept of Computing, University of Lancaster, LA 1 4 YW, UK. *
* *
* Thought for today: Most prologs chew everything very slowly anyway, *
***just being polite I guess*********************************************
Gilbert Cockton
>social "sciences" in toto are afflicted by conceptual confusion at every
>level. Ideologues, special interest groups, purveyors of outworn
>dogma (Marxists, Freudians, et alia) continue to plague the social
>sciences in a way that would be almost unimaginable in the sciences,
>even in a field as slippery, ill-defined, and protean as AI.
like Townsend and Halsey on Age, Poverty, Health and social mobility, you might
find something less concerned with theory and more with rigorous investigation.
I find the conflict in the humanities and behvioural "sciences" far more healthy
than the uncritical following of fashions of paradigms in science. Whilst the
former areas encourage an understanding of methodology and epistemology, the
sciences assume their core methods are correct and get on with it. A lot boils
down to personality (Liam Hudson, Contrary Imaginations). The reason that
ideology and methodological pluralism would be unimaginable in the sciences may
have something to do with the nature (and please, not the LACK) of the
scientific imagination compared to the humanist imagination. Note that
materialism, determinism, statistical inference and positivism are no less
outworn dogmas and ideologies than are Marxism, Freudianism, etc. My
experience is that someone from a humanist critical tradition will have a better
understanding of the assumptions behind methodologies than will scientists and
even more so, engineers. Out of such understandings came the rejection of first
Medieval Catholicism, then Seventeeth Century materialism, Twentieth Century
Behaviourism and Systems Theory, and now the "pure" AI position. Assumptions
behind AI are similar to many which have been around since the warm humility of
Renaissance Humanism cooled into the mechanical fascination of the Baroque.
>So talk about "philistine technical vacuums" if you wish, but
>remember that by and large people know which emperor has no clothes.
So who is it who is deciding strategy for most Western social programmes?
Clothes or no clothes, social administrators have an empire which extends
beyond academia and many of them draw on sociological concepts and results in
their work. It is in their complete ignorance of socialisation that AI workers
fall down in their study of machine learning. Most human learning always takes
place in a social context, with only the private interests of marginal
adolescents and adults taking place in isolation - but here they draw on problem
solving capabilities which were nutured in a social context. The starkest
examples of the nature and role of primary socialisation come from those few
unfortunate children who had been isolated from birth. They are savage animals.
If parents had to interact with their children in FOPC or connectionist inputs,
the same would be true, until the children were taken into care.
>Also, if you want to say "one dead end after another," you might adduce actual
>dead ends pursued by AI research and contrast them with non-dead ends.
DEAD ENDS
Computational Lingusitics, continuous speech understanding, intelligent vision,
reliable expert systems which do not require endless maintenance, human
problem solving, the physical symbol system hypothesis, knowledge representation
formalisms using computable models. Largely areas where some other paradigm
within another discipline can make progress as the lead weight of computability
is not suffocating research. Generally due to knowledge representation problems
- even the Novel has problems here :-) If you can't write it in a text-book
(e.g. clinical diagnosis, teaching techniques, advocacy), you'll never get it
on a machine - impossible in superset (NL) => inpossible in subset (FOPC,
computationally denotable/constructable). A problem in AI is trying to solve
other people's problems, where those other people know more about the problem
than you ever will - they live it day in day out.
NON-DEAD ENDS
Much work done under the name of AI is good - low-to-medium level vision,
restricted natural language, knowledge-based programming formalisms,
theorem-proving and highly-constrained technical planning problems. Indeed,
most technical knowledge, being artificial and symbolic from the outset, is an
obvious candidate for AI modelling and there is nothing in the humanist
tradition which would doubt the viability of this work. Here knowledge
representation is easy, because the domain will generally be so boring (but
economically/environmentally/security critical) that no-one wants to argue
about it. Much technical expertise executed by humans is best suited to
machines. In HCI research, sensible work on intelligent (=supportive) user
interfaces is getting somewhere, but then coming up with a computer model of a
computer system is hardly a major challenge in knowledge representation
techniques. Coming up with a computer model of a user is also possible, as long
as we don't try to model anything controversial, but stick to observable
behaviour and user-negotiated input.
The main objection to AI is when it claims to approach our humanity.
It cannot.
G.Gleason
>The main objection to AI is when it claims to approach our humanity.
> It cannot.
That's a pretty strong claim to make without backing it up.
I'm not saying that I disagree with you, and I also object to all the
hype which makes this claim for current AI, or anything that is likely
to come out of current research. I'm also not saying your claim is
wrong, only that it is unjustified; there is more to learn before we
can really say.
There are new ideas in biology that build upon "systems theory," and
probably can be tied in with the physical symbol systems theory (I
hope I got that right) that suggest that information or "linguistic
interaction" is fundamental to living organisms.
In the May/June issue of "The Sciences," I found an article called
"The Life of Meaning." It was in a regular column (The Information Age).
I won't summarize the whole article, but it does present some compelling
examples, and arguments for extending the language of language to talking
about cellular mechanisms. One is how cyclic AMP acts as an internal
message in E. coli. When an E. coli lands in an environment without
food, cyclic AMP binds to the DNA, and switches the cell over to a
"motion" program. Cyclic AMP in this role has all the attributes of
a symbolic (or linguistic) message: the choice of symbol is arbitrary,
and the "meaning" is context dependant. This becomes even more clear
with the example of human adrenaline response in liver cells. The
hormone binds to sites on the outside of the cell which causes an
internal message to be generated, which just happens to be cyclic AMP.
The cell responds to the cyclic AMP (not by a DNA based mechanism as
in E. coli) by producing more glucose. The composition of the message
has nothing to do with the trigger or the response, it is symbolic.
So, how is this relevant to the original discussion. I don't see any
fundamental difference between exchanging chemical messages or electronic
ones. Although this does not imply that configurations of electronic and
electromechanical components that we would call "alive" are possible or
that it is possible to design and build one, it doesn't rule it out, and
more importantly it suggests a fundamental similarity between living
organisms and "information processors." The only difference is how they
arise. Possibly an important difference, but we have no way to prove this
now.
Gerry Gleason
Jeff Dalton
>of nonsense, and indeed the social "sciences" in toto are afflicted by
>conceptual confusion at every level. Ideologues, special interest groups,
>purveyors of outworn dogma (Marxists, Freudians, et alia) continue to
>plague the social sciences in a way that would be almost unimaginable in
>the sciences, even in a field as slippery, ill-defined, and protean as AI.
Speaking of outworn dogmas, AI seems to be plagued by behaviorists,
or at least people who seem to think that having the right behavior
is all that is of interest: hence the popularity of the Turing Test.
>Also, if you want to say "one dead end after another," you might
>adduce actual dead ends pursued by AI research and contrast them
>with non-dead ends so that the innocent who stumbles across your
>remark won't be utterly misled by your unsupported assertions.
Does anyone actually think the current techniques are capable of
producing human-level intelligence just by scaling up? They are all
likely to be dead ends in that sense though they may well be useful
for something else.
Jeff Dalton, JANET: J.Da...@uk.ac.ed
AI Applications Institute, ARPA: J.Dalton%uk.a...@nss.cs.ucl.ac.uk
Edinburgh University. UUCP: ...!ukc!ed.ac.uk!J.Dalton
Barry Shein
Re: the social sciences, AI etc...
The important event that has happened in psychology in the past twenty
or so years is the movement from a descriptive model (eg. poke a brain
with a stimulus like a question or a light to react to, record the
result, try to fit it into a statistical model and relate the repeated
results to other variables) towards a functional model (try to build a
machine which exhibits the same behavior as a mind on the assumption
that this can provide insight into how the mind must work.)
In many ways I think this is critical to psychology becoming a true
science, probably an engineering science as well. It was the movement
from observing it externally to the development of models. Just as
physics went from sitting and watching things move and developing
stories about why they might do that to producing mathematical and
other models which predict and model the behavior observed.
At some point we might be able to make such a paradigm shift in the
other social sciences. I don't know whether or not it is critical to
view something like a society as the sum of its individual minds and,
thus, you must first understand the mind to understand the interaction
of many minds.
For example, one did not need to know atomic physics to write down a
useful theory of mechanics. It has been helpful to grind out the noise
and make more accurate models (eg. molecular models of friction no
doubt make our theories of mechanics more accurate, but they were
hardly necessary to basically understand the principals of a ball
rolling down a hill.)
Although there is little doubt that our social sciences are in their
fetal stages (ie. their methodologies probably have to undergo radical
shifts) I believe that by being able to use computers to build
functioning models to study we may be getting a glimpse of what that
future methodology will have to be.
Simulation is the new mathematics of science. Computers are its
pencil and paper.
-Barry Shein, Boston University