AIList-REQUEST@SRI-AI.ARPA (AIList Moderator Kenneth Laws) (11/18/85)
AIList Digest Monday, 18 Nov 1985 Volume 3 : Issue 172
Today's Topics:
Queries - Fictional Machines That Talk & Object-Oriented Programming,
AI Tools - Typed Languages and Lisp,
Psychology - Prejudice,
Inference - Abduction & Translating Representations,
Intelligence - IQ Test for AI,
Review - TI's Satellite Symposium and AI Hype,
News - RCA Chair at Penn
----------------------------------------------------------------------
Date: 15 Nov 85 16:17 PST
From: Halvorsen.pa@Xerox.ARPA
Subject: Fictional accounts of machines that talk
I am looking for references to early (or ideally the earliest) mention
of machines with natural language capabilities in fiction or film.
------------------------------
Date: Wed, 13 Nov 85 12:09
From: Nick Davies (at GEC Research) <YE85%mrca.co.uk@ucl-cs.arpa>
Subject: Object oriented programming in Common Lisp
Does anyone have or know of an implementation of Flavors or any other
object-oriented programming system in Common Lisp ?
Nick Davies (ye85%uk.co.mrca@cs.ucl.ac.uk).
Thanks in advance
------------------------------
Date: Sat, 16 Nov 85 14:09:10 gmt
From: Don Sannella <dts%cstvax.edinburgh.ac.uk@ucl-cs.arpa>
Subject: Re: Typed languages and Lisp
>From: Skef Wholey <Wholey@C.CS.CMU.EDU>
A type checker can find some bugs, but it isn't clear that such bugs would
take much time to find and fix relative the to the "real" bugs a programmer
spends most of his time on. Also, actually entering type information can
add to program development time. Controlled experiments are required ...
I have been programming in HOPE and ML (both of which have the same strong but
flexible type system, due to Robin Milner) for about seven years. I do not
know of any controlled experiments which show that strong type checking
decreases program development time; I can only report that in my experience
using these languages, the type checker catches so many bugs (I would guess
far in excess of 95% of non-syntax errors) that programs really do usually
run correctly the first time they get past the type checker. I have heard
other people who use these languages say the same thing, and I haven't met
anybody who has really tried to use them complaining that entering type
information doesn't pay off. (This is especially true of ML, where the
compiler is able to infer types of functions so that a programmer is not
required to say much about types at all.) I shudder to think of how much
time I would have wasted in LISP trying to track down some of the (often
subtle) type bugs the compiler has caught for me!
Disclaimer: I haven't tried writing things like operating systems or compilers
in a language like HOPE or ML (although other people have done so), so I don't
know how much the strong type system gets in the way in cases like these. My
experience is with rather mathematically-oriented programs of 1000 lines or so.
------------------------------
Date: Thu, 14 Nov 85 22:32:12 PST
From: Richard K. Jennings <jennings@AEROSPACE.ARPA>
Subject: Removing Prejudice
Concerning removing prejudice, let me suggest reading "The Nature of
Prejudice" by Gordon Allport and Patrick Monihan (then a Prof at Harvard).
I think modeling prejudice accurately is the mandate for AI systems, not
removing it.
Rich.
------------------------------
Date: Fri, 15 Nov 85 03:16 EST
From: jcma@MIT-MC.ARPA
Subject: Abduction Makes The Big Time
See Charniak and McDermott, Introduction to AI, Addison-Wesley, 1985 for
several big sections on abduction.
------------------------------
Date: 15 Nov 85 10:27 PST
From: Shrager.pa@Xerox.ARPA
Subject: Abduction
Isn't this what *all* script and schema inference systems do?
Isn't this what *all* pattern recognition systems do?
In fact almost every AI system that is not proof-based does something
like what you are calling abduction. I can't understand why anyone
thinks that this is news, other than having a fancy new name for what
we've been doing for years.
-- Jeff
------------------------------
Date: 17 Nov 85 21:18 PST
From: Shrager.pa@Xerox.ARPA
Subject: On translating representations
"[...] for psychological reasons, in order to guess new theories,
[mathematically equivalent theories] may be very far from equivalent,
because one gives a man different ideas from the other. [...] every
theoretical physicist who is any good knows six or seven different
theoretical representations for exactly the same physics. He knows they
are equivalent, and that nobody is ever going to be able to decide which
one is right at that level, but he keeps them in his head, hoping that
they will give him different ideas for guessing [new theories]."
From: Feynman, R (1965). The Character of Physical Law. MIT press,
Cambridge, Mass. pp. 168.
------------------------------
Date: Thu, 14 Nov 85 10:39:47 est
From: Geoff Loker <gkloker%utai%toronto.csnet@CSNET-RELAY.ARPA>
Subject: Re: IQ test for AI. (AIList Digest V3 #164)
> I don't remember anyone suggesting to test AI programs with the
> traditional human IQ test. [...]
> Has any one tried to write such a computer program ?
> Rene Bach (Bach@score)
Thomas Evans' PhD thesis (MIT 1963) was concerned with the solution
by machine of so-called "geometric-analogy" IQ test questions. The
program he developped was called ANALOGY, and it apparently did quite
well with the "A is to B as C is to __" picture problems.
A revised form of his thesis can be found in "Semantic Information
Processing", edited by Marvin Minsky, published 1968 by MIT Press.
Also, the second edition of Winston's "Artificial Intelligence"
(1984) refers to Evans' work and may have pointers to follow-up
work as well.
------------------------------
Date: Tue, 12 Nov 85 12:04:21 EST
From: Jakob Nielsen <nielsen.yktvmv%ibm-sj.csnet@CSNET-RELAY.ARPA>
Subject: Re: Future IQ
I have actually for some time used the term "AIQ"
(Artificial Intelligence Quotient) as a shorthand when discussing
computers (e.g. "this system has an incredibly low AIQ").
I don't think that it would have any meaning to use human IQ tests
on computers however - just understanding the questions themselves
(without trying to *answer* them) would require a higher AIQ than
normally seen these days.
Of course in 2085, who knows ...
------------------------------
Date: Fri, 15 Nov 1985 03:39 EST
From: "David D. Story" <FTD%MIT-OZ @ MIT-MC.ARPA>
Subject: TI's Satellite Symposium & AI Hype
I went to this thing. It was terrible. The technical content
was zilch. I also question the way that "expert systems" and
rule-production systems are lumped together. Expert systems
are defined as systems that can handle areas of knowledge in
the manner that a human expert does. They may or may not be
rule-production systems. Rule Production systems may not
exhibit even vaguest qualities of a human expert.
This type of lumpage will only serve to confuse the general public
and lend the credence of the speakers and TI to real consumer
fraud. There is not even mention of the performance criteria that
originally generated the name of "EXPERT SYSTEM". This is further
solidified by a apprx. 25 year old student saying that he wrote an "EXPERT
SYSTEM" with a couple of rules. If I say that you have a gross income
of over $6000 dollars and you have to file, does that make me an
EXPERT. I think not.
There is no mention of the degree of expertice of such systems should
possess before being named even decision support tools.
What compounds this was I attended the Press viewing of the symposium
and there was no one there to field guestions. Most novices come with
a pre-concieved idea of AI and leave being more reinforced with
errant notions.
This feeds into Geoff Goodfellow's previous posting on this list. The
furtherence of this hype is just another blackeye AI researchers are
going to have to bear when it comes to how much funding AI is
receiving. Generating interest by speculation is one thing - opening
the doors for real consumer fraud other !
Dave Story
FTD@MIT-OZ
------------------------------
Date: Mon, 18 Nov 85 02:03 EST
From: Tim Finin <Tim%upenn.csnet@CSNET-RELAY.ARPA>
Subject: RCA Chair at Penn
RCA ESTABLISHES AN AI CHAIR AT THE UNIVERSITY OF PENNSYLVANIA
The University of Pennsylvania is pleased to announce the establishment
of the RCA Professorship in Artificial Intelligence in the Department
of Computer and Information Science. Established through the
generosity of RCA, the chair is intended to enhance the University's
position as a leading center for research and education in artificial
intelligence.
The University's tradition of pioneering work in the computing field, from
the development of ENIAC, the offering of the first computer course, and the
graduation of the first PhD in computer science, manifests itself today in a
strong focus in the Computer and Information Science Department on
artificial intelligence. The department is involved in a variety of research
programs at the forefront of the field, including work in natural language
processing, knowledge representation and reasoning, expert systems, computer
vision and robotics, computer graphics, parallel processing, programming
languages and environments, computer architecture and systems, theory of
computation, software engineering, and database systems. These research
programs involve interactions with the Departments of Electrical and
Mechanical Engineering (in sensors and robotics), the Department of Chemical
Engineering (in expert systems), the Departments of Linguistics, Philosophy,
and Psychology (in cognitive science, in particular, in the computational
aspects of language and perception), the Wharton School (in database
systems, expert systems, and software engineering), and the Medical School
(in expert systems and image processing).
The person appointed to the RCA Professorship will be a distinguished
scholar who has performed outstanding research in artificial intelligence
with both theoretical and practical significance, especially in the areas of
knowledge representation and reasoning with potential applications to the
new generation of expert systems technology.
------------------------------
End of AIList Digest
********************