gary@rochester.UUCP (Gary Cottrell) (03/09/84)
From: Gary Cottrell <gary>
>From seismo!harpo!decvax!decwrl!rhea!orphan!benson Fri Mar 2 20:24:18 1984
Received: by sen.rochester (3.327.3N) id AA12246; 2 Mar 84 20:24:13 EST (Fri)
Date: Thursday, 1 Mar 1984 13:45:43-PST
From: seismo!decvax!decwrl!rhea!orphan!benson
Subject: Re: Seminar Announcement
To: decvax!harpo!seismo!rochester!gary
Status: R
29-Feb-1984
Garrison W. Cottrell
University of Cottage Street
55 Cottage Street
Rochester, New York 14608
Dear Mr. Cottrell:
Although I was unable to attend your recent seminar, "New
Directions in Connectionist Dog Modeling," I am compelled to
comment on your work as presented in your published works, along
with the new ideas briefly discussed in the seminar announcement.
Having read your "Dog: A Canine Architecture" in late 1981, I
approached "Toward Connectionist Dog Modeling" the following year
with cautious optimism. The former work encouraged me that perhaps
a consistent dog model was, in fact, obtainable; at the same time,
it caused me to wonder why it was desirable. Nontheless, "Toward
Connectionist Dog Modeling" proved to be a landmark in this
emerging science, and my resulting enthusiasm quieted those nagging
suggestions of futility.
You may not be familiar with my work in the field of artificial
ignorance, which, I would like to suggest, shares several goals
with your own work, with different emphasis. "Artificial Ignorance
- An Achievable Goal?" (Benson 79) was the first of my published
papers on the subject. Briefly, it promoted the idea that although
creation of an "artificially intelligent" machine is a worthy
scientific goal, design and implementation of an "artificially
ignorant" one is much more sensible. It presented several
arguments supporting the notion that, compared to artificial
intelligence, artificial ignorance is easily achievable, and is
therefore the logical first step.
As a demonstration of the power of artificial ignorance (AI), I
spent the latter half of 1979 producing CHESS1, a chess system for
the VAX-11/780. CHESS1 was written primarily in LISP, a language
of my own invention (Language for Ingorance Simulation
Programming). In a resounding victory, CHESS1 lost to even the
most ignorant human players, being unable to distinguish between
the pieces. CHESS2, a more sophisticated implementation completed
in April of 1980, lost just as effectively by moving the pieces in
a clockwise progression around the edge of the board.
Ignored by overly ambitious, grant-hungry researchers, artificial
ignorance seemed to become my own personal discipline. After only
three issues, the fledgling SIGIGN newsletter was discontinued, and
the special interest group it served was disbanded.
Undaunted, I published a series of three papers in 1980. The first
two described several techniques I had developed toward simulating
ignorant behavior ("Misunderstanding Human Speech", and "Pattern
Misidentification", Benson 80). The third presented a simple
conversion method for producing artificially ignorant programs from
artificially intelligent ones, using a heuristic bug insertion
algorithm ("Artificial Brain Damage", Benson 80).
Despite these technical triumphs, interest in AI seemed to be
dwindling. By the spring of 1981, I, too, had lost interest,
convinced that my AI research had been little more than an
interesting intellectual exercise.
It is for this reason that your dog modeling thesis so thoroughly
captured my interest. Surely the phrases (to quote from your
announcement) "impoverished phoneme," "decimated world view," and
"no brain" imply "ignorance." And, if I may paraphrase from your
original treatise, the generic dog is essentially the equivalent of
an intellectually stunted human who has been forced to bear fur and
eat off the floor.
Clearly dog modeling and AI have much in common. To prove my
point, I have simulated the Wagging Response in a LISP application,
and am working toward a procedural representation of the Tail
Chasing Activity. The latter is a classic demonstration of genuine
ignorance, as well as a natural application of recursive
programming techniques.
I welcome any suggestions you have on these experiments, and look
forward to the continued success of your dog modeling research.
Sincerely,
Tom Benson