SIAM's annual DiPrima Prize of $1,000 will be given to a
young scientist based on an outstanding doctoral dissertation
in SIAM-relevant applied mathematics (including theoretical
computer science). Send nominations by 12/31/95 to Richard
C. DiPrima Prize, c/o Donna Blackmore, Society for Industrial
and Applied Mathematics, 3600 University City Science Center,
Philadelphia, PA 19104-2688. [Zvi Galil ,
info.theorynt, 4/18/95.] ("Young scientist" means a PhD completed
from 7/1/93 to 6/30/95 and awarded by 12/31/95.)
Robert Epstein reported on 3/13/95 that the Loebner Prize
Competition is in disarray due to changes requested or insisted
upon by Dr. Hugh Loebner . Loebner's 4/21/95
announcement of the competition rules confirms several of the
changes. (Applications are due by 11/1/95.) Loebner is
requiring audio and visual input, with the exception of this
year's competition, and that all programs run on-site.
No telecommunications or online databases will be permitted.
(Last year's winner used text retrieval to answer questions
about sexual matters, an approach called "canned intelligence."
Loebner apparently sees no theoretical AI advance in retrieval of
human-written text, but is currently willing to permit CD ROM
drives or large disc packs for the unlimited-domain test.)
According to Epstein, Loebner further insists that either
he or his representative must be fooled by the winning entry
if it is to win the grand prize. (I see no mention of this in
this year's announcement.) Epstein has directed the Loebner
Prize Competition since it began in 1990. He and the Prize
Committee have unanimously rejected Loebner's suggestions as
being unfair to many contestants (e.g., graduate students using
supercomputers or custom hardware) and not in the spirit of
Turing. Most former competitors are also refusing to enter.
Loebner appears to be proceeding on his own, and will award a
$2,000 prize if even one program competes. The Prize Committee
has submitted a formal protest to the Board of Trustees of the
administering research institute, and is seeking a new underwriter
for the event; an annual Turing Test is still expected.
[, comp.ai, 3/13/95. David Joslin.]
Louis Savain supports Loebner's changes,
saying they will speed the development of truly intelligent
computers based on general learning theories. ASCII I/O makes it
too easy for an "AI priesthood" to win with "intricate toys."
Savain considers Turing's ideas obsolete, and wants a test
that requires autonomous learning. [comp.ai, 4/17/95.]
(The priesthood comment is at odds with Minsky's complaints
about the competition, and rebuttals that "anyone" can win based
on demonstrated competence rather than theory or connections.
And I suspect that most human intelligence is canned -- via neural
adaptation and case-based reasoning -- rather than derived from
first principles. That's why it took us so long to invent the
plow and other artifacts of civilization. We're just very good
at patching together past partial solutions in ways that conceal
their origins and imprecision. But Loebner and Savain may be
right that the rules should keep changing to defeat dead-end
optimizations. What, for example, have we really learned from
chess machines other than efficient alpha-beta search?
Advances in learning, reasoning by analogy, and reasoning from
first principles are worthy of encouragement, whereas the
"fool some journalists for a few minutes" criterion is not
-- as Minsky and others have repeatedly pointed out.
Neither is "the spirit of Turing" sacrosanct, as few in AI
or philosophy have ever accepted the imitation game to be a good
test of computational intelligence. (It may be sufficient but
is not necessary. We just haven't come up with anything better.)
The question here is whether the original rules should be
abandoned -- harming some current and future contestants -- or
whether Loebner should take his lumps and put up a second $100K
prize for his new vision. And, within the new competition,
what rules and metarules will best lead step by feasible step
to intelligent machines. The on-site requirement is silly,
and visual input seems too big a jump right now -- although
Microsoft and OGI are seeking researchers for such work.
And even Loebner's new format discriminates against programs
that compose music, create fine art (e.g., AARON), write stories,
or solve other problems at or beyond the limits of human