What's a fair competition between
a person and a computer?
Goal: to make the "user" believe it's human.
Intermediary hides irrelevant appearances.
Criterion for success is a statistical
one:
The computer is intelligent
if, on the average, it wins 50%
of the
time.
How does it reason?
-- logical inference?
plausible reasoning?
In what way does it learn?
-- acquisition of
facts? fine-tuning of skills?
What kind of perception is
it doing?
-- vision, speech
understanding, reading, feeling?
How does its perception work?
-- exact matching
of patterns? structural analysis?
Does it use and understand
human language?
-- can it hear, read,
take commands?
Can it solve problems?
-- if so, must the
problem be already clearly posed?
Does it possess knowledge?
-- in what form is
the knowledge represented?
acquisition of
facts
inducing rules
from examples
building skills
forming a library
of plans for reaching goals
building semantic
networks to represent situations
Learning needs to be measured
relative to a criterion function,
or function
of merit,
f : states --> R
Let s1 and s2
be the system's states at times t1 and t2, respectively.
The system learns
during [t1, t2] with respect to f provided
f(s1) < f(s2)
and the change of state occurs because of the system's
information
processing activity between t1 and t2.
Poor problem solving ability
does not necessarily mean
that a system lacks intelligence....
The intelligence might reside in its NLU ability.
Or it may be
good at recognizing your face in
a video image.
Hybrid systems often are best, combining
-- intelligent processing,
-- large, high-quality knowledge
and data sources,
-- lots of computing cycles
Last modified: September 30 1998
Steve Tanimoto