[ The
Problem Solver | The Math Explorer | The General Discoverer ]
ne
of the mark of intelligence is the ability to acquire and sort knowledge to produce a
coherent paradigm of the world by oneself. One way to do this is to explore through
what is known to discover what is unknown, an approach that has been adapted by numerous
AI programs. A few famous ones are discussed here.
After developing Logic Theorist, its limitations soon became evident to creators Allen
Newell and Herbert Simon--two men who came to believe that any type of decision-making can
be broken down and guided by rules called heuristics.
Simon expressed the two's realization that Logic Theorist could only solve very specific
problems that had to be stated in a way that directed its decision-making process on what
to do(i.e. tasks). On the other hand, people use independent means to extract the
basic problem and use their own heuristics to develop original strategies and tasks to
solve the problem.
This is where General Problem Solver(GPS) came into the picture in 1957. Again,
with heuristics guiding the decision-making processes, GPS employed a strategy called
"means-end analysis" in which the computer essentially determines the difference
between the goal and the current state of the situation and then develops tasks that will
narrow that difference. One problem GPS was given to solve was to make the computer
think it was a monkey who wanted to get to a banana too high up. Also within the
situation, the "monkey" was given a chair and basic functions for actions like
"move self," "climb on chair," "move chair," and
"jump." With that knowledge, GPS was on its own able to figure out a way
to get the banana.
GPS first tried out its actions to see if it would reduce the difference of the goal
and the current situation. After testing out several other possible paths towards
achieving its goal, GPS soon recognized the obstacles it needed to overcome--i.e.
sub-problems--before it could achieve its primary goal. Soon, it began developing
tasks like getting to the chair in order to move the chair--a consequence of means-end
analysis. In end, the monkey did get his banana!
GPS would go on to solve various puzzles, break secret codes, and even do symbolic
integration. Other programs that employed strategies similar to GPS's means-end
analysis and heuristics like the Geometry Theorem Prover allowed computers to start
reasoning more like humans.
Another problem that used heuristics to reason was Douglas Lenat's Automated
Mathematician(AM). What was different about AM was that instead of being given a
problem to solve like GPS, it was given only a few simple abilities and knowledge in
mathematics explore the realm of numbers. Some of AM's abilities and their results
were:
- The starting information it has to work with is stored in lumps of related information
called frames. From there, AM can create and
organize new frames to store the new knowledge it learns.
- The second ability AM started with is the sense of "play," or what people
describe as curiosity. It was able to try out certain actions like finding the
opposite of something or find the extreme cases of that thing of interest. So when
it discovered addition, it explored and later found subtraction. When it discovered
integers that only divided evenly with only a few divisors, it looked towards numbers with
extremely few divisors and discovered prime numbers.
- To control what AM explores in mathematics before it got lost in the sea of logic paths,
it was given a sense of aesthetics. Using 59 heuristics, AM could assign a value
from 0 to 1,000 that denoted the importance or relevance of a discovery and then decided
whether it should proceed in particular direction of exploration or not.
AM proved to be an important step in AI research because not only did it find out and
could prove mathematical concepts from simple arithmetic to traditional number theory to a
concept called composite numbers. However, because of the limits of its original
heuristics, AM skipped over many important mathematical ideas like remainders and largest
common denominators and soon became lost in thought without producing much results as it
was did when it first started. Lenat recognized AM's limited self-modification
abilities and soon proceeded to build Eurisko.
Working as a teacher at Carnegie-Mellon University, Lenat started developing a better
version of AM with Eurisko(from the Greek word meaning "I discover").
Eurisko was endowed with similar abilities as AM had, but the new program
could modify the heuristics that governed how it learned. Because of the limitations
of the programming language Lenat was using, LISP, he
wrote his own language called Representation Language Language(RLL). RLL allowed
Lenat to store whole heuristics into frames like AM did with knowledge so that it could
efficiently edit its heuristics such as when Eurisko found a rule that could do the same
thing as two other rules in less time.
Eurisko's self-modification abilities allowed it to work on more general topics like
circuit design as well as play a space-war strategy game called Traveler TCS. Of
course, Eurisko's supreme playing and learning abilities made the program a champion by
discovering that making an unorthodoxally small, fast ship would render it invincible and
thus at least produce a draw during battle. Though the organizers of Traveler TCS
changed the rules so that Eurisko couldn't employ that fool-proof strategy again, the
computer once again found a way to be the best and won a second time the next year. After
that, the organizers threatened to cancel the entire competition if Eurisko was allowed to
enter again for the third year, thus Lenat decided to bow out.
So who deserved credit for Eurisko's accomplishments? The computer or the
inventor? Lenat feels a little over half of the credit should go to him while the
rest should go to Eurisko since the computer would not have accomplished what it did
without Lenat's creativity while Eurisko discovered things that Lenat would not have been
able to do. Clearly, AI was becoming a mind of its own in his eyes.
