Re:Google's Plop is implemented in Lisp
by cabalamat on Thursday November 13, 2008 @05:10PM
(#363657)
You're right that most possible programs are useless and uninteresting. So I agree with you that if a learning program was given problems to solve that are generated by generating random programs, that wouldn't be interesting.
I am not suggesting doing that. What I am suggesting is that the problems-to-solve be generated by humans.
Incidently what I am suggesting isn't quite inductive programming but something slightly more general, in that the job of the problem-solver isn't to generate a program that solves the input/output pairs, but to guess what the next output will be depending on its input. (The problem-solver may or may not do this by internally creating a program that solves the problem).
Anyway I've done a quick write-up of what I propose: http://www.includipedia.com/wiki/User:Cabalamat/Function_pre...
To do AI you have to realize that in the sentences: "Humans have general intelligence and can solve problems in many domains" and "Breadth-first search is a problem-general technique that will always find a solution if one exists" the word "general" does not mean the same thing at all! (cf. the huge literature on inductive bias in human cognition).
For a nice list of program induction problems solvable by search (with a system called ADATE), see http://www-ia.hiof.no/~rolando/Examples/index.html .