General intelligence as Transfer; automatic programming; meta-level transitions

[gusl]

General intelligence means being able to transfer skills and knowledge to similar or analogous tasks ("transfer", when used as a noun, is a measure of success for education). In some cases, it means being able to deal with context.

Lisp is wonderful. My programs can't program themselves yet, but we're getting there...

Agents with high "general intelligence" can do the "meta-level transition" and start programming what's on their mind. My ideas are still vague.

What is it that I have that my programs don't when I read a specification and start implementing it? If we translated "easy" programming puzzles into a formal language, would current AIs tackle them successfully? If not, why not?

Comments

[jbouwens.livejournal.com]

Formal specifications are completely impractical for all but the smallest of programs.

And yes, if you would make a formal specification of an "easy" programming puzzle, I think it would be _possible_ to automatically translate this into an implementation, even without AI. This doesn't mean it would be an efficient implementation, only that it would work.

Just out of curiousity, can you give an example of a formal specification for some small, easy problem?

[gustavolacerda.livejournal.com]

I'm looking for an example where the formal specification is far from the implementation.... the trouble is that Lisp brings the two things so close...

But you can think of sorting as having a trivial specification, and yet a non-trivial implementation... especially if you have an average efficiency requirement to meet.