On the other hand, if I knew enough math to be able to solve unsolved problems, I would not spend my time on exercises where I can look up the solution if I don’t know further. My point was that you do know enough about programming to do things that other people have not done before.

Well it is obvious, isn’t it? It is educational and fun. Programming tickles my mind and is the substitute for the pleasure I sadly can’t derive from math.

I just implemented a telnet client in python, ignoring the built in module “libtelnet”. It works. I won’t use it. But it was educational. Today, I also got 11 out of 12 in my math midterm exam, which was not half the fun. I envy you for enjoying something so pure as math. I can’t (yet). But I might be getting there. I’m certainly already beyond a point of no return.

(I’ve been doing LISP for two months now)

However, the question is not whether I prefer math to programming in general, but which tool is most appropriate for what I want to do. You might as well have asked “How did your goals change?” and I would have quoted two paragraphs from my current “about me” page:

“I don’t want to find the shortest or fastest path towards artificial intelligence. I want to map the territory. I want to know what’s learnable, and how, and which limits there are for an intelligent agent’s ability to improve solely by learning from its interaction with the environment.

I don’t consider my work engineering intelligence, I think of it as a kind of metaphysics. Physics is about learning about the structure of our environment. Learning theory is about what intelligent agents can learn about arbitrary environments in principle.”