I used to build tic tac toe games It was never complete or useful but it helped me learn the tech I was using. Now days creating a blog engine is my new tic tac toe.
What was so cool about it was watching it learn, and realizing that unless I programmed it to make random mistakes, it would settle for sub-optimal strategies.
If you haven't already, check out _Essentials of Metaheuristics_ (http://cs.gmu.edu/~sean/book/metaheuristics/), a free textbook / set of lecture notes. :)
Not necessarily. You could get much the same effect (the program learning by playing itself over and over) with many Reinforcement Learning algorithms (like TD learning, say).
There are major differences between Value Function (loosely, perceived_state -> perceived_long_term_reward map) based RL algorithms and algorithms that work only in the policy (loosely, the perceived-state -> chosen_action map) space like Simulated Annealing (or Genetic Algorithms). Barto and Sutton (somewhat loosely) use the term "evolutionary algorithms" to distinguish Value Function based algorithms from those that only manipulate policy
By then you should know a good chunk (IO, loops, built in data structures ...)
o | o | o ---------- x | x | x ---------- o | o | o
That (if HN doesn't mangle it) should be a fully working Tic-Tac-Toe board. You can store it as a string, hashtable, whatever you want. You could label the squares 1-9 for the prompt asking the user where to move.
This is all doable with perl and a command line. Try it out, see what you learn.
o | o | o
---------
x | x | x
---------
o | o | o"The good thing about reinventing the wheel is that you can get a round one."
Douglas Crockford -- creator of JSON