Generating random programs/algorithms to solve a given problem.
Honestly, though, I haven't put much thought into this, so I'm not sure how
feasible it would be. If I had to undertake this task, I'd probably try to
generate code for a functional language with zero side effects (so variable
assignments don't exist, and so variables don't exist, and so I don't have
to worry about checking that a variable is declared before it's used, for
example).

   An easy mutation scheme would be to select a random point in the tree,
discard everything below it, and generate a new subtree branch. The problem
with that is that the degree of "mutancy" depends on how high up in the tree
the random point is chosen, and depending on the grammar, there may be zero
randomness involved in selecting the children below a certain point. I'm not
sure how one would implement cross-over with trees, however. Did you
implement cross-over for permutations, and if so, how did you do it?

   - Oliver