I've just read the paper Better Computer Go Player with Neural Network and Long-term Prediction which seems to make a big step in direction of getting a better algorithm playing Go than any human.
Now I was wondering how the game could be changed to something which is well-defined and where humans are still much better than computers.
Does a Go variant on an infinite board exist?
One would have to play it with a computer, of course. Having such an infinite board (of which only the parts which were looked at are calculated, of course) isn't really a programming problem.
As the boundaries seem to be important, one could add random borders to the game. The winning criterion could be an absolute / relative difference in points. What would be the implications of such a game?
Another idea would be to make a (pseudo-)fractal board.