![]() StartingPoints Referenced by
|
Writing A Go Program To Teach Beginners
(inspired by Jan De Wit's thoughts about why the HikaruHandheldGame plays so badly) OK, computers can't play Go properly. At least not in the coming twenty years or so. But could they play teaching Go games? I'm sure most of the people here have gone through the motions of explaining why a group with only two connected liberties is dead (and what dead means), why the life of three liberties in a row depends on whose move it is, exploring the various possibilities for four and so on. This is standard knowledge, required for everyone and every program: if you don't know as much as your opponent about life and death, you probably lose. The teaching process is the same, regardless of country. This makes me think that there are at least some forms of Go knowledge that can be imparted without human intervention such as: which shapes are dead, play in the corners first, count count count, too many diagonal moves are bad. Why not have a computer do that? But don't do it blindly - adapt to the level of the pupil, by having the program say 'You spotted that right! But what happens in the corner?' to someone who has demonstrated her abiltiies at killing a 3x2 group. I know that it is impossible to react to vague inputs like 'I thought it is was a good move', but since Go is basically a multiple choice test with reasons, you could narrow the pupil's idea space down to a few cases. All of the above can be summed down to: 'You don't know anything about something until you can explain it to somebody else, and a computer is the stupidest of pupils' :-) </rambling> This is a copy of the living page "Writing A Go Program To Teach Beginners" at Sensei's Library. ![]() |