[Computer-go] Chess vs Go // AI vs IA

Mark Boon tesujisoftware at gmail.com
Wed Jun 2 09:28:20 PDT 2010


On Jun 1, 2010, at 8:58 PM, Olivier Teytaud wrote:

> This improvement per doubling speed is interesting to me... it was claimed
> in the early times of MCTS that the winning rate with 2N simulations against N simulations
> was roughly constant, and then various studies (even in the best case, i.e. for a program against itself!) 
> have shown that this does not hold, by far. 

But in any case that would make the case that software matters only stronger.

What curls my toes is the notion often expressed as if there's 'traditional AI' and today's techniques. First of all, 'intelligence' is so poorly defined that discussions about it often get completely tangled up. And it's still rather poorly understood how the human brains comes to be intelligent (by whatever definition you choose to use) so that we don't even know to what extent solutions, old or new, are 'artificial'.

The way I see it is AI is the process of making a computer complete tasks we find interesting or challenging as humans. Along the way there have been many attempts that ultimately are discarded for more successful approaches. But without the first attempts we wouldn't have gotten where we are today. Maybe it didn't always get us the results we wanted, but it often increased our learning to allow us to make the next step.

With regards to the discussion which is harder, chess or Go, I'd like to observe that both are very hard for humans. To what extent one is harder for humans than the other is a difficult question to answer simply because the human brain is so poorly understood still. Although I think it's clear that Go is harder than chess, if you want to have a real discussion about it you have to be more precise in the definitions. Go is harder for computers than chess when compared to humans but it's tricky even to be absolutely sure about that. There are so many approaches we haven't tried yet.

Mark




More information about the Computer-go mailing list