There are a lot of thoughts about AlphaGo these days… The play level they have reached is awesome…
Since I am very fond of the 9x9 board I cannot help thinking about this… could the AlphaGo team and all their resources solve 9x9 go?
What I mean is something like this: we adapted the neural network, etc, etc, to work on a 9x9 board. We left it evolve
and created an AlphaGo Zero (for 9x9 board). ,Then, we left the program play itself 1 million times… The result is that,
with a 5.5 komi black won (or lost) all of them. For me that would be equivalent to say 9x9 has a predictable result with prefect play (for a specific komi).
In fact, I like the idea of taking the same approach of starting an AI on 9x9, then taking it to 13x13 and then up to 19x19 - just to see how it evolves differently.
It would be very interesting to see such a thing compete with the current AlphaGo Zero.
AlphaGo plays with 7.5 komi under Chinese rules, as I understand it.
If it does, and this was unbalanced, I’d kind of assumed they’d have tweaked it?
That in itself seems like a good indication already that it’s a pretty reasonable value.
Of course, lots of people are interested in this topic, and it seems like a good one for a heated debate judging by the threads out there. One common theme is “What if AlphaGo needs 20 stones as white to beat black? Does this mean humans should take 20 stones?”
IE is komi even static with play ability?
But that brings us back to my original question: if komi for AlphaGo were any different to 7.5, wouldn’t we all have noticed by now?