AI takes over Go

[coldcrow]

It finally happened: https://www.youtube.com/channel/UCP7jMXSY2xbc3KCAE0MHQ-A

World's strongest player is being crushed by google AI.
I strongly detest the amateur commentator though (the pro is cool), he goes on and on how beautiful Go is and totally doesn't understand how the AI can play so well.

 

[coldcrow]

So 4:1 for the AI. Now I don't have a good knowledge of Go, but am a rather strong chess player, let me share some of my thoughts about this.

Go is a more complex game where humans still don't have full knowledge of the evaluation of positions - especially how far away groups of stones are connected exactly, transitioning from opening to middlegame and so forth. This is in stark contrast to chess - research about how to evaluate was finished in the last century (probably around K&K era). This knowledge found entrance in the evaluation functions of the chess programs leading to the total dominance of modern programs on normal home pcs, as they can calculate about 10 full moves deep in a few seconds.

Now go is different, there is no full knowledge about how to evaluate + the amount of possible moves is higher. So as I understand they use a self-learning NN with which constantly hones its evaluation function and let it play oodles of games against itself. Likely after programming some knowledge into it. This is a rather huge step to creating the Golem. Now self-learning NNs are common in research, but the more these get accepted into our reality the less we are able to control this creation.
Because you could see that the Go-commentators of the match were sometimes flabbergasted by the AI moves, they couldn't understand how the AI reached it. Of course, it was the product of a nonhuman KI. I predict the inception of a self-conscious AI by 2050 latest.

 

[Severian Silk]

This thread already exists.