I played darkforest (a computer go program developed by Facebook, based on deep learning techniques using a convolutional neural network combines the techniques of its predecessor with Monte Carlo tree search) last night on KGS. The difference between AlphaGo and darkforest is AlphaGo can only play based on Chinese rules only while darkforest based on Japanese rules)
And here’s what I think.
The AI to me is very strong (it’s 5 dan on KGS anyway), however the moves is typical and I see no special move or ear-reddening move but the play is solid. In the middle game I was thinking that I have lead by 10-15 pts so I played defensively and not attacking however at the end game I tested darkforest and it’s very silly (darkforest did some silly moves too).
I have 4 stones handicap and win by 5.5 a very close one and I am happily promoted to 2 dan after the game 🙂
Here’s the game: