The match started today on Fox server. In the first game AI lost to Dang Yifei and second one it won against Yang Dinxing.
More details here
http://weiqi.qq.com/news/8575.html

Games

So it ended up going 7-1.

Here are the games. I think I have the player names right.

Can someone tell me what's the weakness of this new version of Fine Art? It gets much stronger but still lose so it has to had some weakness or bug

I think it's not so easy to say what the weakness is. It's not perfect of coarse. Can you tell me what Ke Jies weakness is? Sometimes he loses so there must be some weakness, but still it's hard to put your finger on it.

It's completely fine if you found FineArt's weakness hard to find. It's impossible for me too, so I'll wait for others.

It's all the same old problem -- sometimes it just miscalculates the stuff. Like in the game with Dang Yifei it didn't see that its group can be killed. So as far as I see the only chance for a human to win against current engines it's to play as more complex fight as they can. Because in simple games it leaves no chance to pros. I'd recommend the amashi strategy -- take as much territory as you can and then just try to not die anywhere. In that case no matter how good engine estimation is -- if opponent lives inside your territory you lost.

The bottom line is the more calculation and the less evaluation position requires -- the better chances you have. The old proverb says: with a stronger opponent try to complicate stuff.

I thought I would never read something like this in my life. Are we at a position where AI's weak points are calculation and not evaluation? This is the opposite of what everyone considered to be difficult for AIs in Go.

Well, the huge variety of variations was always the main problem for Go engines compared to chess. So in simple positions you just can't miscalculate and make a losing move. You can play not-optimal move but never a losing one. It's not the same in complex fightings. You missed an exact sequence and you're doomed. Since all go ai still use a Monte Carlo trees for calculating then there will always be a probability of just missing right sequence. And it's not the same with chess where a minmax algorithm is used.

This is pretty cool. It makes more sense once you understand what actually goes into the AI. The big breakthrough for recent computer Go has been finding how to create a neural network to evaluate the board and provide whole-board positional judgment, and so that's precisely the area where they're relatively strong now. You can think of neural networks of this sort as being the computer analogue to human pattern recognition, instantaneous feeling, snap judgment, etc (an important part of what Kahneman's calls "system 1", if you are familiar with that literature about human cognition). In other fields, these neural nets have provided huge advances in image and facial recognition, speech processing, etc.

Over millions of training games, more data than any human could possibly process, the "value net" is able to refine its overall global judgment to be much better than any human's, simply by learning over that massive amount of data exactly how much to value different kinds of influence and territory and thickness, and a finely tuned instinct for when a result is better or worse based on the whole board situation, so they outperform humans in the opening and in overall judgment.

But the value net doesn't provide quite as much help in tactical fights and trying to kill/live with a large dragon. To reiterate, the value net's role is most analogous to human snap judgment about what results are good and bad and overall board feeling. Complex tactics involving large dragons aren't so easily reduced to simple snap judgment and overall feeling - it requires exact reading to check whether you actually can get the second eye or connect to something or not. While snap judgment, pattern recognition, etc, can still help spot possible moves that might be tactically useful, and judge whether living might be more or less likely, in the end you still have to do the reading. While bots aren't exactly weak at reading any more and are still steadily improving, it's not the primarily the aspect of the game improved by the latest breakthrough.

I feel like in Wuzhen summit there are gonna be another breakthrough or two. They prepare the whole second day for AI conference.

FineArt lost 2 more games.
Even the time control is normal for server (60 minutes + 3 byoyomi of a minute) FineArt play fast, when human enter byoyomi period it still has more than 45 minutes main time.
For FineArt vs. top Fox Go player event this month the record is 16 wins and 3 loses, 3 more games will be play tomorrow.

I wonder... Would it improve humans' chances against current AI opponents if human players adopted (at least partially) the ancient style of playing. Before all the theoretical breakthroughs by Dosaku, Go and Kitani, players simply fought fiercely from the beginning to the end with little regard for global consideration. The game was more a test of players' reading ability and tactical skills.

Of course, I'm not suggesting playing without considering the whole board - because that'd be suicidal; perhaps humans could play in such a way as to draw the AI into more frequent tactical fights, to pit their strength against AI's weakness. This could be viable for a time before the machines' reading ability inevitably improves beyond that of humans'.

FineArt 10d vs. Mr. Chang Hao 9p (9d)

FineArt vs. Mr. Wang Yao: