Zen won the second game with 4 stone handi against Takemiya Masaki 9p. Edit: I fixed a mistake, I now get B+20, same as they did.

Post-game commentary:

My own transcriptions, the first game I probably have quite a few major order errors (mostly related to the ko on top left). The second should mostly be fine, though in both games the endgame definitely have minor errors.

Bye bye...

That's a pretty scary, seeing a pro being taken down on 4 stones by Zen. I don't think I would have a chance on 4 stones against the mighty T.

_________________
North Lecale

They should have kept going to 3H!

The thing is, Zen's positional awareness is so good that while it may not see the consequence of some of it's aji, it makes very good (and consistent) judgement calls. It's style is very well suited to handi go.

_________________
Tactics yes, Tact no...

Official SGFs:

Wow, that's truly impressive. I'm going to have to eat my hat, Computer Go is improving more quickly at the mid dan ranks than I suspected would be possible before at least 2020.

Now, we got a problem chess has for some years now: players that cheat in running a programm besides their play on the internet. I don't see a effective solution other than blitz games.

I also think one should match a pro vs. Zen in 2013 to play an even game with a price tag of 10K, sponsored by some firm. I think it's time that Zen can legitimately look to beat pro's, esp. in rapid games.

I think Zen is nowhere near enough to beat pros even. I also suspect that Takemiya could give 5 stones or even more if he didn't care about pretty game records :p

Zen is using a lot of power to deliver these results, isn't it? The Monte Carlo bots have made huge strides (well, at least Zen has!). That has forced a lot of people to reconsider their guesstimates about how strong Zen is, how quickly it's improving, and things like that. But how strong would Zen be if it were running on my laptop?

I am pretty impressed, however I think people are expecting result too quickly. MCTS has a limit because of size of the Branching factor. Also I'd be interested to know what the time limits were like for these games.

Some versions of Zen are running on standard hardware (though upper range, nonetheless) or small clusters (4-8 PCs).
Given the natural hardware evolution (hail Moore's law!), that's how powerful new laptops will be in two to four years.

I would really like to read Takemiya's comments on the game. Considering that both Takemiya and Zen are known for their use of influence, I wonder what he thinks of Zen's unorthodox style, and whether he plans on studying the strategy that the computer employs.

_________________
Patience, grasshopper.

Hm? MCTS is a way of minimizing the effect of huge branching factors...

_________________
That which can be destroyed by the truth should be.
--
My (sadly neglected, but not forgotten) project: http://dailyjoseki.com

Who was making the moves on the board for Zen?

Sorry, what I was trying to say was that MCTS is only useful to an extent because the huge branching factor prevents MCTS from seeing all the moves. to play like a pro, I think an MCTS bot will have to run so many simulations that it seems all of the possible moves many moves deep, and then still run a considerable number of moves for each possible branch

I think Mr. Kato Hideki of team DeepZen.

Yup. He is often on KGS as gghideki when Zen plays tournament games (he speaks English well).