I'm one of the six people who vote for AlphaGo 5-0, wondering who's the other five? Wanna know their reason

I wish I’d voted for a 5:0 victory for Ali (AlphaGo’s nickname) here also like I did later (before the games) on FB. My decision was not based on any wishful thinking since I don’t have a preference for who “should” win, but rather on how I perceived the DeepMind team’s communication and scheduling.

_________________
“The only difference between me and a madman is that I’m not mad.” — Salvador Dali ★ Play a slooooow correspondence game with me on OGS?

Lee Sedol wins game 4!

On Twitter Demis said:

Myungwan Kim also pointed out this mistake in his commentary: it should have been atari from below. From then on AlphaGo played a bunch of silly sentes reminiscent of the Monte Carlo bots.

AlphaGo is stronger when playing White. I wonder if komi gives it some kind of extra confidence!

If you are going anthropomorphic, it might make more sense to say it takes the komi to be (too) large.

I try to remember in computer go mailing list discuss that the correct Komi might be 4.5-5.5, but I can't find that thread anymore.

It's equally possible that an unfairly large Komi would weaken the bot because it only tries to win by 0.5 points with high probability. The fact that the bot's fuseki is strong, as white and black, indicates to me that the Komi is not too far off.

Personally, I think Lee Se-dol was clutching at straws when he claimed that AlphaGo struggled when holding the black stones. Before move 78, the bot was ahead by country miles - way more than Komi. Afterwards, it blundered many times in a row and snatched defeat from the jaws of victory. By that time, who played first was entirely irrelevant!

Also, let's not forget that the bot won game 2. You can quibble but a win is still a win.

No, not what this (and other strong go playing programs are doing).

They make the move to have the highest probability of winning the game. They aren't trying to end up with half a point more than the opponent but to end up with AT LEAST half a point more than the opponent. That is the definition of winning.

Many of us are making an assumption that isn't true. Assuming that if (currently) ahead by X points a move that maintains that lead of X points necessarily has a higher probability of ending up at leastT a half point ahead (ie: winning the game) than some other move which instead gives up a point

You're right. My wording was careless. You have a better way of saying it.

My point still stands - we cannot assume that incorrect Komi will have a predictable effect on the bot's play. It might be fair to say it would have *an* effect on the bot's play but assuming which way it would swing would require us to understand the bot's play and we simply don't.

Alternative stream on game 5
https://www.youtube.com/watch?v=YZPKR7HzM_s

So after a thrilling match it ends 4-1 to AlphaGo, well done those 7 people who predicted it correctly. Before the match I had though Lee would win 5-0 or 4-1, but with a big uncertainty due to not knowing how much AlphaGo had improved since October, and it turns out to have improved a lot. How much more will it improve, will we see it giving 3 stones to top pros in a few years, as pros have in the past suggested they should take from God?

I also wrong too since I predict AlphaGo to win 5-0.
4-1 is also okay i guess

Just a little postscript that I saw Lee Sedol beat Kim Jiseok in the round of 8 of the Maxim cup today (with some very nice use of aji), so he's back from his holiday and still winning. I was a bit worried he might go into a slump after the loss to AlphaGo.

A very interesting summary of the AlphaGo event and chosen comments of the games from Chinese professional point of view:

http://www.eurogofed.org/index.html?id=43

Translated reviews/summaries of the games from Lee Sedol himself:

http://badukinkorea.tumblr.com/

Does anyone know what's going on with AlphaGo? I thought we would hear something now if they had scheduled a match with Ke Jie. Are they still working on improving it at this time or are they moving on to doing something else with the AI they are developing?

We might heard more when 2016 US Go congress come since DeepMind members join it.

It's at end of July right? That's more than 3 months away.

I don't think anything is decided.

Does anyone know if they are still working on improving it at this time? They admitted it was playing on Tygem before the Lee Sedol match right? Is it still playing there?