American Go E-Journal » Computer Go/AI

AlphaGo-AlphaGo Game 7: Go Seigen-like attachments, a 3-3 variation and a running fight

Friday September 15, 2017

“In this game we will see some Go Seigen-like attachments that White plays against a Black shimari,” says Michael Redmond 9p in his game 2017.09.15_ag-ag-thumb-7commentary on AlphaGo-AlphaGo Game 7. There’s also “an AlphaGo variation for the early 3-3 invasion, and after White makes a moyo there will be a running fight in the center.”

Click here for Redmond’s video commentary, just posted on the AGA’s YouTube channel and hosted by the AGA E-Journal’s Chris Garlock.

The video is produced by Michael Wanek and Andrew Jackson. The sgf file was created by Redmond, with editing and transcription by Garlock and Myron Souris.

[link]

Share

Your Move/Readers Write: AlphaGo is unbeatable; get over it

Sunday September 10, 2017

“Apparently, some people believe that someday a human will be able to defeat AlphaGo,” writes Joel Sanet. “It’s not gonna happen. The reason is biological, not technological. No human being is capable of thinking about the game the way AlphaGo does. AlphaGo’s way of thinking is better than the human way; ergo it is no longer possible for a human to beat AlphaGo. We human beings are not capable of considering a choice of moves by determining a concrete number for each called “the probability of winning” then choosing the one with the highest value, but this is what AlphaGo does.

“Thinking that it is possible for a human to win now is due to anthropomorphization, the application of human attributes to something that is not human, a process rampant in the go community. I have heard people say, ‘AlphaGo likes the early 3-3 invasion’ or ‘He (or she) likes thickness.’ AlphaGo can’t ‘like’ anything because it has no emotions. It plays the early 3-3 invasion because it maximizes its probability of winning in certain openings. Also, as far as I know, AlphaGo has no concept of thickness. It has nothing to do with how AlphaGo derives its moves. Furthermore, AlphaGo is not a ‘he’ or a ‘she’. AlphaGo is an ‘it’.

To attribute thinking to AlphaGo is also a mistake. I wrote that it chooses the option with the highest probability of winning. It doesn’t “choose” anything because it isn’t self-aware. AlphaGo receives input, does what it is programmed to do, and produces output. To me this is more akin to a human knee jerk than to true thought. A doctor’s percussion hammer causes sensory neurons to fire off a signal to the spinal cord where it is processed and returned to the knee via motor neurons without intercession of the brain. This is analogous to AlphaGo’s input-programming-output. AlphaGo’s programming is immutable. The day AlphaGo changes its own programming is the day I’ll say it thinks.

Nevertheless, humans can learn from AlphaGo. We have learned that the shoulder hit is a lot more useful than anyone thought. AlphaGo’s new 3-3 invasion joseki makes sense so we can benefit from that, but I advise you not to do the early invasions until you are able to read the rest of the game to the end.

Alphago’s supremacy over humans is no reason to feel that studying go is a dead end. Your study is de facto open-ended because you will never reach the end of it. People study go to improve, not to become the strongest player on the planet.”

Share

AlphaGo vs AlphaGo Game 6: Flexibility and a bias for complications

Sunday September 10, 2017

“In this game AlphaGo shows its flexibility when Black abandons a running fight and tries to control the open lower side of the board instead,”2017-09-10-alphago-game-5-video says Michael Redmond 9p in his game commentary on AlphaGo-AlphaGo Game 6. “In the second fight of the game, White deals with two weak groups masterfully. Finally, Alphago shows its bias for complications when White allows a dangerous ko in the corner.”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock. As usual, the commentary in the sgf file here includes variations not covered in the video commentary, and the sgf commentary includes additional comments transcribed from the video.

The video is produced by Michael Wanek and Andrew Jackson. The sgf file was created by Redmond, with editing and transcription by Garlock and Myron Souris.

[link]

Share

DeepZenGo wins all-AI go competition

Wednesday September 6, 2017

At an all-AI go competition held in China, Japan’s DeepZenGo took first place and then bested top Chinese pro Kong Jie 9p, who was being assisted by one of the AI runners up, CGI, from Taiwan. Jie was able to choose which AI to use as an assistant, and opted for CGI over the Chinese system FineArt, which had come into the tournament as a favorite. Among the 12 contestant systems, one North American entrant, MuGo by Brian Lee, came in 11th. MuGo was only six or so months old, and not that strong yet, but Lee was pleased to play against other systems, and preparing for this event with a short lead time was good motivation to work harder. “It was good to have a goal. I’d been working on it alone for four months, and it’s difficult to construct imaginary castles when there’s no one looking at it but yourself.” Scheduling conflicts kept other North American Go programmers, like Dave Fotland, away and Facebook has not been working on its system actively. The AlphaGo group did not attend, with the final versions of AlphaGo having retired from competition after the match with Ke Jie 9p in May. The competition took place in Ordos City, China, at the first ever Chinese Go Congress, a well-attended event that brought together 5,000 mostly amateur attendees, according to organizers.
- Andy Okun, Special Correspondent

[link]

Share
Categories: China,Computer Go/AI
Share

AlphaGo vs AlphaGo Game 5: An AI-like opening, then one fight at a time and a beautiful endgame

Saturday September 2, 2017

“In the opening this game looks very AI-like to me, in that I think the order of moves is not consistent,” says Michael Redmond 9p in his game 2017.09.02_alphago-game5commentary on AlphaGo-AlphaGo Game 5. “In the middle game Black controls the center of the board. Our reading skills are tested as Black invades White’s moyo, and then White lives with three weak groups inside Black’s sphere of influence. Unlike in other games we’ve seen so far in this series, the middle game fights are one at a time instead of all over the place, like in Game 2, for example. It’s more organized, you might say, so in that way, it’s easier for me to explain what’s going on. The game winds up with a very nice endgame, in fact I think it’s a beautiful endgame.”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock. As usual, the commentary in the sgf file here includes variations not covered in the video commentary, and for the first time, the sgf commentary now includes additional comments transcribed from the video. Both include the news that Redmond and Garlock are now working on an e-book about the AlphaGo-AlphaGo games. Redmond and Garlock discuss their plans for more AlphaGo-AlphaGo commentaries in this brief video.

The video is produced by Michael Wanek and Andrew Jackson. The sgf file was created by Redmond, with editing and transcription by Garlock and Myron Souris.

[link]

Share

Your Move/Readers Write: Gary Kasparov on AlphaGo

Tuesday August 29, 2017

By Michael Bacon

Enjoyed the coverage of the Go Congress immensely! Could not help but poke a few of my chess friends in the eye while contrasting all the coverage it received with all the coverage the recent US Open did not receive on the organ of US chess, the USCF webpage. I’ve also been transfixed by Michael Redmond’s videos. The man is a national treasure!

Former World Human Chess Champion Gary Kasparov, who will always be remembered as the human who lost to a ‘machine,’ in his apologia for having lost to the computer chess ‘engine’ called ‘Deep Blue’ — not for having turned Kasparov a deep shade of blue, and a whiter shade of pale, I might add — writes about go in ‘Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins‘:2017.08.27_Deep Thinking-Kasparov
“The nineteen-by-nineteen Go board with its 361 black and white stones is too big of a matrix to crack by brute force, too subtle to be decided by the tactical blunders that define human losses to computers at chess. In that 1990 article on Go as a new target for AI, a team of Go programmers said they were roughly twenty years behind chess. This turned out to be remarkably accurate. In 2016, nineteen years after my loss to Deep Blue, the Google-backed AI project DeepMind and its Go-playing offshoot AlphaGo defeated the world’s top Go player, Lee Sedol. More importantly, and also as predicted, the methods used to create AlphaGo were more interesting as an AI project than anything had produced the top chess machines. It uses machine learning and nural networks to teach itself how to play better, as well as other sophisticated techniques beyond the usual alpha-beta search. Deep Blue was the end; AlphaGo is the beginning.” (pgs. 74-75)

Please note the author capitalizes “Go,” but not “chess.” I find that curious as I have always capitalized “Chess.” (note: the EJ does not capitalize go, consistent with AP style) In addition, Lee Sedol, as all go players know, was not the “…world’s top Go player,” when he lost to the computer program known as AlphaGo.

2017.08.27_kasparov-bookWe move along to page 104 where one finds this:2017.08.27_Kasparov-playing
“The machine-learning approach might have eventually worked with chess, and some attempts have been made. Google’s AlphaGo uses these techniques extensively with a database of around thirty million moves. As predicted, rules and brute force alone weren’t enough to beat the top Go players. But by 1989, Deep Thought had made it quite clear that such experimental techniques weren’t necessary to be good enough at chess to challenge the world’s best players.”

Finally, on page 121, Kasparov, or his co-author Mig Greengard, writes this paragraph:
“More success was had with another method for allowing machines to extend their thinking into the hypothetical outside of the direct search tree. Monte Carlo tree search simulates entire games played out from positions in the search and records them as wins, draws, or losses. It stores the results and uses them to decide which positions to play out next, over and over. Playing out millions of “games within the game” like this was not particularly effective or necessary for chess, but it turned out to be essential in Go and other games where accurate evaluation is very difficult for machines. The Monte Carlo method doesn’t require evaluation knowledge or hand-crafted rules; it just keeps track of the numbers and moves toward the better ones.”

While reading I continually thought of former World Human Chess Champion Emanuel Lasker’s famous quote, “If there are sentient beings on other planets, then they play Go.”

Not chess; go!

Share

Redmond AlphaGo commentaries generate big response

Monday August 7, 2017

Response to the four Michael Redmond 9P AlphaGo-AlphaGo commentaries released last week has been overwhelmingly enthusiastic, with nearly 25,000 views so far in just a week.
“I loved the master series, but this is even better,” wrote Alek Erickson7. “I have been waiting for English pro commentary on the self-play 2017.08.07_AlphaGo-Redmond-dawggames for so long.” And oncedidactic1 said the commentaries are “Really really valuable, both entertaining and enlightening, to hear Michael’s perspective on this game, which I’ve seen a lot of commentary on. I feel like Michael has wonderful insights into where alphago is at.” Bla Bla6 added that “This makes me realize that Ke Jie and the players in the Master series didn’t even come close of testing the limits of AlphaGo.”
Then on Sunday, Mr. AlphaGo himself, DeepMind CEO Demis Hassabis tweeted “1st #AlphaGo vs AlphaGo Redmond commentary: https://goo.gl/unK6dy amazing game and analysis: ‘AG has invented a whole new opening theory’!”

Check out the video commentaries here, with links to the sgf commentaries (in italic):
AlphaGo vs. Alphago with Michael Redmond 9p: Game 1
Redmond’s AlphaGo-AlphaGo commentaries launched

AlphaGo vs. Alphago with Michael Redmond 9p: Game 2
AlphaGo-AlphaGo Game 2; Fighting throughout, a surprising sacrifice, a final huge ko 

AlphaGo vs. Alphago with Michael Redmond 9p: Game 3
AlphaGo-AlphaGo Game 3: Three 3-3 invasions, a big moyo and a fight that fills the center of the board 

AlphaGo vs. Alphago with Michael Redmond 9p: Game 4
AlphaGo-AlphaGo Game 4: Reminders of Go Seigen, escalating trades and semeais, and a final ko 

“If only I could ‘like’ a 100,000 times, it would not be enough,” said From Fear to Trust1, while mmKALLL said “I think my head imploded here. Crazy to think of all the 47 games ahead… Thank you!” And Tokenias3 chimed in with “That dog is cute.”

“Guys, slow down!” pleaded trucid22. “1.5 hour review each day is a bit much. I can barely keep up! Spread out the games a bit.” The next set in the series is in pre-production now so trucid22 and the rest of the AlphaGo fans have some time to catch up. We’ll keep you posted on plans for the next release.

E-Journal Managing Editor Chris Garlock hosts the commentaries, which are produced by Michael Wanek and Andrew Jackson (who created the snazzy introduction).

Share

AlphaGo-AlphaGo Game 4: Reminders of Go Seigen, escalating trades and semeais, and a final ko

Friday August 4, 2017

“With this game I get to talk about some moves in AlphaGo’s opening that remind me of the great player Go Seigen,” says Michael Redmond 9p 2017.08.04_alphago-alphago-game4in his game commentary on AlphaGo-AlphaGo Game 4. “The territory is very close throughout the game, while fighting in the center gradually escalates with trades and semeais to be calculated and discarded, and even during a final ko to kill a huge Black group, the correct variations leads to a half point difference.”

Click here for Redmond’s nearly 90-minute video commentary, hosted by the AGA E-Journal’s Chris Garlock, and follow along with the sgf below, which as usual includes extra variations.

The video is produced by Michael Wanek and Andrew Jackson.

[link]

Share

AlphaGo-AlphaGo Game 3: Three 3-3 invasions, a big moyo and a fight that fills the center of the board

Thursday August 3, 2017

This exciting game features an astonishing three invasions at the 3-3 point, prompting Michael Redmond 9p to note that “This version of AlphaGo 2017.08.03_AlphaGo vs. Alphago Game 3will invade here at any time when there is no urgent fighting going on. AlphaGo played an early invasion at the 3-3 in just two of the 60 Master series games, but that was shocking, as it defied the common knowledge of pros that such an early invasion should be bad. In this 50-game series AlphaGo played an early 3-3 invasion about 40 times.”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock, and follow along with the sgf below, which as usual includes extra variations.

“Black plays a big moyo game, and then chases an eyeless White group into Black’s moyo, to start a fight that fills the center of the board,” adds Redmond.

The video is produced by Michael Wanek and Andrew Jackson.

[link]

Share

AlphaGo-AlphaGo Game 2; Fighting throughout, a surprising sacrifice, a final huge ko

Wednesday August 2, 2017

“In this game AlphaGo shows great flexibility in the early stages, and also its ability to calculate extremely complicated fights later in the game,” 2017.08.02_AlphaGo vs. Alphago2says Michael Redmond 9P in his commentary on Game 2 in the AlphaGo-AlphaGo self-played series. Click here for his video commentary, hosted by the AGA E-Journal’s Chris Garlock, and follow along with the sgf below, which includes the extra variations Redmond refers to in the video. “Against Black’s sanrensei, White plays two unusual moves at 10 and 16 to create a unique opening,” says Redmond. “As the fighting starts, White makes a surprising sacrifice, abandoning a group to take the offensive in the center. Fighting continues throughout the game to climax in a final huge ko.” The video is produced by Michael Wanek and Andrew Jackson.

[link]

Share