American Go E-Journal » Master/AlphaGo Commentaries

DeepMind has yet to find out how smart its AlphaGo Zero AI could be

Monday November 13, 2017

“Perhaps the most interesting thing about AlphaGo Zero, though, isn’t how fast it was able to do what it did, or with such efficacy, but also that2017.11.12_hassabis-techcrunch it ultimately didn’t even achieve its full potential,” reports TechCrunch. “DeepMind CEO and co-founder Demis Hassabis explained on stage at Google’s Go North conference in Toronto that the company actually shut down the experiment before it could determine the upper limits of AlphaGo Zero’s maximum intelligence.”

“We never actually found the limit of how good this version of AlphaGo could get,” he said. “We needed the computers for something else.”

Hassabis said that DeepMind may spin up AlphaGo Zero again in future to find out how much further it can go, though the main benefit of that exercise might be to help teach human AlphaGo players about additional, “alien” moves and stratagems that they can study to improve their own play.

PLUS: The October issue of Games magazine includes “A God of Go: AlphaGo Crosses the Next Frontier of Artificial Intelligence”

Share

AlphaGo Zero-AlphaGo Master: Two openings, less variety

Saturday November 11, 2017

“In the set of 20 games between AG Zero and AG Master, there are pretty much just two openings — i.e. identical moves for about the first 202017.11.11_ag-ag-zero-opening moves — one with Zero as Black and one with Master as Black,” says Michael Redmond 9p in this first commentary on the recently released AG Zero games. “This provides an opportunity to examine how Zero differs from Master, as well as how Master differs from earlier versions. ”

“When AGMaster plays against AGZero, it does not show the variety that it had before,” says Redmond. “As AG does not change within a version, I find it hard to accept that it apparently does not have the option to play moves that it played before in identical board positions. In the ‘Master series’, 60 games played against top pros in Dec 2016 to Jan 2017, Master could play the 3-4 point as it’s first move in about 1/4 of the games when it had Black. Incidentally, AGKeJie also could play the 3-4 point in some of it’s games. The fact that Master repeats the same opening every time in these games against AGZero bothers me and makes me question, is this truly the same version of AGMaster that played the Master series, and if so, what caused it to play the same opening every time in this series, when it was allowed to have variety in previous games with identical board positions? The difference in calculated winning percentage between A and B should be extremely small and I would expect it to have little or no effect on the ultimate win-loss record. This set of games would be much more valuable if Master had been allowed to vary in it’s choices for moves.”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock, and see below for the two sgf commentaries. Note that these commentaries focus only on the two openings; watch for a full-game Zero-Master commentary next week. Click here for a playlist of all the Redmond AG commentaries.

Video produced by Michael Wanek and Andrew Jackson. The sgf files were created by Redmond, with editing and transcription by Garlock and Myron Souris.

[link]

[link]

Share

AlphaGo-AlphaGo Game 15: New, different and possibly bad

Friday November 10, 2017

“I don’t like to call it weird, but in this game we’re going to see some new and different stuff that AlphaGo is doing with the joseki that I don’t 2017.11.10_ag-ag-thumb-15really understand and I don’t like it,” says Michael Redmond 9p in his commentary on Game 15. “It’ll be interesting to see if eventually I change my mind, but for now I’m going to say it’s a bad move.” Redmond adds that “We’ll also see another typical AlphaGo move later in the game that’s pretty exciting too.”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock. Click here for a playlist of all the Redmond AG commentaries.

And keep an eye out here and on the AGA YouTube channel for the launch of Redmond’s commentaries on the AlphaGo Zero-Master games, coming very soon!

The Game 15 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

Redmond announces new series on AlphaGo Master vs AG Zero

Saturday November 4, 2017

“We were expecting DeepMind to make some sort of an announcement (about a new version of AlphaGo),” says Michael Redmond 9p, “But 802017.11.04_agzupdatethumb games was a big present.” (Self-taught AlphaGo Zero bests all previous versions in record time Redmond discusses AlphaGo Zero with the E-Journal’s Chris Garlock in a brief video announcing the launch of a new series of game commentaries. DeepMind released four sets of games for the self-taught AI, including training games, games against the Fan Hui version, the Lee Sedol version and the Master version, which defeated 60 top human opponents earlier this year. “I’m going to be looking at the games where Master plays Zero, mainly because Master is such a popular version of AlphaGo,” Redmond says. Master’s tactics, including big shimaris and emphasizing the center “people wanted to play, but were afraid because that way of playing is weak in territory. Master showed us some successful ways…and is still having an effect on how professionals play. So it’s going to be really interesting to see Master playing against a stronger version of AlphaGo.”

Share

AlphaGo-AlphaGo Game 14: A double kakari and a new joseki

Saturday November 4, 2017

“In this game we’re going to see a double kakari against a star point, a first for this series of games,” says Michael Redmond 9p in his game 2017.11.03_AG-14commentary on Game 14. “In the Master vs. human series back in January, Master would play away when the opponent played a kakari against a star point, sometimes. Now we’ll get to see how Master plays this with White, and it has a special move. Its a new joseki that actually make s some sense, so it’s going to be interesting.”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock.

The Game 14 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

AlphaGo-AlphaGo Game 13: The large knight’s move enclosure workout

Sunday October 29, 2017

“This is a very different game, in that there are three corner enclosures,” says Michael Redmond 9p in his game commentary on AlphaGo-2017.10.27_AG-13AlphaGo Game 13. “Black isn’t playing a kakari, which is different from what human players do now. So we’re going to have kind of a workout in how to deal with the large knight’s move enclosure.”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock.

The Game 13 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

AlphaGo-AlphaGo Game 12: An interesting opening and an unusual collapse

Saturday October 21, 2017

“In this game we start with a variation of the mini Chinese opening, so we’re going to have another of these 3-3 invasions,” says Michael 2017.10.20_ag-12Redmond 9p in his game commentary on AlphaGo-AlphaGo Game 12. “It’s a really interesting opening, and it’s one of the games that ends in a collapse, so it’s an unusual game in that way.”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock.

The Game 12 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

Self-taught AlphaGo Zero bests all previous versions in record time

Wednesday October 18, 2017

AlphaGo Zero, the latest version of the go-playing AlphaGo AI, defeated all previous versions of AlphaGo in just forty days. More importantly, AlphaGo Zero was taught go’s rules, but given no additional instructions, instead learning the best moves by playing 2017.10.18_alphago-zero-saran-poroongmillions of games against itself. Details of the new program were published Wednesday in the journal Nature.

“AlphaGo Zero independently found, used and occasionally transcended many established sequences of moves used by human players,” write the AGA’s Andy Okun and Andrew Jackson in an accompanying article. “In particular, the AI’s opening choices and end-game methods have converged on ours — seeing it arrive at our sequences from first principles suggests that we haven’t been on entirely the wrong track. On the other hand, some of its middle-game judgements are truly mysterious and give observing human players the feeling that they are seeing a strong human play, rather than watching a computer calculate.”

The ability to self-train without human input is a crucial step towards the dream of creating a general AI that can tackle any task, reports Nature. In the nearer-term, though, it could enable programs to take on scientific challenges such as protein folding or materials research, said DeepMind chief executive Demis Hassabis at a press briefing. “We’re quite excited because we think this is now good enough to make some real progress on some real problems.”

Prof Satinder Singh, a computer scientist at Michigan University, who reviewed the findings for the journal said: “The AI massively outperforms the already superhuman AlphaGo and, in my view, is one of the biggest advances, in terms of applications, for the field of reinforcement learning so far.”
- Chris Garlock; Image credit: Saran_Poroong Getty Images

Read more…

The AI That Has Nothing to Learn From Humans
DeepMind’s Go-playing AI doesn’t need human help to beat us anymore
AlphaGo Zero: Google DeepMind supercomputer learns 3,000 years of human knowledge in 40 days
DeepMind has a bigger plan for its newest Go-playing AI
AI versus AI: Self-Taught AlphaGo Zero Vanquishes Its Predecessor
AlphaGo Zero Goes From Rank Beginner to Grandmaster in Three Days—Without Any Help
‘It’s able to create knowledge itself’: Google unveils AI that learns on its own
DeepMind AlphaGo Zero learns on its own without meatbag intervention
This more powerful version of AlphaGo learns on its own

Share

AlphaGo-AlphaGo Game 11: A calmer game, with hidden reading

Friday October 13, 2017

“This game is a lot calmer than Game 10,” says Michael Redmond 9p in his game commentary on AlphaGo-AlphaGo Game 11. “There’s a lot of 2017.10.13_ag-ag-thumb-11fighting that doesn’t actually come into the game, but I’ll be showing a lot of variations about things that could have happened, so there’s a lot of, you might say, hidden reading. And then there’s a ko at the end, for the life of a group. ”

Click here for Redmond’s video commentary, hosted by the AGA E-Journal’s Chris Garlock.

The Game 11 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

Redmond AlphaGo Q&A released

Wednesday October 11, 2017

Michael Redmond’s series of commentaries on the fascinating AlphaGo-AlphaGo games has proven extremely popular, with nearly 90,0002017.10.11_ag-ag-thumb-qa views so far, and lots of comments from viewers. Today Redmond, along with host Chris Garlock, releases his first Q&A video, responding to some of those questions. “It’s been a wonderful challenge, not only trying to understand these complex, historic games, but figuring out how to explain them,” says Redmond, “so the response to the videos has been quite gratifying and we’re pleased to acknowledge and respond in this new series of Q&A videos.”

Share