American Go E-Journal » Computer Go/AI

Man beats machine in human victory over AI

Sunday February 19, 2023

A human player has comprehensively defeated a top-ranked AI system at the board game Go, in a surprise reversal of the 2016 computer victory that was seen as a milestone in the rise of artificial intelligence.

Kellin Pelrine, an American player who is one level below the top amateur ranking, beat the machine by taking advantage of a previously unknown flaw that had been identified by another computer.

The triumph, which has not previously been reported, highlighted a weakness in the best Go computer programs that is shared by most of today’s widely used AI systems, including the ChatGPT chatbot created by San Francisco-based OpenAI.

Financial Times; click here to read more.

Share

AGF Launches Kids Go Server

Saturday September 10, 2022

Kidsgoserver.com, a new site designed specifically for kids aged 6 to 12, is now live on the web as a beta release. A joint project of the American Go Foundation and the British Go Association, the server is designed to be a totally safe and kid friendly environment that can be used by anyone, and in schools. The site is accessed through a website, and requires no software, and no login. Graphics and animations help kids realize when stones are in atari. Your choice of rocket ships on the home page will take you to a lessons page or a playing page on one of several planets. For now, the lessons are only text, but voice actors will read them in the future. More lessons, as well as go problems, will be coming as well.  Kids can play either 9×9 or 13×13 against weak AI bots, as well as against other kids. Chat is limited to preselected phrases so there are no safety concerns. Kids who wish to save their avatar can generate a login code or just pick a new avatar every time they come.

The graphics and characters were all designed by Nick Snyder, a professional animator and artist. The coding is by Akita Noek from OGS, with lessons by Paul Barchilon, and conceptual design by Barchilon and AGF President Terry Benson.

Click here to visit the site, and here to give us feedback on the beta.

Share

AlphaGo vs. AlphaGo with Michael Redmond 9p: Game 50 (Part 2)

Friday July 9, 2021

“Wow, it’s been a while! Welcome back,” Nyarlah commented on the latest Michael Redmond AlphaGo game commentary; Part 1 was published last week. “It’s been too long AGA!” agreed Clay Tin. Some commenters noted the unusually long pre-game chat as Redmond and host Chris Garlock caught up on news from the previous 8 months; note that the game commentary starts at 22:56.

Part 2 of the AlphaGo-AlphaGo Game 50 commentary will be published at 6p ET tonight.  The game series was published by Deepmind after AlphaGo’s victory over world champion Ke Jie 9p in May 2017. The video was produced by Eva-Dee Beech, Chris Garlock and Stephen Hu.

[link]

Share

Redmond AlphaGo game commentary live tonight at 7p EDT

Sunday November 1, 2020

Michael Redmond 9P returns with more live commentary, this time on AlphaGo vs AlphaGo Game 49, tonight at 7p EDT on the AGA’s Twitch channel. His review of the relatively peaceful Game 48 was released Friday night and the latest in the AlphaGo vs. The World series features AlphaGo Master vs. Meng Tailing 9P.

Share

Redmond live commentary Sunday night; latest Redmond Review posted

Saturday August 15, 2020

Michael Redmond 9P’s live commentary series resumes this Sunday, August 16 at 8p EDT on Twitch when he and E-Journal Managing Editor Chris Garlock will review Game 45 in the AlphaGo vs. AlphaGo series.

In the latest episode in the Redmond’s Reviews series — which originally aired live on Twitch back in May on Mother’s Day — Redmond takes a look at his game against Yao Zhi Teng 5P. Once again, Michael tries some AlphaGo moves and “things get messy,” he says. Yao Zhi Teng 5p “is one of the younger really strong players,” and this game was played in late 2019 before the competition hiatus due to the COVID pandemic.

Also, Redmond continues to post new content on his YouTube channel, including life and death problems and his 1988 Judan final against Sakata Eio.

Share

Redmond on AlphaGo vs AlphaGo Game 44 this Sunday

Saturday July 25, 2020

Tune in to the AGA’s Twitch channel at 8p Sunday night for Michael Redmond 9Ps live commentary on AlphaGo vs AlphaGo game 44. “It’s going to be wild,” says Redmond, “with two big groups that could have died in a trade and it’s not clear which is better or more important, various kos, and then suddenly it’s all over with the groups surviving, and a half point difference of course.” That’s 8p EDT this Sunday, July 26 on the AGA’s Twitch channel.

Share

Redmond on Go Seigen vs Kitani, The Tengen Game, and AlphaGo vs The World

Saturday July 4, 2020

Tune in to the AGA’s Twitch channel at 8p Sunday night for Part 2 of Michael Redmond 9Ps live commentary on the 1939 game between Go Seigen and his friend and rival Kitani Minoru. If you need your Redmond fix sooner, not to worry, check out the latest Redmond Review featuring The Tengen Game between Honinbo Dosaku and Yasui Santetsu, which originally aired back on May 3. “What a treat!” says Sev.

And Redmond’s brand-new series, AlphaGo vs. The World — in which he and AlphaGo to Zero co-author Chris Garlock do short reviews of the 2016-17 Master vs human games — is now up to Game 11. “Some really beautiful moves by white in this great game,” says Rory Mitchell. “The placement of stones looks so calm, light, and impressive. Thanks so much for the video!”

Also, for the French Redmond fans, jonathan4055 has just posted subtitles in French for Redmond’s recent AlphaGo to Zero: Revisiting AlphaGo vs Lee Sedol, Game 2 commentary. And of course there’s always Redmond’s YouTube channel, as well.

Share

AlphaGo vs. Lee Sedol Game 2, revisited, plus AlphaGo vs. The World

Sunday June 28, 2020

If you missed the live commentary — originally aired on May 17 on Twitch — by Michael Redmond 9P and Chris Garlock on the second game of the historic 2016 AlphaGo vs. Lee Sedol match, it’s now been released on the AGA’s YouTube channel. Click here to check it out. “Those days of Lee Sedol and you two doing the commentary were some of the most enjoyable times during that AlphaGo match,” said Mike Young. “I got really excited about Go for awhile.”

While there was no live broadcast this Sunday, Redmond and Garlock will return on Sunday July 5 at 8 PM EDT on AGA Twitch. Meanwhile, you can check out their new AlphaGo vs. The World series playlist, with new releases of commentaries on the AlphaGo Master games every Tuesday, Thursday and Saturday at 11 AM EDT.

Share

Redmond launches new AlphaGo video commentary series

Monday June 8, 2020

In the first days of 2017, rumors started to ricochet around the online go community. A mysterious online player had been making huge waves by defeating dozens of top professionals on go sites in Asia in recent days. “Master” first appeared on December 29, 2016, registering from Korea. Achieving 30 consecutive wins against many former and current world go champions, Master defeated Park Junghwan four times and Ke Jie twice. After that, Master appeared on a different go site and logged another 30 consecutive wins. That made it 60 games in a row with no losses. Was AlphaGo the Master?

Michael Redmond had been on holiday when the games were being played but word had of course spread swiftly through the professional community about the mysterious online player racking up win after win. “That kind of record was simply mindblowing,” he says, “and even before I got the game files off the internet it was clear that something new was happening. Anyone – or anything – that could win 60 straight games could probably give a 2-stone handicap, and these were top professionals who were losing to a player no-one had ever heard of.” The 60 Master games were not only evidence that AlphaGo had attained a whole new level of play, but an incredibly rich “treasure trove for professionals,” says Redmond.

Working again with American Go E-Journal Managing Editor Chris Garlock, Redmond is launching a new series of videos in which he’ll focus on the key points of each of the 60 Master games in brief videos, most of which will be 15 minutes or less. The commentaries will also be the basis for Volume 2 of AlphaGo to Zero: The Complete Games, “as well as a chance to introduce viewers to the professional go players who tackled Master,” notes Garlock.

The first video will be released on Tuesday, June 9 on Redmond’s YouTube channel and the series will be linked on the AGA’s YouTube channel  as well. Stephen Hu is producing the series.

The human player in this first game is 15-year-old Pan Tingyu 1P, who finished #1 in the Chinese professional qualification tournament in 2015. Pan has Black and plays a modern version of the mini-Chinese, and AlphaGo shows a new move in the upper left corner, which has since become the standard move for White in the Chinese opening pattern.

NOTE: The AlphaGo vs AlphaGo series, now up to Game 41, will continue through Game 55. 

Share

AlphaGo, KataGo, and the future of AI

Saturday June 6, 2020

Visualization of ownership predictions by KataGo

“There’s something magical about the game of go,” writes Branton DeMoss in a recent blog post. “For thousands of years, it has captured the imagination of those who want to learn what it is to learn, to think about what thinking means. With the recent advent of strong, open source go AI that can beat top professionals, it’s worth tracing the history of the game, why it remained so difficult to beat humans for so long, and what the future of go may hold.”

DeMoss explores the evolution of computer go, and then discusses how AlphaGo differs from the open source Katago. “KataGo attempts to predict a greater number of game outcomes than just value,’ says DeMoss, “in particular, KataGo also predicts final territory control, final score difference, and from each board state the opponent’s next move. As a result of these improvements, KataGo massively outperforms Leela Zero and Facebook’s ELF bot in learning efficiency. KataGo achieves a factor of fifty improvement in training efficiency vs. ELF”.

The creator of KataGo, David J. Wu, answers some of DeMoss’s questions at the end of the article. “I think the AlphaZero-style training loop using MCTS (Monte Carlo Tree Search) is not the last word on [things like] this,” says Wu. “Blind spots are just the most visible of the flaws, but there are some technical and theoretical details you can dig into that start to make it clear that there are some practical problems with how exploration and move discovery work in this loop, some basic theoretical flaws involving mismatches between the neural net’s training distribution and usage, and also some fundamental ‘missing’ capabilities in current bots in terms of using search effectively.” The full blog post can be read here. -Story by Paul Barchilon. image from Accelerating Self-Play Learning in Go, by David J. Wu.

Share