Archive for the ‘go’ Category

AlphaGo: Implications for Machine Translation

Thursday, April 7th, 2016

In this March 15, 2016 photo, South Korean professional Go player Lee Sedol reviews the match after finishing the final match of the Google DeepMind Challenge Match against Google's artificial intelligence program, AlphaGo, in Seoul, South Korea. Google's Go-playing computer program again defeated its human opponent in a final match on Tuesday that sealed its 4:1 victory. (AP Photo/Lee Jin-man, File)

Machine defeats man at the game of Go.

The entire world was stunned at the 4–1 win by Google’s Deep Mind over Lee Se-Dol, one of the world’s best Go players. Some say that Deep Mind is a highly specialized intelligence that only knows how to play Go. But the principles, techniques, and algorithms underlying Deep Mind do in fact have wider application to so-called AI-complete problems. What do they mean for Machine Translation (MT)?

The development of go programs and machine translation programs have followed a parallel path.

The initial generation of solutions to both problems were based on “classical” AI techniques of encoding human knowledge. The go programs used rules of “good shape”, and human-style “reading”. The MT programs used grammars and rulesets built by human linguists. The results, in the case of go, were programs which played at the amateur sub-dan level (meaning the top 10% of all players). The results, in the case of MT, were programs which could at best produce vaguely understandable translations.


The brain of the go player

Saturday, January 7th, 2006

Chinese researchers have analyzed the brains of go players with fMRI (abstract), finding “a modest degree of stronger activation in the right parietal area than in left. This type of right hemisphere lateralization differs from the modest left hemisphere lateralization observed during chess playing.”

To abstract away basic visual processing, the researchers presented the subjects—half a dozen mid-level amateur players—with three alternating go boards, one empty, one with random stone placement, and one with an actual game position

Go playing requires the participation of a network of cortical areas. As shown in Fig. 2c, these areas include the mid-dorsal prefrontal area (BA9), the dorsal prefrontal area (BA6), the parietal areas (BA7, 40), the posterior cingulate areas (BA30/31), the occipital area (BA19), and the posterior temporal area (BA37). These areas are generally engaged in attention, spatial perception, imagery, manipulation and storage in working memory, retrieval in episodic memory, and problem solving [2].

An interesting finding was that the primary somatosensory and motor areas were also activated. The authors’ explanation is that the player is getting ready, in his mind, to pick up a real stone and play it. A similar phenomenon was found when subjects were shown Chinese characters (earlier post): their motor areas indicated they were ‘thinking” of picking up a pen or brush to draw the character!

Concerning the right lateralization, the researchers hypothesize that this may be attributed to the storage component of spatial working memory, global spatial pattern analysis, and sustained spatial attention all being biased in the right parietal areas. They also found lateralized activity in the left dorsal lateral prefrontal area, normally involved in language functions. They hypothesize this may be due to the subjects verbalizing go terms internally.

As topics for futher research, the authors propose neuro-imaging studies of professional go players. They make the fascinating prediction that the pros will show even less activity in the frontal areas, since they know the game so well they are not really thinking as they play.

The game of go and the brain

Thursday, May 26th, 2005

A group of Chinese researchers has studied go players using fMRIs. The experimental design had the subjects looking at an empty board, a board with randomly placed stones, and a real game position.

Two findings struck me. First, the researchers found players may be internally verbalizing their play.

Lateralized activity was also found in the left dorsal lateral prefrontal area (BA44/45): it was activated when the subjects carried out the GO playing task. Because left BA44/45 is usually involved in language functions, [2], this result suggests that the subjects may be verbalizing GO terms when they were playing GO.

Second, they found players may be imagining the actual physical motion of playing stones on the board.

The primary somatosensory and motor areas (BA3-1-2/4) were more active in the GO playing task. A possible explanation for this activiation may be that subjects were imagining that they were picking up stones and placing them on the game board int he process of finding out the next reasonable solution.

There are other interesting findings as well, such as a modest right lateralization in the parietal area as compared to chess, which they attribute to the heavily spatial nature of the game. They found activation in many other cortical areas, such as dorsal prefrontal, parietal, occipital, and posterior temporal areas.

Abstract of the paper, “A functional MRI study of high-level cognition. II. The game of GO”, is here . Full paper is here.

Bob’s Overview of Go

Wednesday, October 22nd, 2003

Bob has written a nice overview of the game of Go, also known as weichi or paduk together with Sangit Chatterjee. It ws published in the July-August 2003 issue of “Science and Culture”, a respected Indian periodical.

Which points on the go board are played when?

Saturday, August 9th, 2003

Ashley Feniello has done some brilliant analysis on when what points on the go board are typically played. He reported the results, in the form of a cool animated GIF, on Sensei’s Library.

Computer Go Overview

Tuesday, June 10th, 2003

Bob has written a good overview of the computer go problem.