AlphaGo versus Fan Hui

AlphaGo versus Fan Hui was a five-game Go match between European champion Fan Hui, a 2-dan (out of 9 dan possible) professional, and AlphaGo, a computer Go program developed by DeepMind, held at DeepMind's headquarters in London in October 2015.^[1] AlphaGo won all five games.^[2][3] This was the first time a computer Go program had beaten a professional human player on a full-sized board without handicap.^[4] This match was not disclosed to the public until 27 January 2016 to coincide with the publication of a paper in the journal Nature^[5] describing the algorithms AlphaGo used.^[2]

Fan described the program as "very strong and stable, it seems like a wall. ... I know AlphaGo is a computer, but if no one told me, maybe I would think the player was a little strange, but a very strong player, a real person."^[6]

Games

Summary

In this match, DeepMind used AlphaGo's distributed version with 1,202 CPUs and 176 GPUs^[5] with Elo rating 3,144.^[7] For each game there was a one-hour set time limit for each player followed by three 30-second byo-yomi overtime periods.

Game	Date	Black	White	Result	Moves
1	5 October 2015	Fan Hui	AlphaGo	White won 2.5 points	272
2	6 October 2015	AlphaGo	Fan Hui	Black won by resignation	183
3	7 October 2015	Fan Hui	AlphaGo	White won by resignation	166
4	8 October 2015	AlphaGo	Fan Hui	Black won by resignation	165
5	9 October 2015	Fan Hui	AlphaGo	White won by resignation	214
Result: AlphaGo 5 – 0 Fan Hui

During this match, AlphaGo and Fan Hui also played another five informal games with shorter time control (each player having just three 30-second byo-yomi) and AlphaGo defeated Fan by three to two.^[5]

Game 1

Fan Hui (black) v. AlphaGo (white), 5 October 2015, AlphaGo won by 2.5 points.^[5]

First 99 moves

Moves 100–199

Moves 200–272 (234 at ; 250 at )

Game 2

AlphaGo (black) v. Fan Hui (white), 6 October 2015, AlphaGo won by resignation.^[5] Although the white stones at the lower-left corner could have been captured if black 135 had been placed at "a", AlphaGo's choice might be safer to win.^[8]

First 99 moves

Moves 100–183 (182 at 169)

Game 3

Fan Hui (black) v. AlphaGo (white), 7 October 2015, AlphaGo won by resignation.^[5]

First 99 moves

Moves 100–166

Game 4

AlphaGo (black) v. Fan Hui (white), 8 October 2015, AlphaGo won by resignation.^[5]

First 99 moves (96 at 10)

Moves 100-165

Game 5

Fan Hui (black) v. AlphaGo (white), 9 October 2015, AlphaGo won by resignation.^[5] Black 75 should be placed at 83, and Fan Hui missed the opportunity.^[9]

First 99 moves (90 at 15)

Moves 100–199 (151/157/163 at 141, 154/160 at 148)

Moves 200–214

Responses

AlphaGo's victory shocked the Go community.^[10][11][12] Lee Sedol commented that AlphaGo reached the top of the amateur level in this match, but had not yet reached the professional level,^[13][14] and he could give AlphaGo one or two stones.^[15] Ke Jie and Mi Yuting thought that the strength of AlphaGo in this match was equal to that of a candidate for Go professional,^[16][17] and extremely close to the professional level,^[18] while Shi Yue thought that it already reached the professional level.^[19][11] "It was terrifying," said Ke Jie, "that AlphaGo could learn and evolve although its power was still limited then."^[20][17][21]

Canadian AI specialist Jonathan Schaeffer, comparing AlphaGo with a "child prodigy" that lacked experience, considered this match "not yet a Deep Blue moment", and said that the real achievement would be "when the program plays a player in the true top echelon".^[22]

See also

• AlphaGo versus Lee Sedol
• AlphaGo versus Ke Jie