字體：小 中 大 |
|
|
|
| 2016/03/10 23:41:29瀏覽566｜回應4｜推薦25 | |
|
人機大戰，巔峰競技 - Game II: AlphaGo再下一城
局後大俠小思：
一、AlphaGo的宏觀思維，整局優勢，整體氣勢，創意之招（黑的第19子），令下得保守安全的李世石（改變第一場比賽的心態與策略），一路追趕，不時長考，以致用盡時限（2小時）而進入讀秒階段，AlphaGo仍有20分20秒。一入讀秒，李世石已至絕境。
二、評論員Michael Redmond九段的評論尚可，但明顯偏向人類；若AlphaGo能一邊下棋一邊評論，肯定比人類評論員的講評更能令人眼睛一亮，例如：目前黑棋實空是xx目、白棋實空是YY目；自己下每一著的理由，等等，保證令人眼界大開。
三、投子認輸後，李世石滿臉倦容，筋疲力竭，無言無語；AlphaGo是不懂、也不會倦容、筋疲、力竭的。
四、AlphaGo的能力棋力深不可測，大俠保守估計，目前的AlphaGo應有11段。
五、人類常說的直覺、感覺、心態與策略等，AlphaGo肯定有不同的認知與說法。
Google stunned the world by defeating Go legend Lee Sedol yesterday (3/10), and it wasn't a fluke — AlphaGo, the AI program developed by Google's DeepMind unit, has just won the second game of a five-game Go match being held in Seoul, South Korea. AlphaGo prevailed in a gripping battle that saw Lee resign after hanging on in the final period of byo-yomi ("second-reading" in Japanese) overtime, which gave him fewer than 60 seconds to carry out each move.
"Yesterday I was surprised but today it's more than that — I am speechless," said Lee in the post-game press conference. "I admit that it was a very clear loss on my part. From the very beginning of the game I did not feel like there was a point that I was leading." DeepMind founder Demis Hassabis was "speechless" too. "I think it's testament to Lee Sedol's incredible skills," he said. "We're very pleased that AlphaGo played some quite surprising and beautiful moves, according to the commentators, which was amazing to see." The close nature of the game appears to offer validation of AlphaGo's evaluative ability, the main roadblock to proficiency for previous Go programs. Hassabis says that AlphaGo was confident in victory from the midway point of the game, even though the professional commentators couldn't tell which player was ahead. After more than four hours of tight play and a rapid-fire endgame, Google’s AlphaGo has won a second contest against grandmaster Lee Sedol, taking a two-games-to-none lead in their historic best-of-five match. The surprisingly skillful AlphaGo now needs only one more win to claim victory in the match. The Korean-born Lee Sedol is widely-regarded as the top Go player of the last decade, after winning more international titles than all but one other player. He is currently ranked number five in the world, and according to Demis Hassabis, the Google AI lab that created AlphaGo, his team chose Lee Sedol for this all-important match because they wanted an opponent who would be remembered as one of history’s great players. Fast Versus Cautious
Attack it did. As commentator Michael Redmond (a 9 Dan Professional Go Player) put it, AlphaGo started “fast,” and seven moves in, the machine made what he called a “slightly unusual move.” He wasn’t prepared to say whether it was a good move or a bad move, but it was aggressive, as if the machine was trying to force Lee Sedol into action. It soon made another move along the same lines. But Lee Sedol didn’t necessarily bite. At this point in the game, he took a rather long time considering his position and then continued with a comparatively cautious approach. On the whole, Redmond said, the game was moving slower than it had at the same point in Game One. “White (Lee Sedol) is playing a much more conservative game,” Redmond said. Perhaps the Korean had adopted an added degree of caution after considering how well AlphaGo performed in Game One, but a certain amount of caution was expected because he was playing white, not black. Regardless, AlphaGo was showing—once again—that it’s significantly more skilled than it was in October when it topped Fan Hui, the European Go champion who was ranked 633rd in the world at the time. The current version of AlphaGo not only plays more aggressively. It makes fewer mistakes. A Creative Move
Then, with its 19th move, AlphaGo made an even more surprising and forceful play, dropping a black piece into some empty space on the right-hand side of the board. Lee Sedol seemed just as surprised as anyone else. He promptly left the match table, taking an (allowed) break as his game clock continued to run. “It’s a creative move,” Redmond said of AlphaGo’s sudden change in tack. “It’s something that I don’t think I’ve seen in a top player’s game.” When Lee Sedol returned to the match table, he rook an usually long time to respond, his game clock running down to an hour and 19 minutes, a full twenty minutes less than the time left on AlphaGo’s clock. “He’s having trouble dealing with a move he has never seen before,” Redmond said. But he also suspected that the Korean grandmaster was feeling a certain “pleasure” after the machine’s big move. “It’s something new and unique he has to think about,” Redmond explained. “This is a reason people become pros.” [Note]
A New Autonomy
During Game One, match commentators Michael Redmond and Chris Garlock didn’t seem to understand how AlphaGo operated. Redmond kept referring to AlphaGo’s “database” of moves—something it doesn’t really have. Once AlphaGo is trained using those machine learning techniques, it plays entirely on its own. By Game Two, Redmond and Garlock were wise on this, after some coaching from the DeepMind team over breakfast at the Four Seasons. During Game Two, the commentators even invited DeepMind research scientist Thore Graepel onto their stage to explain the system’s rather autonomous nature. “Although we have programmed this machine to play, we have no idea what moves it will come up with,” Graepel said. “Its moves are an emergent phenomenon from the training. We just create the data sets and the training algorithms. But the moves it then comes up with are out of our hands—and much better than we, as Go players, could come up with.”
Maximizing Probability
After AlphaGo’s rather unexpected play on his 19th move, the Google machine was very much the aggressor. But then things tightened up, with Lee Sedol commanding some notable territory. Match commentators were unable to make a real call on who was ahead and who was behind.
This reflects another aspect of the machine learning technology that underpins AlphaGo. As Graepel explained, AlphaGo does not attempt to maximize its points or its margin of victory. It tries to maximize its probability of winning. So, Graepel said, if AlphaGo must choose between a scenario where it will win by 20 points with 80 percent probability and another where it will win by 1 and a half points with 99 percent probability, it will choose the latter. Thus, late in Game One, the system made some moves that Redmond considered mistakes—“slow” in his terminology. These moves seemed to give up points, but from where Graepel was sitting, AlphaGo was merely trying to maximize its chances.
Near the two-hour mark in Game Two, Lee Sedol made a move in the top left-hand corner, at the heart of the territory commanded by AlphaGo, and Redmond said: “Things are really going to get fun now.” The Korean was back on the offensive, and Redmond predicted a “close game”—something he said so often during Game One. “My assessment will likely be changing,” Redmond later said, “with every move.”
Overtime at Speed
Unlike Game One, when Lee Sedol resigned after about three and a half hours, the Korean kept playing as the match approached its fourth hour. At the three and a half hour mark, Redmond felt that Lee Sedol might have a small territorial advantage. Shortly thereafter, the grandmaster’s clock ran out, which meant he was forced to play each of his remaining moves in under 60 seconds, while AlphaGo still had 20 minutes and 20 seconds left. But Redmond believed that Lee Sedol had made most of his major decisions and that he could easily play out the game at speed. (Redmond’s belief here might be a ‘very naive belief’) The result was a rapid-fire end game. Lee Sedol began rocking back and forth in his chair during his first 60 seconds of overtime and continued to do so even after he made his move. In terms of territory, the Korean seemed to hold his own. But time was indeed an issue. Twice, he let his 60 second clock run out (on the third time, his window goes down to 30 seconds), an indication that we was still unsure how things should play out. “I have a feeling—and maybe Lee Sedol has a feeling—that black is ahead,” Redmond said. Lee Sedol started to rock again, and the other English commentator, Chris Garlock, insisted he saw sweat on the Korean’s brow. Then the grandmaster began punctuating his moves with an almost despondent shaking of the head. As the match stretched well into its fourth hour, AlphaGo entered overtime as well, and both players were limited to 60 seconds per move. The pace picked up again, before only for so long. Just a few minutes later, Lee Sedol resigned.
The complex, tense 2nd game ran nearly five hours. The 3rd game will be on Saturday, March 12, (Friday night 8p PST, 11p EST) when the Korean will play black and move first. Friday is a rest day for the two players. That favors the Korean. The match will be livestreamed on DeepMind’s YouTube channel with commentary by Redmond and Garlock. [Note]: Back in 1997, during its second match with Gary Kasparov, IBM Deep Blue made a similar move very early in the second game. But it didn’t give Kasparov much pleasure. Kasparov was completely flummoxed, and much to the surprise of the chess world, he soon resigned the game. It highlighted a certain advantage that machines carry in a match like this. They don’t get upset. And they can rile opponents simply by doing something that no human would do—or at least that no human would anticipate from a machine. “Computers are able to make moves that are unexpected by people,” says Murray Campbell, who was part of the team that built Deep Blue and is closely watching this week’s match back in the States. References: Google's DeepMind beats Lee Sedol again to go 2-0 up in historic Go series Human ingenuity beats human intuition again By Sam Byford on March 10, 2016 03:26 am http://www.theverge.com/2016/3/10/11191184/lee-sedol-alphago-go-deepmind-google-match-2-result AlphaGo Scores Another Win to Go Up 2-0 Against Lee Sedol in DeepMind Challenge Match Mar 10, 2016 05:41 am | Chris Garlock AlphaGo-Lee Sedol Match Draws Global News Coverage Mar 09, 2016 04:39 pm | Chris Garlock Google’s AI Wins First Game in Historic Match With Go Champion Master of Go Board Game Is Walloped by Google Computer Program Google’s software beats human Go champion in first match Google’s DeepMind AI makes history by defeating Go champion Lee Se-dol Google’s DeepMind defeats legendary Go player Lee Se-dol in historic victory Google’s AlphaGo AI defeats human in first game of Go contest Google’s AlphaGo Wins First Match Against Go Grandmaster Lee Sedol Google DeepMind’s AlphaGo beats Go champion Lee Sedol in AI milestone in Seoul ‘I’m in shock!’ How an AI beat the world’s best human at Go
|
|
| ( 不分類｜不分類 ) |
