评论
打印
Google’s computer program AlphaGo has defeated the world’s best Go player in the second out of three games, scoring an overall win for the artificial intelligence algorithm in the fiendishly complex board game.
谷歌(Google)电脑程序AlphaGo在三局比赛的第二局战胜世界排名第一的围棋选手，意味着人工智能算法在这种极其复杂的棋盘游戏领域大获全胜。
Chinese champion Ke Jie gave up on Thursday after both the computer and human surprised the audience with their innovative strategies.
中国围棋冠军柯洁周四投子认负，这场对弈中电脑和人脑都以创新的战术给观众带来了惊喜。
Although the ancient board game of Go has few rules, its complexity means it had been considered unimaginable for a computer to beat a human champion. However, AlphaGo, developed by Google’s DeepMind unit, beat the South Korean grandmaster Lee Se-dol last year.
尽管古老的围棋规则很少，但其复杂性意味着电脑击败世界冠军曾被认为是难以想象的。然而，由谷歌旗下的DeepMind公司开发的AlphaGo去年击败了韩国围棋大师李世石(Lee Se-dol)。
Google brought AlphaGo to China in order to challenge Mr Ke and showcase its AI breakthroughs in front of local experts and government officials at a five-day tournament. The competition is the company’s biggest event in collaboration with the Chinese government since its search engine was blocked seven years ago.
谷歌将AlphaGo带到中国的目的是，挑战柯洁并在为期5天的赛事中在中国专家和政府官员面前展示其在人工智能领域的突破。自其搜索引擎7年前在中国被屏蔽以来，这是谷歌与中国政府合作举办的规模最大的活动。
This year’s version, called AlphaGo Master, is even stronger than last year’s, explained Demis Hassabis, co-founder of DeepMind.
DeepMind联合创始人杰米斯?哈萨比斯(Demis Hassabis)解释道，今年的版本名为AlphaGo Master，比去年的版本更为强大。
AlphaGo Master is able to learn from playing itself, rather than needing to learn from human games. It is also more efficient, requiring 10 times less computing power than the previous version.
AlphaGo Master能够从与自己对弈中学习，而不是需要从人类比赛中学习。其效率也更高，需要的计算能力是上一个版本的10分之一。
“The main innovation is that AlphaGo has become its own teacher,” said Dave Silver, lead programmer on AlphaGo. “This makes it learn in a much more general way, meaning it does not only play Go better, but could lead to more general applications.”
“主要的创新是AlphaGo已变成自己的老师，”AlphaGo首席程序员戴夫?西尔弗(Dave Silver)表示，“这令其能够用更一般化的方式学习，意味着不仅棋艺增强，而且还能带来更一般化的应用。”
AlphaGo Master also runs on better hardware. It uses “tensor processing units”, microchips that Google has designed for use with AI programs. According to Google, TPUs can perform up to 80 times better per unit of energy used compared with conventional chips.
AlphaGo Master的硬件也更好了。它使用“张量处理单元”，这是谷歌为用于人工智能程序而设计的微芯片。据谷歌称，“张量处理单元”每单位能量的性能优于常规芯片最多80倍。
In a case of man imitating machine, Mr Ke started using strategies learnt from watching AlphaGo’s matches.
柯洁也开始使用从观看AlphaGo比赛中学到的战术，表明人类也在模仿机器。
Although IBM’s Deep Blue beat the chess champion Garry Kasparov in 1997, advancements in cracking Go have been more difficult.
尽管IBM的“深蓝”(Deep Blue)计算机在1997年击败了国际象棋冠军加里?卡斯帕罗夫(Garry Kasparov)，但要在攻克围棋领域取得进展更为困难。
Unlike chess, Go cannot be won easily by using brute power to work out all the possible moves. Instead, AlphaGo uses styles of learning that mimic the way neurons work in the human brain.
与国际象棋不同，围棋不可能通过利用纯粹的计算能力计算出全部可能的走法来轻易取胜。AlphaGo采用的学习模式是模仿神经元在人类大脑中的工作原理。
DeepMind is now working on developing programs to play games that require more complex visual input, such as Blizzard Entertainment’s online strategy game Starcraft.
DeepMind现在正致力于开发程序来玩需要更复杂视觉输入的游戏，例如暴雪娱乐(Blizzard Entertainment)的在线策略游戏《星际争霸》(Starcraft)。
