下载掌阅APP，畅读海量书库
立即打开
新的AlphaZero现在不但会下围棋，还会下国际象棋。
测试中可能遇到的词汇和知识：
prior[ˈpraɪə(r)] adj.之前的
drosophila[drɒˈsɒfɪlə] n.果蝇
workhorse[ˈwɜ:khɔ:s] n.驮马
amino-acid[æ'mɪnəʊ'æsɪd] n.氨基酸
crystallography[ˌkrɪstəˈlɒgrəfi] n.结晶学
cryo-electron microscopy n.低温电子显微镜
阅读马上开始，建议您计算一下阅读整篇文章所用的时间，对照下方的参考值就可以评估出您的英文阅读水平。
-----------------------------------------------------
DeepMind, the London-based artificial intelligence company, has announced two research successes — the strongest game-playing AI so far and the world’s most powerful predictor of the shape of protein molecules — as it returns to its scientific roots after agreeing to transfer its health division to parent company Google.
In the journal Science, DeepMind presents AlphaZero, a development of its celebrated AlphaGomachine. With no prior knowledge or data input beyond the rules of the game, AlphaZero taught itself to play not only Go but also chess and shogi (Japanese chess) at a level that defeated all human players and the best existing game-playing computer programs.
Chess masters enthused about AlphaZero not only as a winner but also for its dynamic and unconventional style. “It was fascinating to see how AlphaZero’s analysis differed from the top chess engines and even top grandmaster play,” said Natasha Regan, Women’s International Master.
Garry Kasparov, the former world chess champion famously beaten in 1997 by IBM’s Deep Blue, wrote an editorial in Science praising AlphaZero’s “dynamic open style like my own”.
“The conventional wisdom was that [chess] machines would approach perfection with endless dry manoeuvring,” Mr Kasparov said. However AlphaZero preferred risky-looking positions over more conservative moves that kept as many pieces as possible in play.
Playing 1,000 games against Stockfish, the current champion chess engine, AlphaZero won 155 games and lost just six. Stockfish is built on thousands of rules crafted by strong human players that try to account for every eventuality in a game. AlphaZero, in contrast, uses a neural network to learn chess from scratch, knowing only the rules of the games, by playing millions of games against itself and learning winning moves.
Games are an ideal test bed for AI — or, as Mr Kasparov put it, “a drosophila of reasoning”, referring to the fruit fly that became a model organism for genetics research. They provide excellent publicity for DeepMind but have negligible sales potential.
The Streams business, which is developing an AI-based assistant for nurses and doctors, is set to bring in revenues from health systems but DeepMind announced last month that it would be transferred to its Google parent for global development. DeepMind’s income so far has come mainly for internal work for Google, such as WaveNet which improves voice synthesis and is being adopted across Google products.
But DeepMind, which Google bought for £400m in 2014, regards developing AI to “drive and accelerate new scientific discoveries” as a particularly promising field, said Demis Hassabis, DeepMind chief executive.
The “first significant milestone” in this quest was achieved this week when its AlphaFold system won an international competition to predict the three-dimensional structure of proteins from their genetic sequence.
How proteins, the workhorse molecules of life, fold their long chains of amino-acid building blocks into a compact 3D shape remains one of the most important unsolved problems in biology, because knowing a protein’s shape is key to understanding its function.
Analytical techniques such as X-ray crystallography and cryo-electron microscopy can sometimes give the answer but a way to find out the structure without having to carry out expensive and time-consuming experiments would give a huge boost to biomedical research.
“DeepMind’s work on applying AI to this longstanding problem in molecular biology is definitely an advance,” said David Jones, a biomedical data specialist at the Francis Crick Institute. “One eventual goal will be to determine accurate structures for every human protein.”
This article has been amended to say Garry Kasparov was famously beaten in 1997 by IBM’s Deep Blue, not DeepMind.
请根据你所读到的文章内容，完成以下自测题目：
A.The strongest game-playing AI so far
B.Deepmind's health division
C.A chess engine
D.It is the same as AlphaGo
答案 (1)
A.AlphaZero can teach itself to play games.
B.AlphaZero is not developed by Deepmind.
C.AlphaZero is a search engine.
D.AlphaZero needs large amount of training data.
答案 (2)
according to Garry Kasparov?
A.Endless dry maneuvering
B.Static conservative style
C.Dynamic open style
D.Elegant slow moving
答案 (3)
A.Management
B.Gaming
C.Weather forecasting
D.Molecular biology
答案 (4)
(1) 答案：A解释：DeepMind, the London-based artificial intelligence company, has announced two research successes — the strongest game-playing AI so far and the world’s most powerful predictor of the shape of protein molecules — as it returns to its scientific roots after agreeing to transfer its health division to parent company Google.
(2) 答案：A解释：With no prior knowledge or data input beyond the rules of the game, AlphaZero taught itself to play not only Go but also chess and shogi (Japanese chess) at a level that defeated all human players and the best existing game-playing computer programs.
(3) 答案：C解释：Garry Kasparov, the former world chess champion famously beaten in 1997 by IBM’s Deep Blue, wrote an editorial in Science praising AlphaZero’s “dynamic open style like my own”.
(4) 答案：D解释：DeepMind’s work on applying AI to this longstanding problem in molecular biology is definitely an advance