下载掌阅APP，畅读海量书库
立即打开
突破性的游戏使用图像来帮助人工智能与人沟通。
测试中可能遇到的词汇和知识：
pit against | 放到对立面
complex [ˈkɒmpleks] adj.复杂的
collaboration [kəˌlæbəˈreɪʃn] n.协作
scratch [skrætʃ] v.挠
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。
The recent history of games that pit man against computer, from IBM's Deep Blue to DeepMind's AlphaGo, has matched human brain power against machines in increasingly complex settings.
Now comes a game that claims to break new ground — this time with a machine intelligence that tries to communicate and collaborate with people, rather than crush them.
Its backers claim that, by setting a new direction for artificial intelligence, it points the way to solving some of the limitations in deep learning, the technique that currently dominates the field.
The game, which goes live online on Tuesday, is somewhat like Pictionary, with one side drawing a picture and the other trying to guess what it represents. The computer plays either part. Developed by the AI institute of the late Microsoft co-founder Paul Allen, it is designed to get people and machines to interpret each other, using guesses and hints when necessary, to arrive at a shared understanding.
“A lot of the [AI] research work hasn’t really explored collaboration,” said Oren Etzioni, head of the Allen Institute for Artificial Intelligence. “How do you build collaborative systems? We see this as a test bed for debugging interaction.”
The AI used in the game, called Iconary, was trained on 100,000 drawing games between human players. The computer recognises 1,200 different icons, which are combined randomly into new phrases for one side to draw and the other to guess — like “coughing in a house” and “man lifting the weight”. The result is a vast number of possible clues that the computer has not seen before, said Ani Kembhavi, one of the Allen researchers.
Future systems built along these lines could have “the ability to abstract common sense knowledge and use it for task completion”, he added. That could include digital assistants and home robots that are much better at interpreting their owners’ desires than today’s voice-powered systems.
A key to developing systems that interact more naturally with people involves training them on more human examples, said Mr Etzioni. By contrast, systems such as DeepMind’s AlphaGo, which beat the best human Go player three years ago, learn by playing a vast number of games against themselves. The machine-to-machine training produces a huge amount of data to refine the computer’s pattern recognition but does nothing to help it understand or collaborate with humans, he added.
The system, called AllenAI, uses natural understanding and image recognition, as well as techniques that the researchers said give it a degree of “common sense” in interpreting its fellow player. Teaching machines common sense is a field that has barely been scratched in computer science, and recently became the focus of an advanced research programme funded by the US military.
By making Iconary generally available online, Mr Kembhavi said the researchers hoped to increase substantially the number of human/machine interactions they could study in order to refine the algorithms further.
请根据你所读到的文章内容，完成以下自测题目：
A.Game theory
B.Brain hack
C.Adaptive learning
D.Deep learning
答案 (1)
A.It is a drawing game.
B.It is trained on a much larger data set.
C.It gets people and machines to interpret each other.
D.It is the first AI game.
答案 (2)
A.Facial recognition.
B.Distributed system.
C.Some "common sense".
D.Quantum computing.
答案 (3)
(1) 答案：D. Deep learning解释：Its backers claim that, by setting a new direction for artificial intelligence, it points the way to solving some of the limitations in deep learning, the technique that currently dominates the field.
(2) 答案：C. It gets people and machines to interpret each other.解释：Developed by the AI institute of the late Microsoft co-founder Paul Allen, it is designed to get people and machines to interpret each other, using guesses and hints when necessary, to arrive at a shared understanding.
(3) 答案：C. Some "common sense".解释：The system, called AllenAI, uses natural understanding and image recognition, as well as techniques that the researchers said give it a degree of “common sense” in interpreting its fellow player.