阅读理解
AlphaGo is a computer program that plays the board game Go.
In March, 2016, the pride of humankind was crushed (粉碎) by a computer. Google's AlphaGo defeated the South Korean grandmaster (围棋大师) Lee Sedol four games to one, as the world looked on with shock and awe (敬畏). Artificial intelligence (AI, 人工智能) had suddenly reached a new and unexpected height.
But as smart as AlphaGo is, it's no longer the best Go “player” in the world. Google's artificial intelligence group, DeepMind, has created the next generation of its Go-playing program, called AlphaGo Zero. The new AI program is unique in the way it learned to play Go. Instead of learning from thousands of human matches, as its predecessor (前任) did, AlphaGo Zero mastered Go in just two days without any human knowledge of the game and defeated AlphaGo by day three, reported The Guardian. It then went on to defeat AlphaGo 100 games to zero.
To learn how to play Go, AlphaGo Zero played millions of matches against itself using only the basic rules of the game to rapidly create its own knowledge of it. Like the previous version, it used “reinforcement (增强) learning to become its own teacher,” according to DeepMind's website.
“It's more powerful than previous approaches,” David Silver, AlphaGo's lead researcher, told The Guardian, “because by not using human data, or human expertise in any fashion, we've removed the constraints (约束) of human knowledge and it is able to create knowledge itself.”
AlphaGo Zero's approach to self-learning is a significant advancement in AI that could be applied to help solve some of the world's biggest problems, according to a recent research report published in the journal Nature. For example, DeepMind co-founder Demis Hassabis argues that AlphaGo Zero could probably find cures for a number of serious diseases within weeks, according to The Telegraph. Indeed, the AI is now being used to study protein folding, which is connected to diseases such as Parkinson's and Alzheimer's.
So now that AI has exceeded (超过) the bounds of human knowledge, perhaps the question is not about what AI can learn from humans, but what humans can learn from AI. We can only wait and see.
