Home > News content

Facebook AI R & D director: blindly imitate the human brain will hinder the development of AI

via:博客园     time:2017/9/19 17:31:04     readed:668

Facebook AI research director pointed out that blindly imitating human brain development AI will hinder the development of AI. Many industry experts also have similar arguments, they believe that the focus of AI research should not be to imitate the working mechanism of the brain, but should be to explore the fundamental principles of intelligence.

Some people think that the Creator is in accordance with their own image to create human beings. And artificial intelligence may be the greatest invention of mankind, mankind has been trying to create AI according to their own. The most typical way is to reconstruct the human brain in digital form. However, top scientists do not agree with this approach, that trying to completely simulate the human brain is a waste of time. Look for inspiration from somewhere else.

"We do not really understand the human brain," says Janna Levin, an astronomer at Barnard College. We tried to explore the brain, but did not work. Levin recently led a roundtable discussion of artificial intelligence and ethical prospects.

"We do not understand the human brain, how to create artificial brain?" & rdquo;

Experts who participated in the roundtable discussion said it was a tricky issue. Human beings can not completely simulate the brain. Therefore, humans should take the time to unlock the fundamentals of intelligence.

MIT Future Life Research Institute director Max Tegmark that excessive attention to the human brain is a kind of "carbon chauvinism". Although scientists do not fully understand the mechanism of brain operation, but it can be said that this is not amazing.

Tegmark said, "We are too obsessed with the functioning of the brain." This is a lack of imagination. & rdquo;

Historically, his view was true. In the Victorian era, an engineer called Cl & eacute; ment Ader was inspired by the bats and built the first flying machine.

Ader has been flying hundreds of meters without the control of the device. But why is the White Brothers famous, and he is not known?

Tegmark believes that this is because the White brothers really from the scientific point of view of the flight principle, and Ader is always in the imitation of biological. The White Brothers develops aircraft by constructing wind tunnels and tests, rather than just imitating creatures. The same applies to artificial intelligence.

But even in accordance with the image of human beings to create artificial intelligence, nor is the most correct path. The roundtable discussion will always return to the topic of biology.

Facebook's director of AI research and development Yann LeCun that "you can get inspiration from biology, but you do not want to simply copy." From an engineering point of view, the evolutionary process will be extremely difficult. & ldquo;

After millions of random mutations, humans can reproduce and we can appear here. This is an identity. And the brain's intelligence and reasoning ability to maximize or simplify, are unwise.

The functioning of the human brain is extremely complex. When the baby is still in the mother, the brain can assemble it. In the human life, the brain also has a self-repair function. The machine does not need these mechanisms. The machine only needs to receive and process the data and learn.

There are two ways to supervise operation and to strengthen learning to learn from artificial intelligence.

LeCun explained that for more traditional supervised learning, humans must enter thousands of examples, the machine can really start work. For example, the image recognition algorithm must first see countless apples in order to achieve accurate identification.

Another way is to strengthen learning. AI algorithms and neural networks can be trained with each other. Usually this way only applies to the game. A chess AI can be at any time with their own millions of games, to learn the chess game in the subtle details.

But these two methods are not perfect, can not really teach AI to know the world. Human beings are still supervising the study to pick the giants; chess AI in addition to the chess other than know nothing.

"We are training the neural network in a very stupid way," LeCun said. & rdquo;

Babies can understand the object forever when they are two months old. At the age of six, you can intuitively understand how the material world works. But so far, humans can not make the machine a similar unsupervised learning.

If someone can achieve unsupervised learning of the machine, these people are likely to be LeCun and his team. Only the giant company has enough resources and architecture to train advanced neural networks. But during the roundtable discussion, LeCun just shrugged and said that we had no idea yet. & rdquo;

For the AI, the biological basis is vital. This foundation is not the complete reconstruction of the human brain. The human brain is a scientific miracle, but not the only answer. AI developers need to keep in mind that humans themselves are not very special, they should try to find some new inspiration.

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments