Lei Feng Network (search "Lei Feng Net" public concern) press: Volkswagen seems to be more concerned about some of the more frightening and subversive (whether good or bad) topic, the performance in the field of artificial intelligence, is the public more Many tend to believe in topics such as "AI annihilation". At the same time, this actually means that everyone has too optimistic expectations of AI. Historically, the low tide of artificial intelligence has occurred due to the blind optimism of AI and the disappointment that the actual situation has failed to achieve expectations. Therefore, while some academics have warned us to be careful about the threat of AI, there are also more scholars who want to constantly remind us that AI is not a scourge, and the current AI is far from reaching the level that will replace us.
At the first speech and smart technology summit held on the 28th, Hong Xiaowen, Dean of Microsoft Asia Research Institute explained to us in detail how he understands AI, intelligence, and how we think of AI. AI problem.
What is intelligence?Hong Xiaowen believes that intelligence consists of four levels of definition. These layers are defined at different levels and represent different levels of "smartness."
The first level is Capability, which is basically uncontroversial. The memory of the machine, the frequency of the CPU, the cache, and the pixels of the camera are all real data. It is a very objective thing.
The second level is Intelligence, but the concept of "smart" has always been changing. Hong Xiaowen said that he still remembers the first “smart TV†he had bought at the time. It was a long time ago that a brand called Janice had added a remote controller to the TV for the first time, and then called it. That TV is called a smart TV. Did that count as a smart TV? Actually, this is really because there were no remote controls for all the TVs before. You want to change channels and you have to go to the front of the TV to change. It is very troublesome, so such an improvement in the past is enough to make people believe that it is a smart TV, but today, I think no one will think that there is a remote control called smart TV.
Today no one thinks that computing power is a manifestation of intelligence. Hong Xiaowen said that even playing chess is the same. Chess is more than anyone else. Of course, AlphaGo already has some reasoning stuff, but it's mainly about calculating ability.
By the third and fourth levels of Intellect and Wisdom, the definition is even more ambiguous. Humans often have to make decisions. Sometimes these decisions are not very intelligent, but one of the key points is that many times people’s decisions are made when data is insufficient, and in this case they still can Maintain a fairly high accuracy. If there is enough data, many computers can make the right decision, but once the data is insufficient, the computer will immediately misfiring.
An extreme example of this is Gravitational Wave: The first time this year people barely detected gravitational waves. This is a big news this year, but this thing was predicted by Einstein as early as a hundred years ago. . How can data be available at that time? Don't say small data, no data at all. And people can make predictions so accurate without data at all. This is where humans are bad.
How should we treat intelligence?
To analyze our brains, Hong Xiaowen said that our brain is divided into two hemispheres. Over the years, the research results of psychologists and brain scientists show that the division of labor between the left and right brains is different. The left brain is more responsible for logic, and the right brain is more responsible for intuition. The left brain has stronger analytical ability, and the right brain has stronger integration ability. The left brain is "I know" and the right brain is "I believe."
In fact, the computer is the best left brain, better than the left brain, because no one will compare computing power with the computer. But there is a very important thing in the computer called algorithm, what is the algorithm? Hong Xiaowen said that the algorithm is a step-by-step set of operaions to solve a problem in accordance with the established order. According to this definition we can derive the definition of creativity. What is creativity? Hong Xiaowen said that his definition is: Solve the problems that have not yet been resolved. For example, Fermat’s Theorem, if you can solve it before the theorem is solved, or the problem you can solve is better than the existing method, then you have a strong creativity.
So we can see that the computer has no creativity at all. The current computer has only calculations, and the algorithms it uses to calculate all come from people. So whenever the machine defeats humans once, I don't think computers are so strong. I think people are so strong that they can design such a powerful algorithm.
Nowadays, robots' victory over people in many fields is totally unsuccessful. For example, let you and your computer calculate the sum of 1,2,3...100 at the same time. We all look at Gauss's algorithm. We think of ourselves as Gaussians. After the first time we heard this question, we went to the question. Solved by clever methods, but we will find that the speed of this calculation is still not as fast as a computer, even if the computer is using the most stupid method, but because of its computational power, it can calculate this result in a flash. Come out, but do we think computers are smarter than people? Obviously not.
Hong Xiaowen said that the computer is really a good thing to be a "left brain." As for strong AI, if we want to be a thing, we must first understand people if we want to approach or even surpass human wisdom and creativity. And we are still not enough to understand ourselves. And, at the end of the day, I would also like to say that we need not need a robot that is very similar to ours. In fact, if the robot is conscious and really intelligent, it means that we have no way to control it (the feasibility of the three laws of robots is actually very low). Hong Xiaowen said that he often gave an example: If we make a robot and want it to help us pour coffee, then we certainly hope that we will call it ten times and it will help us to pour it ten times. But if you ask for a conscious thing, for example, I told my wife to help me pour coffee. My wife may tell me five times with ten times that you have hands and feet. Why don't you go by yourself? Then I can only say yes. That is to say, we want to create a controlled robot. The meaning of control is that it is not self-conscious. I don't want us to build a robot and call it to help me pour coffee. As a result, it says the owner doesn't want to go for it. ? If we have to get something and hope it is the same as us, even smarter than we are, consciously and uncontrollably, then our experience in this area will be too rich. We will have a baby.
Finally, Hong Xiaowen said: We do not need to fear AI, and there is not much need to create a universal AI. Just as he had just analyzed, building an AI is certainly of great academic significance, but in practical terms, it may not be as good as the current AI. Although the current AI is weak AI, it is already It is very useful. The computer helps us collect and process data. We then make final decisions based on it. This type of person+machine is the best combination, which is another meaning of AI development.