(Original title: People cannot defeat AI on Go, but people can invent Go.)
Columnist @ Pan Yiming Original Release
Al JaGo has already ended the topic of AI and has begun to be widely discussed. The AI ​​in people's eyes is more likely to be the thinking and extremely powerful existence such as Terminator and Western World. Those who do not understand AI either underestimate AI or overestimate AI. This kind of argument belongs to the latter.
1. Does AI think?
Before answering this question, share an answer. Knowing that I answered a question - does AlphaGo "understand" Go?
The answer is as follows:
A thing that looks like a duck and quacks like a duck, then this is a duck.
How do we judge people to understand Go? You can play chess with your opponent and understand the rules. There is a detailed operation of the layout, according to opponents targeted law, know how to win the game. So we think people understand Go. Machines can do the same thing, even better than people. So the machine also understands Go.
Only people do not understand the machine. Machine learning algorithms, SVMs, and neural networks are also good. It is not that the machine does not understand the entered sample. Instead, people do not understand how to express the logic of the machine's output, expressed in terms of logic that people can understand.
AI thinks about Go. However, does AI think? I tend to think that the current AI does not think, or from another perspective, the current AI will not think like humans.
2. The learning mechanism of people and AI
How do people learn? As a senior rookie game player at DotA, when I first played in the glory of the king, I chose a hero and saw a skill introduction. I got very good data in the first game. Before this, I have not seen any video or introduction of the glory of the king. And the entire learning process is only a few minutes.
How does AI learn? If there is a dota that crushes professional AI and wants to learn about the king's glory, the previously trained model will momentarily fail. At this time, only a large number of professional players' video games can be obtained for algorithm training. The entire learning process requires a lot of computing resources and requires a long training assessment.
This shows that humans have evolved as a biological algorithm for 4.6 billion years in nature. Compared with Infant's AI algorithm, it has an absolute advantage: the ability to quickly transfer learning. At present, AI does not have this capability, and it is impossible to have this capability within a predictable time.
3. Human and AI information processing mechanisms
There is no breakthrough in brain science research. We do not know how humans think. In the field of machine learning, after the algorithm was upgraded from the rule algorithm to machine learning, we did not know how the machine thinks.
But if people and machines are used as a black box, we find that people and machines are different systems and their information processing capabilities are very different.
In the new scene, people get very little information and they can adapt quickly. For example, Robinson drifts away from a desert island and can adapt quickly and survive. Machine learning does not work in the absence of clear data and rules.
In the stable scenario, with a large amount of feedback data, people's decision-making ability is not as good as AI. This is also the determination of individual human brain power. In the face of large amounts of data, people will summarize the data, analyze the data, and extract a large number of rules from the data. Use rules to make quick decisions when you encounter problems.
The process of summarizing data and analyzing data is itself the process of information loss. In other words, getting 1,000 data and getting 10,000 data may not be significant for people. Computers, based on statistical algorithms and learning models, have the largest amount of information stored, the amount of data continues to grow, and the efficiency of decision-making continues to increase.
In summary, people can perform well in a new scenario with no clear goals through complex analogy, association, and other strategies. AI can surpass human decision-making in a clear scenario and with sufficient data feedback.
4. What kind of human work will be replaced
What are the areas where artificial intelligence will replace? This is a cliché problem.
In summary: For long-term fixed-type mental workers, the decision-making data comes from online databases or can be collected in large quantities by online data, and the work effect can be feedback or recorded on the data on the cable.
In these industries, you need to make judgments based on the data. Due to data overload, a large number of data can not be fully utilized in the decision-making process. Only some rules in the data can be mined and solutions can be made. In this field, people are easily replaced.
Such as operating schedules, advertising, material design, doctor's diagnosis, and so on.
In the Internet, activity scheduling usually depends on the manual configuration of operations. In the face of personalized recommendation engines, the activity of thousands of people is inherently flawed. The personalized recommendation engine can stage all historical user behaviors and give optimal results. Material design Taobao is currently on the line, although there may be obvious Badcase, but according to the historical materials of each person's banner material, the drainage effect is definitely better than the banner style of thousands of people.
Advertising, or financial analysis, is the result of the need to analyze history, as well as continuous optimization of process data, and efficient processing of data is the strength of the algorithm.
In the field of doctors' diagnosis, a large number of examination data needs to be handled effectively, and doctors are required to have enough background knowledge. No doctor, no matter how full his experience, can not understand all the cases. On the one hand, artificial intelligence knowledge reserves and case reserves far exceed people. On the other hand, the ability to process and identify images has also slowly increased. Artificial Intelligence Through a large number of samples, the current state of the art of artificial intelligence has surpassed the level of human doctors in some areas.
Of course, with the increasing degree of informationization, the development of semantic recognition image recognition, and the improvement of computer information collection capabilities. More and more fixed may be replaced directly: simultaneous interpretation, financial analysis, autopilot and so on.
5. Where is the person's position in the future of AI popularity?
According to the current computer computing capabilities and the development of AI algorithms, AI will quickly become popular in the short term, and it will infiltrate into industries with relatively high degree of informatization and high labor costs. For example: Internet advertising, Internet operations, finance, medical care and so on. But at the same time, people in these industries will not completely disappear, but they will need fewer people and need more skills from these people.
Before the age of mechanization, when farmers farmed land, a person could take care of the land for a few acres. It would be enough if he could turn to sowing and weeding and fertilizing. And now modern farmers. One can take care of the hectare-level land. But it is necessary to drive a harvester, a tractor, use mechanical means to fertilize, and sow.
The invention of the computer, numerous mathematicians physicists from the cumbersome calculations to break away, and explore new areas. Future people need to constantly discover and explore new scenes through creativity, while AI can use data to reach the ultimate in defined scenarios.
Although people cannot overcome AI in medical diagnosis, people can research new diagnostic techniques and methods. While people can't beat personalized recommendations when ranking feeds, people can explore new user scenarios and introduce new content models. A specific medical diagnosis, a certain type of feed flow, is a board, and people cannot compete with the AI ​​on this board, but it is not replaced. What humans need to do is not to defeat AI, but to use their unique thinking skills to constantly explore the use of AI.
People cannot defeat AI on Go, but people can invent Go.
燑br>
Outdoor Bluetooth Speaker,Sound Equipment,Active Pa Speaker,Professional Speaker
NINGBO RFUN AUDIO TECHNOLOGY CO.,LTD , https://www.mosensound.com