News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
Let AI reach the human level? First of all, we need to cross the five mountains"
Let AI reach the human level? First of all, we need to cross the five mountains"
Google's London based artificial Intelligent Company Deep Mind made a significant breakthrough in machine learning last week, announcing that AlphaGo Zero can learn autonomously without human intervention. Artificial intelligence has been able to do a lot of things that human beings can not do, but how far are we from artificial intelligence like "human beings"? What are the key issues that we need to address before we reach our goals?
To answer this question, I will argue that in order to make machines as smart as humans, five milestones must be overcome: universality, transfer learning, autonomous learning, common sense, and self-awareness. Let's look at the development of AI in turn.
Universality: this means that we develop a method, or a system architecture that can be applied to any other domain independent problem. I think this problem has been solved to a great extent. The probabilistic methods of artificial intelligence, such as deep tilt networks (e.g., expert systems), have proved their versatility. We can use the same depth of learning networks and algorithms to solve almost all problems - this is a good example of machine learning.
Common sense: This is a very difficult problem. For example, "Michael Phelps won the 400 meter butterfly gold medal at the Beijing Olympics."". When you read this sentence, you will immediately and vaguely think of a long list of things; for example, Phelps is wet when he gets the medal, he must take off his socks before entering the swimming pool and so on. The connection between the logical hypothesis and the original statement is extremely difficult to code in the computer. We still have a long way to go to solve common sense problems. But a good beginning is the study of neuroscience, which teaches us how to form, retain and use memory. The function of human memory may be the key to developing common sense in machines.
Self consciousness: human self consciousness is still a mystery, although neuroscientists have made breakthrough progress, when we realize that some things, such as "I" or "self", when we have the subjective experience, what happens to the brain. For many people, the high level of consciousness may be the last bastion of human beings to maintain some advantages in the face of future intelligent machines". However, it is not impossible to create a machine that mimics self consciousness. The reason I say "imitation" is that unless we find an objective way to measure human consciousness, we will never be able to determine whether a machine is "really" conscious. Machines that allow us to believe that they have self or personality should be relatively easy to develop, but whether they will really have self-awareness, we can only know whether we can solve the problem of consciousness".
To answer this question, I will argue that in order to make machines as smart as humans, five milestones must be overcome: universality, transfer learning, autonomous learning, common sense, and self-awareness. Let's look at the development of AI in turn.
Universality: this means that we develop a method, or a system architecture that can be applied to any other domain independent problem. I think this problem has been solved to a great extent. The probabilistic methods of artificial intelligence, such as deep tilt networks (e.g., expert systems), have proved their versatility. We can use the same depth of learning networks and algorithms to solve almost all problems - this is a good example of machine learning.
Autonomous Learning: This is the achievement of Deep Mind's "AlphaGo Zero". Through the adjustment and simplification of the original reinforcement learning method to the first use of AlphaGo, they demonstrated the neural network of a given target (such as "win") can be their own learning, and to achieve this goal and strategy of invention. It's a big breakthrough, and it's a step closer to artificial intelligence.
Common sense: This is a very difficult problem. For example, "Michael Phelps won the 400 meter butterfly gold medal at the Beijing Olympics."". When you read this sentence, you will immediately and vaguely think of a long list of things; for example, Phelps is wet when he gets the medal, he must take off his socks before entering the swimming pool and so on. The connection between the logical hypothesis and the original statement is extremely difficult to code in the computer. We still have a long way to go to solve common sense problems. But a good beginning is the study of neuroscience, which teaches us how to form, retain and use memory. The function of human memory may be the key to developing common sense in machines.
Self consciousness: human self consciousness is still a mystery, although neuroscientists have made breakthrough progress, when we realize that some things, such as "I" or "self", when we have the subjective experience, what happens to the brain. For many people, the high level of consciousness may be the last bastion of human beings to maintain some advantages in the face of future intelligent machines". However, it is not impossible to create a machine that mimics self consciousness. The reason I say "imitation" is that unless we find an objective way to measure human consciousness, we will never be able to determine whether a machine is "really" conscious. Machines that allow us to believe that they have self or personality should be relatively easy to develop, but whether they will really have self-awareness, we can only know whether we can solve the problem of consciousness".