Brain Science & Artificial Intelligence - Artificial Intelligence Human Brain Technology

If AI is to evolve further, it needs to be inspired by brain science.

The industry generally believes that the future evolution direction of artificial intelligence is computational intelligence, perceptual intelligence and cognitive intelligence, and the real breakthrough in this period is to make computers have the ability to understand, think and self-learn, and brain science provides an important foundation for the development of brain-like computing systems and devices to get rid of the shackles of traditional computer architecture.

Although artificial intelligence (AI) is developing rapidly, there is always a limit that cannot be crossed, and that is the thinking ability of the human brain. For example, AlphaGo, which has convolutional neural network technology, may not be as good as a baby in terms of other abilities besides playing chess.

There is a growing consensus among scientists around the world that in order to break through the technical barriers of artificial intelligence, it is necessary to make achievements in the field of brain science. Brain science and brain-like research have also been listed as one of the important research directions in the "New Generation Artificial Intelligence Development Plan" in mainland China.

"If AI is to develop further in the future, it needs to be inspired by brain science." As a leader of China's "Brain Project", Pu Muming, academician of the Chinese Academy of Sciences, academician of the National Academy of Sciences of the United States, and director of the Institute of Neuroscience of the Chinese Academy of Sciences, is working hard to promote the integration and development of artificial intelligence and brain science.

At the 2018 Tencent WE Conference held recently, Pu Muming said that how to design new computing models, new devices, chips, and even robots similar to the neural structure of the human brain based on brain-inspired concepts are all problems that need to be solved in the future.

Brain-inspired AI is at the forefront

Can machines become smarter than humans?

In a program called "Super Intelligence", Zhang Jianwei, director of the Cognitive Center of the Shenzhen Institute of Advanced Technology of the Chinese Academy of Sciences, found a blurry portrait of a little girl in surveillance video, and asked machines and police to identify and point out the little girl's parents. As a result, the machine made a mistake, and the police managed to lock up the two pairs of parents through the child's portrait. Obviously, humans have triumphed over machines.

"Machines don't make small mistakes, but they make big mistakes, while the human brain does the opposite. How to further improve and enhance artificial intelligence with reference to the human brain model is a direction of future research. Zhang Jianwei said.

The human brain is one of the most complex systems in the universe, and once it learns one thing, it can be applied to other situations, which is an ability that artificial intelligence cannot match. "The forefront of artificial intelligence is brain-like artificial intelligence, because only the human brain is the highest point of intelligence, and it is the intelligence that has been exerted to the highest limit in the process of evolution." Pu Muming said.

The industry generally believes that the future evolution direction of artificial intelligence is computational intelligence, perceptual intelligence and cognitive intelligence, and the real breakthrough in this period is to make computers have the ability to understand, think and self-learn, and brain science provides an important foundation for the development of brain-like computing systems and devices to get rid of the shackles of traditional computer architecture.

To understand how the brain works, we need an accurate "map" of the brain, known as a map of brain connectivity. The human brain has 100 billion neurons, and the brain connection map is an important foundation for understanding the brain and developing brain-inspired artificial intelligence technology.

At the 2018 Zhongguancun Life Science Park Development Forum held in Beijing, Zhang Xu, academician of the Chinese Academy of Sciences and vice president of the Shanghai branch of the Chinese Academy of Sciences, said that from a global perspective, the brain connectivity map will produce systematic research results in the next few years, so as to analyze the structure of our brains more accurately.

"Whether it is the principle design of artificial intelligence, such as smart chips or intelligent machines, or engineering design, it will be more and more closely coordinated with brain science." Zhang Xu said that the development of intelligent technology is facing new bottlenecks, and it needs to be inspired by brain science and neuroscience. The development of intelligent technologies will also lead to further breakthroughs in brain science, such as deep learning neural network processors, speech recognition, and multilingual translation technology.

How does AI learn from the human brain?

However, how can we develop artificial intelligence by mimicking the human brain when we don't even understand how it works?

"It's too late to think about AI after we've studied the brain well." In Pu's view, the architecture of machine learning networks can also be learned and shaped like brain networks. In addition, machine learning networks can borrow from many other features of the human brain. For example, there can be multiple processing units (neurons) that process signals in different ways (excitability, inhibition, etc.) at the same time, and the connections between the units can be varied, not only feedforward, but also feedback and lateral connections.

"Machine learning relies entirely on supervised learning, unlike the human brain. The human brain's network, which can be constantly pruned and changed during the learning process, has another set of unsupervised learning methods that can effectively find the best path to the most likely success. Therefore, we need to look at the efficient and low-cost network structure from the perspective of unsupervised learning of the human brain. Pu Muming said.

"This also shows the importance of the 'China Brain Project'." Pu Muming said that the "China Brain Project" aims to solve the problem of the basic circuit mentioned above, that is, what kind of network can create such effective functions. "Starting from this kind of effective network, we can design effective artificial network algorithms, or hardware, equipment, chips, etc. This is the prospect of the future. ”

Fudan University is the first in the world to construct a dynamic map of the brain and discovered the variability of the brain network, which means that humans can control the learning ability of the brain. For example, brain-computer fusion technology has been developed to achieve precise regulation of brain function.

However, Zhang Xu said that at present, there are no biosensors, processors, computers, etc. similar to brain intelligence, combined with neural network analysis of brain science, including mutual coordination and control between different brain regions, resource utilization, etc., and these basic theories need to be further developed. From the perspective of mathematics and computational science, both models that are closer to the structural network of the cerebral cortex and models closer to computing are constantly being developed.

"We will never be able to build an identical brain, and the possibility of completely mimicking a bionic brain is very small, so we can only achieve a closer bionic one." Zhang Xu said that brain-like is the most inclusive, and some basic principles of brain science or neuroscience and the basic principles of brain operation are applied to the structure and functional design of intelligent device chips, neural network computing, and intelligent robots.

Zhang Xu's research team is currently committed to breaking through a series of key intelligent technologies based on neural networks, strengthening the research and development of brain imaging technology equipment, neural network-related technologies, and intelligent basic components through brain perception function maps and Chinese brain molecule, structure, and function maps, and building the ultimate intelligent system.

"In terms of key technologies of neural networks, Cambrian and iFLYTEK have participated in the research and development, and have made breakthroughs in R&D results and principles from the cognitive technology of models to intelligence and manufacturing." Zhang Xu said.

Cross-integration is not easy

However, both Pu Muming and Zhang Xu admit that the integration of brain science and artificial intelligence is still very difficult.

"People who work in the field of artificial intelligence or information have completely different backgrounds and languages than people who work in neuroscience and brain science. People in both fields speak their own language, and the other cannot understand what they are saying. In Pu's view, in order to integrate the two, artificial intelligence must understand the progress of brain science, and brain science must also understand what artificial intelligence is doing.

To this end, the mainland has also begun to strengthen team building in the field of "artificial intelligence + brain science". For example, the Center for Excellence in Brain Science and Intelligent Technology of the Chinese Academy of Sciences is an interdisciplinary and cross-institutional organization that solves major problems in the two frontier fields of brain science and brain-like intelligence technology through teamwork and interdisciplinary integration.

Zhang Xu said that brain-like artificial intelligence is a gathering point for collective innovation in a distributed society, and Shanghai is also establishing a brain science and brain-like research center, on the one hand, to meet the needs of national science and technology strategic development, and on the other hand, to build a research and development platform and a platform for talents and knowledge.

"Just like when we set up many forward-looking advanced interdisciplinary research projects, we must break through the limitations of traditional disciplines and specialties in order to open up new research fields and develop new development directions." Zhang Xu said.