Zhang Xu: Interdisciplinary integration is key to brain-like intelligence

Brain science research has become an important focus in the current global technology competition.

The concept of brain-like intelligence was first proposed by American scientist Carver Mead in the late 1980s. It refers to machine intelligence inspired by the neural and cognitive mechanisms of the human brain and achieved through the synergy of software and hardware using computational modeling. Brain-like intelligence serves as the cornerstone for developing artificial general intelligence.

Zhang Xu, an academician at the Chinese Academy of Sciences and director of the Guangdong Institute of Intelligence Science and Technology (GDIIST), delivers a speech at 2023 Greater Bay Area Science Forum.[Photo/gsf.org.cn]

Interdisciplinary integration is the key

Zhang Xu, an academician at the Chinese Academy of Sciences and director of the Guangdong Institute of Intelligence Science and Technology (GDIIST), said that to lead the world in brain-like intelligence, China needs to make breakthroughs in multiple disciplines to develop an ideal brain-like chip. "First, it's the artificial neurons, which are the result of integrating brain-like science principles, neural science principles, computer technology, brain-like intelligence algorithms, and more. Second, connecting these neurons to form a neural network capable of processing information requires joint efforts from brain scientists, computer scientists, mathematicians and other information technology and chip-design personnel," said Zhang.

In Zhang's view, the key to the current development of brain-like chips lies in the integration and exchange of various disciplines to study how to design neural networks.

Currently, many universities in China have established research centers to conduct brain-like intelligence research.

Tsinghua University and the Institute of Automation of the Chinese Academy of Sciences established their brain-like computing research center in 2014 and 2015 respectively. In 2018, the Chinese Institute for Brain Research in Beijing and the Shanghai Center for Brain Science and Brain-Inspired Technology were established.

Interdisciplinary integration requires the support and collaboration of talented individuals from various fields. Zhang emphasized that scientific and technological research is about solving talent problems, and there is no other way.

"Hengqin in Guangdong province has an excellent scientific research and living environment and can take advantage of Guangdong's connection to the Chinese mainland and Macao's internationalization. It also has talent policy support and great potential for the development of brain-like intelligence," Zhang said.

How far is it to commercialize brain-like intelligence?

Brain-like chips mimic the human brain in design and are a type of artificial intelligence chips. Compared to traditional chips, brain-like chips have greater advantages in power consumption and learning capabilities. Currently, AI chips have a relatively wide range of applications in such fields as 5G communication, autonomous driving, smartphones, drones and intelligent security.

"Brain-like intelligence will have broad future applications. In specific fields, brain-like intelligence is more sensitive and faster in tracking sudden events,"said Zhang. At night, if a person, animal, or car suddenly runs across the vision of a moving vehicle, ordinary vehicle equipment may not capture this event in time, but brain-like chips can capture it clearly.

Zhang stated that the development of brain-like intelligence is essentially a market behavior, with research institutes and enterprises playing different roles. "Research institutes mainly focus on basic research, early-stage development of core technologies, and the promotion of industrial technology. This will lead to the emergence of some related companies that focus on product research, application technologies for products and industrial technological research to meet the market demands," Zhang explained.

Many industry insiders believe that the slow progress of commercialization is due to the current models that cannot fully simulate the complex multi-brain region structure of the human brain. For example, current deep learning can only simulate single brain regions such as visual recognition and speech recognition, but it is difficult to achieve the ability for multiple brain regions to learn jointly.

Zhang said that from sci-tech research to technological application, brain-like intelligence is the future trend and a frontier field. In the future, international brain-like intelligence will accelerate the research and development of brain-like intelligence technology and products of its kind, break through brain-machine interface technology and focus on utilizing neural computing to simulate the process of information processing in human brains.