Brain specialists and experts in artificial intelligence should work together to create AI as complex as the human brain, a top scientist said at a Beijing seminar on Monday.

Poo Mu-ming, director of the Chinese Academy of Sciences' Institute of Neuroscience, said AI should be able to learn quickly, be versatile and be energy efficient.

"The human brain is the most complex object in the universe," he said. "Both the brain and AI involve big data training, but the brain develops its neural network organically with significant structural changes over decades, whereas AI's learning structure is static and requires a huge amount of data and energy."

The brain's ability to rewire and adapt to changing environments is called plasticity, and this may hold the "key to overcoming the current limits of machine learning", he said.

Most AI research focuses on imitating and maximizing a small part of the brain's function, such as sensory recognition. However, higher cognitive functions like language and emotions are still too complex for computers, said Guo Aike, a biophysicist with the academy.

"Machines excel in activities that have clearly defined rules and straightforward goals, such as video or board games, but they struggle to perform tasks in environments that involve many changing variables," Guo said when talking about a match scheduled this month between world Go champion Ke Jie and AlphaGo, an AI program developed by Google.

He said computer scientists can learn from neuroscientists about how the brain stores and processes information, to develop AI that can learn dynamically, transferring skills learned from one task to new ones, while keeping its energy requirement low.

There should be more platforms and opportunities for computer scientists and brain specialists to work together and share their findings, Guo said.

One application of a brain-inspired AI technology is self-driving cars, which rely heavily on sensors and a massive quantity of live and preprogrammed data, according to Li Deyi, an academician at the Chinese Academy of Engineering.

No matter how meticulously programmed, "self-driving cars still encounter problems when faced with challenging road conditions", Li said. One solution, he added, is to materialize a human driver's cognition by mimicking human brain activities, and to build this "mechanical driving brain" into self-driving cars.