香港研究人員在原始模型——人腦中尋找先進人工智慧晶片的關鍵

A team of scientists in Hong Kong is embarking on an ambitious project to advance artificial intelligence (AI) beyond its current capabilities. They are drawing inspiration from the most complex and efficient model available - the human brain. The human brain excels at lifelong learning and task performance, and the team aims to develop AI that can emulate these abilities. The challenge lies in creating more efficient and powerful hardware than what is currently available.

Leading the team is Li Can, an assistant professor at the University of Hong Kong's Department of Electrical and Electronic Engineering. He and his team are working on creating AI circuits and systems that mimic the functioning of the human brain. Li points out that unlike computers, the human brain can function despite the presence of dead cells. He also highlights that humans can learn from experience, a trait that current computers and AI lack. For instance, a chatbot may not be aware of the latest news due to its knowledge being cut-off at a certain point. Furthermore, humans can reason based on vague information, whereas computers require clear instructions.

However, computers do have their advantages. They are more adept at performing scientific calculations, repetitive operations, and deriving values from patterns than the human brain. To help computers better emulate the human brain, Li and his team are exploring an emerging memory device known as a memristor. This advanced microelectronic platform can replicate the behaviour of biological synapses and neurons, the pathways of the human brain. These brain-inspired devices could potentially revolutionize technology and significantly speed up graphics processing units used to train AI systems. Looking ahead, Li believes that these memristor chips could offload AI tasks from energy-intensive data centres to devices like smartphones and watches with high energy efficiency. They could also be used in subdermal health-monitoring implants and expedite virus genome sequencing.