A team of Hong Kong researchers is delving into the human brain to unlock the secrets of advanced artificial intelligence (AI), aiming to develop AI systems capable of lifelong learning and task performance. The scientists seek to overcome one of the limitations of conventional computers by creating more efficient and powerful hardware inspired by the brain’s functionality. Led by Li Can, an assistant professor at the University of Hong Kong, the team is exploring the potential of the memristor, an emerging memory device that mimics the brain’s storage and processing abilities. The innovative microelectronic platform has the capacity to replicate the behavior of biological synapses and neurons, enabling computers to emulate the human brain more effectively.
Li explains that unlike computers, human brains can function even with dead cells, tolerate defects, and learn from experience. The team aims to harness these unique traits while also leveraging the computational strengths of computers for scientific calculations and pattern recognition. By studying the memristor, which operates by computing directly within memory, Li and his team hope to eliminate the need for data transfers between memory and processing units, making AI systems more energy-efficient.
The potential applications of this technology are vast. Li envisions wearable sensors for disease monitoring that are highly energy-efficient without compromising functionality, enabling devices to run for years on a single charge. Memristor chips could also expedite virus genome sequencing, reducing the time needed from days or weeks to a much shorter duration. The technology could be used in subdermal health-monitoring implants, smartphones, and watches, effectively offloading AI tasks from energy-intensive data centers.
Li’s innovative work has been recognized with a HK$5 million grant from the Croucher Foundation, which supports local scientists. He plans to use the funding to hire new team members, support experiments, and explore cutting-edge research.
As AI models continue to grow in complexity with the inclusion of trillions of parameters, efficient memory systems are becoming increasingly important. The memristor could revolutionize the field by providing energy-efficient hardware to meet the escalating computational demands of advanced AI systems, presenting a promising future for AI development beyond conventional computing.
