A team of researchers from the Chinese Academy of Sciences has created an innovative brain-inspired network model that focuses on internal complexity to overcome the limitations of traditional AI models, such as high computational resource demands. This study is detailed in the journal Nature Computational Science.
One of the main goals in advancing artificial intelligence (AI) is to create more general-purpose models that exhibit broader cognitive abilities. Current popular methods often involve constructing larger and deeper neural networks to achieve this general intelligence, relying on what is termed "external complexity." According to researcher Li Guoqi, this approach, while prevalent, faces challenges like unsustainable energy consumption and a lack of interpretability.
In contrast, the human brain consists of approximately 100 billion neurons and nearly 1,000 trillion synaptic connections, with each neuron featuring a rich internal structure. Remarkably, the brain operates on roughly 20 watts of power.
Motivated by the brain's dynamics, the research team—comprising scientists from the Institute of Automation, Tsinghua University, and Peking University—employed an "internal complexity" strategy to achieve general intelligence. Their experiments demonstrated the effectiveness and reliability of this internal complexity model in managing complex tasks, providing new theoretical insights and practical solutions for integrating neuroscientific principles into AI. This approach may enhance the performance and efficiency of AI models moving forward.