Scientists in China have made a significant stride towards achieving artificial general intelligence (AGI) with the creation of a novel computing architecture inspired by the human brain. This breakthrough, published in the journal Nature Computational Science, offers a potential solution to the limitations of current AI models, particularly large language models (LLMs) like ChatGPT and Claude 3.
While LLMs have demonstrated impressive capabilities, they remain constrained by their training data and struggle with human-level reasoning. The pursuit of AGI, a hypothetical system capable of performing any intellectual task a human can, has primarily focused on scaling up neural networks. However, this approach faces challenges, including escalating energy consumption and computational resource demands.
The new architecture, inspired by the intricate structure and efficiency of the human brain, takes a different approach. Instead of focusing on external complexity by scaling up neural networks, it emphasizes internal complexity within individual artificial neurons. This concept draws inspiration from the human brain’s 100 billion neurons, each boasting a rich internal structure, yet consuming a mere 20 watts of power.
The researchers developed a Hodgkin-Huxley (HH) network, where each artificial neuron is a complex HH model capable of scaling in internal complexity. HH models are known for their accuracy in simulating neural activity and representing the firing patterns of real neurons. This makes them ideal for constructing deep neural network architectures that aim to replicate human cognitive processes.
The study demonstrates that this architecture can handle complex tasks efficiently and reliably. Notably, a small model based on this architecture performs as well as a much larger conventional model, showcasing its potential for greater efficiency and resource optimization. The development of this brain-inspired architecture could pave the way for more efficient and powerful AI systems, potentially bringing us closer to realizing the long-sought goal of AGI.
