University of Massachusetts Amherst Develops Asynchronous AI That Significantly Reduces Computing Energy Consumption and Supports Continuous Learning
As artificial intelligence (AI) systems continue to scale up and improve in performance, their energy consumption has also surged dramatically.
According to foreign media reports, a research team led by Hava Siegelmann, a distinguished professor at the Manning College of Information and Computer Sciences at the University of Massachusetts Amherst, has developed a new type of AI that operates more closely to the core working patterns of the human brain. The study focuses on two complementary goals: enabling AI to engage in real-time continuous learning, rather than learning only during fixed training phases, and significantly reducing the energy consumption of intelligent computing.
This research, recently published in Nature Communications, demonstrates that AI can achieve advanced capabilities while substantially lowering energy consumption.
Hava Siegelmann stated: “Current AI is powerful but extremely energy-intensive. Our research proves that it is entirely possible to build high-performance AI with greatly improved operational efficiency.”
The energy consumption gap between the human brain and existing AI is vast. According to data from the U.S. National Institutes of Health, the human brain operates with approximately 86 billion neurons working in parallel, consuming only about 20 watts of power—equivalent to the energy usage of a small LED light bulb. In contrast, top-tier AI models today can consume tens of millions of watts during training and rely on large data centers to function.
The key reason for the human brain’s exceptional energy efficiency lies in its asynchronous operation. When performing specific tasks or updating information, only a small fraction of neurons are activated, allowing the brain to carry out complex behaviors with extremely low energy consumption.
