Garrett Kenyon’s A.I. Research Featured in Scientific American
Garrett Kenyon’s A.I. Research Featured in Scientific American
Garrett Kenyon, a physicist and neuroscientist at Los Alamos National Laboratory and the New Mexico Consortium, has his artificial intelligence (AI) research featured in Scientific American in an article titled Lack of Sleep Could Be a Problem for AIs. The article highlights groundbreaking research suggesting that some forms of artificial intelligence may benefit from a process remarkably similar to sleep.
While machines such as refrigerators, calculators, and conventional computers can operate continuously without rest, Kenyon’s research indicates that certain biologically inspired artificial intelligence systems may require periods of sleep-like activity to function optimally. In this respect, these advanced neural networks appear to share an important characteristic with the human brain.
The discovery emerged from research focused on developing neural networks that learn to recognize and classify objects in a manner similar to how humans learn. Kenyon and his team were training AI systems to “see” and interpret visual information through repeated exposure to images, much like a young child gradually learns to identify objects through experience.
However, the researchers observed an unexpected problem. After extended periods of continuous learning, the neural networks began to exhibit signs of instability. Their performance degraded, and the systems became less effective at processing information accurately.
To address this issue, the team experimented with introducing a sleep-like state into the neural networks. Rather than shutting the systems down, they exposed the networks to patterns of activity resembling the electrical waves experienced by the human brain during sleep. Remarkably, the networks regained stability and continued to function effectively after these periods of artificial “rest.”
Further investigation revealed that the most effective way to induce this sleep-like state was to expose the neural networks to a broad range of random, static-like signals. These signals mimic the neural activity associated with slow-wave sleep, also known as deep sleep, which plays a critical role in maintaining healthy brain function in humans.
Importantly, the need for sleep was not observed in all forms of artificial intelligence. The phenomenon appears to be specific to neural networks that use biologically realistic processors designed to emulate the behavior of neurons in the brain. These systems more closely mirror the structure and function of biological neural networks than traditional AI architectures.
The findings suggest that sleep serves a fundamental purpose not only in biological brains but also in artificial systems modeled after them. Rather than representing a period of inactivity, sleep may be an active process that helps stabilize learning and maintain proper function.
As Kenyon explains, “In neural networks as well as living creatures, a sleep-like state is not inactivity, but a different kind of activity that is crucial to the proper functioning of neurons.”
This research offers new insights into both neuroscience and artificial intelligence, highlighting unexpected connections between biological and machine learning systems. As AI technologies continue to become more sophisticated and brain-like in their design, understanding the role of sleep-like processes may prove essential for building more stable, efficient, and capable intelligent systems.
Read the entire article at: https://www.scientificamerican.com/article/lack-of-sleep-could-be-a-problem-for-ais/
Article by Carrie Talus.