Human brain consumes significantly less power compared to a light bulb, while AI systems require massive electricity to perform similar operations. Researchers have now developed a highly efficient Pocket AI model that demonstrates how biological brains manage complex tasks with minimal energy. The findings were published in the journal Nature.
The model is an imitation of the visual system of the brain. It was initially written to take 60M variables but was reduced to only 10 000 variables, with similar performance. It is vastly small, said Ben Cowley an assistant professor at Cold Spring Harbor Laboratory. He further said that the model is small enough to fit in a tweet or email.
Tiny AI Model Mimics the Brain’s Efficiency
The miniature Pocket AI system seems to be more of a biological brain. This resemblance may be useful in investigating the neurons related disorders like the Alzheimer disease. Researchers believe that biology-based AI models can uncover the way in which the human brain handles information in such an efficient manner. Such discoveries will transform neuroscience studies.
The Simons Foundation Flatiron Institute Mitya Chklovskii stated that these kinds of models could result in more human-like systems of artificial intelligence. He did not participate in the study. The study focuses on finding out the ways in which the visual system converts light into the visualized signs, which can be identified as landscapes or recognizable faces.
Monkey Data Powers Visual AI Breakthrough
Scientists cannot directly observe this process in the human brain. So the team trained AI systems using data from macaque monkeys.
Cowley and his colleagues worked with researchers from Carnegie Mellon University and Princeton University. They built a simplified AI model that mimics V4 neurons. These neurons help detect colors, textures, and complex object shapes.
The classical deep neural networks require large computing power. In comparison, the compact model removes superfluous components and statistical compression. The outcome is an AI model so small that it can be delivered in an email attachment, but it is effective at major visual recognition tasks.
Smaller Models May Shape the Future of AI
Due to simplicity of the model, researchers are able to analyze the mechanism of work of its artificial neurons closely. Other neurons were very sensitive to forms which curve such as the fruits in grocery stores. Small dots were mostly responded to by other neurons, which scientists associated with eye-oriented primates. This specialization could be the reason that brains take advantage of visuals.
These results show that developers can make Pocket AI systems smaller, more efficient, and more powerful without sacrificing performance. The potential uses are self-driving vehicles that require less computing capabilities.
specialists warn that AI still struggles with complex recognition tasks. For example, it often fails to recognize a friend in changing or dynamic conditions. Scientists believe updating AI foundations with the latest neuroscience insights can close that gap.
Read :Â Telemedicine Visits Dramatically Lower Patient Costs
