Pocket-Sized AI Brain Uses Monkey Neurons

The research, a collaboration between scientists at Cold Spring Harbor Laboratory, Carnegie Mellon University, and Princeton University, leveraged data from macaque monkeys to drastically shrink an AI vision model. The original model contained 60 million variables, but the team successfully compressed it to a version with just 10,000 variables that performed nearly as well. This compressed model is so small it could be sent in a tweet or an email. The study specifically focused on V4 neurons, which are part of the brain's visual system and are responsible for encoding colors, textures, and complex shapes. By analyzing how the miniaturized AI processed images, the researchers found that certain artificial neurons showed a strong preference for dots. This is significant because dots are a key component of one of the most vital visual cues for primates: eyes. This new model offers more than just efficiency; it provides a clearer window into how the brain itself functions. Traditional, large-scale AI models are often referred to as "black boxes" because their internal workings are too complex to fully understand. By creating a simpler, more interpretable model, scientists can better study the underlying mechanisms of vision and potentially apply this knowledge to understanding brain disorders like Alzheimer's disease. This breakthrough is part of a growing field known as "organoid intelligence" or biocomputing, which combines living brain cells with artificial systems. Companies like Cortical Labs have already demonstrated the ability to grow real neurons on silicon chips, creating systems that can learn tasks like playing the video game "Pong". The goal is to develop computers that are far more energy-efficient and can learn with significantly less data than current AI.