A human brain consumes less power than a light bulb. The AI systems trying to match it guzzle electricity by the megawatt.
Now scientists have built an AI model so compact it hints at how nature solved this efficiency problem hundreds of millions of years ago — and the answer is almost absurdly small.
Researchers at Cold Spring Harbor Laboratory, working with teams from Carnegie Mellon and Princeton, started with a standard AI vision model containing 60 million variables. Using data from macaque monkey neurons and statistical compression techniques similar to those used for digital photos, they shrunk it to just 10,000 variables.
That's a 6,000-fold reduction. The compressed model could literally fit in an email attachment.
"That is incredibly small," said Ben Cowley, an assistant professor at Cold Spring Harbor Laboratory and lead author of the study, published in Nature. "This is something we could send in a tweet or an email."
The remarkable part? The tiny model performed nearly as well as the original.
The team focused on V4 neurons — cells in the visual system that process colours, textures, curves, and 'proto-objects.' By studying what real monkey neurons responded to and stripping away everything redundant in the AI model, they revealed something fascinating about how biological brains might achieve their efficiency.
Some artificial neurons in the compressed model responded strongly to shapes with curves and edges — 'When you go into the supermarket and you see the arranged fruit, your V4 neurons love that,' Cowley explained. Others responded specifically to small dots, which the researchers suspect relates to primates' innate attention to eyes.
The implications ripple outward. If brains run on models this compact yet outperform AI systems running on supercomputers, something fundamental is different about how they're designed. Self-driving cars might run on far less powerful computers. AI assistants might work offline on phones. And scientists might finally crack open the black box of how we see.
"If our brains have less complex models and yet can do more than these AI systems, that tells us something about our AI systems," Cowley said.
Smaller, simpler, closer to nature. That's the direction. 🧠
