Imagine if your smartphone could think as efficiently as your brain, using barely any power. Sounds like science fiction, right? But scientists have just taken a giant leap toward making this a reality by creating a pocket-sized AI brain inspired by monkey neurons.
Here’s the mind-blowing part: while a human brain uses less energy than a light bulb, AI systems often guzzle electricity to perform even simple tasks. Now, a team of researchers has developed a highly efficient AI model that mimics part of the brain’s visual system, offering a glimpse into how nature accomplishes so much with so little. Their findings were published in the journal Nature (https://www.nature.com/articles/s41586-026-10150-1).
The model initially relied on 60 million variables, but the team managed to shrink it to just 10,000 variables while maintaining nearly the same performance. And this is the part most people miss: this compact version is so small it could be sent in a tweet or email. “That is incredibly small,” says Ben Cowley (https://www.cshl.edu/research/faculty-staff/benjamin-cowley/), an assistant professor at Cold Spring Harbor Laboratory and study author. “This could revolutionize how we think about AI efficiency.”
But here’s where it gets controversial: this tiny model doesn’t just save space—it also behaves more like a living brain. This could help scientists unravel mysteries behind diseases like Alzheimer’s, Cowley explains. Is this the key to unlocking the secrets of the human brain, or are we oversimplifying its complexity?
Mitya Chklovskii (https://www.simonsfoundation.org/people/dmitri-mitya-chklovskii/), a group leader at the Simons Foundation’s Flatiron Institute, who wasn’t involved in the study, points out that if this AI truly replicates natural strategies, it could deepen our understanding of the brain’s inner workings. He also suggests it could lead to “more powerful and humanlike artificial intelligence.” But here’s the catch: current AI models are based on outdated 20th-century knowledge of the brain. Should we be updating the foundations of AI networks to reflect modern neuroscience?
The study focused on the human visual system, which transforms light into recognizable images, like your grandmother’s face or the Grand Canyon. Scientists have long puzzled over questions like, “How do we recognize a cat or a dog?” Since directly observing a human brain in action isn’t feasible, Cowley turned to AI systems that perform similar tasks. However, these systems are often black boxes—we don’t fully understand how they work, much like our own brains.
Working with researchers from Carnegie Mellon and Princeton, Cowley created an AI model that simulates V4 neurons, a specific part of the visual system. These neurons encode colors, textures, curves, and complex proto-objects. While existing AI uses deep neural networks that require massive computing power, Cowley’s team aimed for efficiency. “We wanted to take these big, clunky models and compress them into something smaller and more manageable,” he says.
They started with data from macaque monkeys, identified redundant parts of the model, and applied statistical techniques similar to those used for compressing digital photos. The result? A model so small it fits in an email attachment. But here’s the real kicker: because the model is so simple, the team could observe its artificial neurons in action. Some V4 neurons responded to shapes with strong edges and curves—think arranged fruit in a grocery store. Others reacted only to small dots, which might explain why primates are drawn to eyes. Could this specialized behavior be the key to how our brains process visual information so efficiently?
The implications for AI are huge. If our brains use less complex models yet outperform AI systems, it suggests AI could be smaller, simpler, and more effective. For instance, self-driving cars might run on less powerful computers while better distinguishing between a pedestrian and a plastic bag. But Chklovskii cautions that AI still falls short in tasks humans find easy, like recognizing a friend’s face from any angle or with a new haircut. Is AI truly ready to match human intelligence, or are we still missing something fundamental?
What do you think? Are we on the brink of creating AI that rivals the human brain, or are we oversimplifying the problem? Let’s discuss in the comments!
