Scientists want to power AI with real human brain cells

Silicon chips are hitting a brick wall. Between power-hungry data centers and the physical limits of squeezing transistors onto silicon, computing is running out of headroom. Now, researchers want to bypass chips entirely by plugging AI directly into real human brain cells.

On Tuesday, a multidisciplinary international team led by researchers at Johns Hopkins University (JHU) published a paper in Frontiers laying out a roadmap for Organoid Intelligence (OI). The idea is to take 3D cultures of human brain cells, called brain organoids, and hook them up to brain-machine interfaces so they can act as living biological hardware.

Moving past silicon

The fundamental pitch here is raw biological efficiency. JHU researcher Thomas Hartung pointed out that we are rapidly reaching the physical limits of silicon computers. The brain, by contrast, operates on a completely different scale, packing roughly 100 billion neurons linked by more than 10^15 connection points, with an estimated storage capacity of 2,500 terabytes.

This is not entirely starting from zero. Some of the scientists involved previously built DishBrain, a system where living neurons wired to a computer interface actually learned how to play the video game Pong.

OI aims to scale that up dramatically. Co-author Lena Smirnova confirmed that they already have functional brain organoids with synchronous electrical activity that react to chemical and electrical stimulation. The current hurdle is deciphering the molecular mechanics of how these clusters learn, paving the way for long-term memory models.

Beyond biocomputing, the team believes OI could transform medicine. Growing disease-specific organoids could unlock new treatments for neurodegenerative disorders and overhaul drug screening without risking human subjects.

Why it matters

Every tech giant is currently burning billions of dollars and megawatts of power trying to make LLMs faster and smarter. Biological tissue operates at a fraction of that energy budget. If organoid computing manages to commercialize even a sliver of that efficiency, it fundamentally resets the physics of compute.

That said, nobody is replacing their Mac with a petri dish anytime soon. Keeping biological cells alive requires sterile perfusion, complex feeding loops, and extreme environmental stability. Silicon may be hot and inefficient, but you can turn it off and put it in your backpack without killing it.

We think positioning this research around medical testing alongside raw biocomputing is the smart move. It gives organoids a pragmatic, near-term commercial purpose while the hardware engineering catches up. But the elephant in the room remains ethics. The moment these cell clusters start demonstrating complex learning behaviors, the philosophical debate over what constitutes synthetic consciousness will quickly stop being a hypothetical question.