Computing has spent seventy-odd years telling one story: silicon gets smaller, and smaller silicon does more. That’s Moore’s Law in a sentence, and it held up remarkably well — until recently. Transistors are now so tiny that quantum effects start messing with their reliability. There’s only so much shrinking left to do. So chip designers are looking past silicon for whatever comes next, and one of the strangest contenders isn’t a material at all. It’s biology. Researchers call it “wetware”: using living, lab-grown neurons to do computational work. The field itself goes by a more formal name — organoid intelligence biological computing.
So What Is Organoid Intelligence, Really?
At its core, organoid intelligence is the study of growing small three-dimensional clumps of living brain cells — brain organoids — to see whether they can learn, hold onto information, and maybe someday do something computationally useful. These clusters start as stem cells. Scientists push the cells through a specific sequence of chemical cues until they turn into neurons and self-organize into something that resembles, structurally and electrically, a very early slice of developing brain tissue. Researchers at Johns Hopkins University, who coined the term, describe organoid intelligence as a genuine form of biological computing built on scientific and bioengineering advances.
It’s worth being upfront about what these are not. They’re not tiny brains. Nobody has found evidence they’re conscious or have any kind of inner experience. In terms of size, most organoids top out somewhere between a few hundred thousand and a few million neurons. That’s nothing compared to the roughly 86 billion packed into an actual human brain. But even at that scale, they still run the basic machinery every animal brain runs on: neurons wiring together, firing signals, and strengthening or weakening those connections based on activity. That last part — synaptic plasticity — is generally considered the mechanism behind learning itself.
Why Bother With Biological Computing?
The pitch for organoid intelligence biological computing boils down to one word: efficiency. And the gap is almost absurd once you look at the numbers. Training a large AI model today can chew through staggering amounts of electricity. Entire data centers full of GPUs run around the clock and need serious cooling just to function. Meanwhile, your brain does things — recognizing faces in real time, parsing language, reasoning through abstract problems — that still stump the best AI systems around. And it runs on roughly the wattage of a dim lightbulb.
That contrast drives the whole pitch. If even a slice of that biological efficiency could be captured, it might mean computation that costs a fraction of what silicon demands for the same task. Scientists in the field also point to something silicon can’t fake: real plasticity. Artificial neural networks “learn” through statistical adjustment during training. That process mimics the outcome of biological learning without actually being it.
How Do You “Program” a Clump of Neurons?
This is where the field diverges hardest from anything resembling normal software work. Nobody writes code and uploads it to an organoid. Instead, researchers rely on microelectrode arrays: small biocompatible sensors that sit against or near the tissue. These arrays pick up the faint electrical chatter of firing neurons, and when needed, they deliver controlled pulses back in.
Most experiments so far are simple by design. Researchers try to get the organoid to respond a certain way to a certain stimulation pattern, then reinforce that response until it sticks. It’s the same feedback logic behind training an artificial network, just running on living cells instead of matrix math. In one widely discussed example, researchers coaxed a brain organoid into playing a stripped-down version of a classic arcade game. Stimulation and feedback shaped its responses toward the “correct” behavior over time.
Some people in the field call this a biological version of programming, but the comparison only goes so far. There’s no instruction set to write. What’s actually happening looks closer to training an animal than writing software — shaping behavior through repeated exposure and reinforcement, not dictating outcomes line by line.
Where the Hard Problems Are
The distance between today’s lab experiments and any kind of usable biological hardware stays enormous, and most researchers in this space will tell you that plainly.
Keeping the tissue alive. A silicon chip needs power and a fan. Living tissue needs nutrients, waste removal, oxygen, and tight temperature control — essentially a whole life-support setup. Organoids also don’t last especially long right now. Keeping them viable for extended stretches remains an unsolved problem.
Getting enough signal in and out. Today’s electrode arrays can only talk to a small slice of an organoid’s neurons at any given time. Researchers are working with a partial picture of what the tissue actually does. Cramming more electrodes into that space without wrecking the tissue is its own serious bioengineering challenge.
Consistency. Organoids are grown, not manufactured, so no two turn out quite the same — even from identical starting cells under identical conditions. Predictable, repeatable behavior is still a long way off.
The ethics of it all. As these systems learn to do more complicated things, the ethical questions around them get louder too, even with zero evidence of anything resembling consciousness. Most labs working in this area push for building oversight and ethical guidelines alongside the science itself, rather than scrambling to catch up after the fact. The Baltimore Declaration toward Organoid Intelligence, drafted at the first OI workshop, lays out exactly this kind of early framework.
Show Image
What It Could Mean Down the Road
Let’s be honest about where things stand: this isn’t about to replace the chip in your phone, your laptop, or the racks of GPUs training the next big AI model. This is still early-stage basic research, happening at a handful of specialized labs. Most of the work right now just tries to answer whether the underlying idea works at all, before anyone seriously talks about applications. For a broader look at where AI hardware is headed next, see our guide to neuromorphic computing.
Still, the long-term potential deserves attention. If even part of that biological efficiency advantage gets tapped eventually, it might open the door to AI systems that use dramatically less power. That matters given how much scrutiny AI’s energy footprint already draws. Some researchers find a second angle just as compelling: studying how these biological systems learn could reveal something fundamental about neural computation in general. That insight might eventually feed back into designing smarter, more efficient artificial neural networks, regardless of whether wetware itself ever becomes practical.
Right now, organoid intelligence biological computing sits in a strange spot between neuroscience and engineering. Researchers are trying to understand biological intelligence and figure out whether that understanding can become hardware, all at once. Whether wetware ends up as a footnote or an entirely new branch of computing is anyone’s guess. But it’s a solid reminder: whatever comes after silicon might look stranger, and more alive, than most people expect.