The Intelligence Isn't in the Dish

*Cortical Labs built a remarkable closed-loop training system. Then they branded it as a brain.*

In 2022, a company called Cortical Labs made a splash with a claim that traveled everywhere from a YouTube explainer to a wave of coverage about synthetic biological intelligence: brain cells in a dish learned to play Pong. The setup, for anyone who missed the splash: neurons extracted from cortex, plated onto a multi-electrode array, wired so that electrical stimulation carried sensory feedback and the cells' own firing moved the paddle. Over hours, the cultures got better at the game than chance would predict. The framing is vivid — neurons learning a game — and it's worth understanding what actually happened before arguing about what it means.

What's missing from that picture is any published evidence that the culture itself is anything more than a standard dissociated cortical prep on a multi-electrode array. Look at the published work and the background on Cortical Labs: cell-type composition that reads like any other dissociated cortical culture, no described emergent structural organization beyond "neurons on an MEA," no morphological characterization that distinguishes these cultures from the ones sitting in hundreds of electrophysiology labs right now. The honest description of the biological component is: dissociated neurons on a chip. That's the substrate.

The intelligence lives in the closed loop, not the cells.

Here's where the actual innovation sits: the system that maps sensory input to electrical stimulation feedback, creates a training environment, and lets any sufficiently plastic culture adapt to it. The neurons do real work — plasticity, signal processing, adaptation — but it's fungible work. Replace the culture with another dissociated cortical prep and you'd expect comparable results, because the thing that makes DishBrain work isn't the specific neurons. It's the loop. The analysis that frames this as a training architecture rather than a biological one is closer to the published evidence than the venture pitch is.

And this isn't one lab's spin — it's a field-wide reassurance pattern. The same move shows up in the human-mouse chimera work, where the standard line is that "no human-like abilities were observed." That's a claim about the ruler, not the animal: nobody has an assay that could detect the thing being reassured about. The dish and the chimera are two instances of the same rhetorical habit — locating the interesting claim in the biology, where the evidence is thinnest, rather than in the system that actually does the work.

The problem is that "brain in a dish" carries enormous rhetorical and ethical weight that "closed-loop MEA training system with dissociated cortical culture" does not. One sounds like it belongs in a bioethics panel about emergent sentience. The other sounds like a control systems paper. The gap between the public framing and the published morphology is exactly where funding hype and premature ethical hand-wringing both take root — the same ambiguity feeds the venture pitch and the discourse about whether a dish of neurons deserves moral consideration. Neither conversation is well-served by pretending the culture is something the published evidence doesn't show it to be.

The engineering is genuinely impressive. Building a closed-loop system that lets biological neurons learn a task through sensory feedback is a real technical achievement and deserves serious recognition. But calling it "synthetic biological intelligence" locates the innovation in the wrong layer of the stack — in the biology, where nothing novel has been demonstrated, rather than in the control architecture, where everything novel actually lives. The overselling makes the real conversation harder to have: not whether a dish of neurons is sentient, but what closed-loop biological control systems actually are, what they can become, and what that means for the engineering of intelligence. That conversation deserves better than branding.