800,000 brain cells once learned to play Pong while seated inside a thumb-sized dish filled with nutrient solution in a Melbourne, Australia laboratory. Not in a symbolic sense. In actual Pong, the 1972 arcade game in which a white ball bounces between paddles on a black rectangle of a screen, the neurons receive electrical signals indicating the location of the ball, produce their own electrical pulses in response, and gradually improve at not missing over the course of roughly five minutes.
This is not in some loose, abstracted sense of the word “play.” On a silicon chip that was linked to a computer, the cells were grown using a combination of human stem cells and mouse embryos. They were blind. Not a nervous system. Technically, no brain at all. However, there was a lesson in that dish.
| Topic | DishBrain — Lab-Grown Brain Cells Playing Pong |
|---|---|
| Lead Organization | Cortical Labs, Melbourne, Australia |
| Collaborating Institutions | University College London (UCL), Monash University |
| Published In | Journal: Neuron (October 2022) |
| Lead Researcher | Dr. Brett Kagan, Chief Scientific Officer, Cortical Labs |
| System Name | DishBrain |
| Cell Count | ~800,000 neurons (human stem cells + mouse embryo cells) |
| Interface | High-density multi-electrode array (silicon chip) |
| Learning Time | 5 minutes to begin playing Pong |
| Feedback Mechanism | Predictable stimulation = reward (hit); random stimulation = punishment (miss) |
| Key Finding | Biological neurons learn faster and use less energy than silicon-based AI |
| Potential Applications | Alzheimer’s research, drug testing, neurodegenerative disease modeling, bio-hybrid computing |
| Latest Development | Cortical Labs now building systems where cells play Doom (as of March 2026) |
| Reference Website | Cortical Labs |
The group responsible for this accomplishment, led by Dr. Brett Kagan at Cortical Labs and working with researchers from Monash University and University College London, published their results in the journal Neuron in October 2022. As the company’s chief scientific officer, Kagan made a claim that fell somewhere between science and philosophy: he referred to DishBrain, the system’s name, as the first “sentient” lab-grown brain. That term was criticized by other experts in the field, who said it went too far. However, very few challenged the fundamental outcome. From the outside, the neurons’ responses to the task and feedback appeared remarkably similar to learning.
The simplicity of the mechanism was elegant. The neurons received signals from a multi-electrode array embedded in the dish’s base that indicated the ball’s location on the screen and its distance from the paddle. The neural equivalent of a reward was given to the cells when they generated electrical activity that correctly moved the paddle and made contact with the ball. Following what Kagan refers to as a basic drive toward predictability, the cells were motivated to avoid the chaotic and random feedback that resulted from their failure. Programming was absent. The system is not instructed by an algorithm. The cells figured it out on their own terms, imperfectly but quantifiably.
Not only did the cells perform, but the way they performed in comparison to silicon alternatives caught the attention of scientists. It is possible to train traditional AI systems to play Pong flawlessly, and they have done so numerous times. However, that training necessitates a substantial amount of energy, computational power, and prolonged reinforced repetition.
The DishBrain neurons learned in five minutes while using a small portion of the energy required to complete the same task on a traditional computer. Additionally, they illustrated biological adaptability, a point that Brett Kagan repeatedly brought up in subsequent interviews. The neurons increased their electrical activity to adjust to the new unpredictability when the game was interrupted, such as when the ball was reset to a random position halfway through. Once they had discovered a pattern, they settled back down.
This kind of mid-task environmental shift is a challenge for silicon systems, even powerful ones. A person can walk into a kitchen they have never been to before and figure out how to make tea, according to Kagan’s straightforward description. No matter how powerful, a computer cannot accomplish that without specific instructions for each unknown variable.
It’s difficult to avoid thinking about that comparison for a little while. The speed at which silicon computing has developed would have seemed unthinkable to those who created the first computers, and the difference between what machines could accomplish in 1972—the year Pong was released—and what they can now accomplish is nearly unfathomable. However, it is still very challenging to replicate artificially the fundamental cognitive flexibility that enables a human to navigate an unfamiliar kitchen or learn a new task from limited feedback in an uncertain environment. The same quality was being displayed in miniature by the neurons in that Melbourne dish. That is not a minor issue.
The study falls under the expanding field of organoid intelligence, which investigates the possibility of using clusters of lab-grown brain tissue, or organoids, as biological processors. This is a worthwhile endeavor just based on the energy efficiency argument. The amount of electricity used by modern data centers is becoming a real barrier to the expansion of the AI sector.
One of the most urgent physical constraints in the field would be resolved by a biological computing substrate that functions similarly while consuming orders of magnitude less power. Dishes like DishBrain and the iterative process of comprehending precisely how neurons acquire and apply learned behavior may be the route to that future. It’s also possible that it will take decades to overcome the enormous technical obstacles separating a functional biological processor from a Pong-playing petri dish. Kagan has taken care to avoid making exaggerated claims about deadlines.
It’s possible that the medical applications will arrive sooner. Disease research was Kagan’s initial stated objective for DishBrain, not computing. Alzheimer’s and other neurodegenerative diseases are notoriously challenging to study because animal models only partially reflect what occurs in the human brain and so much of the pertinent biology is inaccessible in living patients.
Research opportunities that were previously unattainable are made possible by a system of living human neurons that can be seen reacting to stimulation in real time and that can be exposed to experimental drug compounds to gauge their effects. Since then, Cortical Labs has expanded into that area, and as of early 2026, researchers associated with the project are reportedly working with neurons performing much more difficult tasks—Doom, the first-person shooter from 1993, has been mentioned. The difference between Doom and Pong is not insignificant. It’s a fair indicator of how rapidly this field is evolving.
It is genuinely unclear whether any of this has an impact on consciousness or whether something is going on inside that dish that warrants a more precise term than “learning.” The neurons are obviously unaware that they are playing Pong, according to the researchers. As of right now, scientists are unable to identify or characterize any subjective experience. However, neither biology nor philosophy have provided a clear answer to the question of where the boundary is between complex information processing and something more, between a system reacting to stimulation and a system experiencing it. DishBrain also didn’t respond. But it did make the question much more pressing.