Definition
DishBrain: DishBrain is an experimental system described by Kagan et al. in Neuron in 2022, in which cultured mouse and human neurons on a high-density microelectrode array controlled the paddle in a simulated Pong-like game. Closed-loop stimulation told the cells where the ball was, and the authors reported improved rally lengths within minutes of play.
Also known as: Kagan et al. 2022, neurons playing Pong, in vitro Pong experiment
DishBrain is an experimental system described by Kagan et al. in Neuron (2022) in which cultured mouse and human neurons, grown on a high-density microelectrode array, controlled the paddle in a simulated Pong-like game. Electrical stimulation encoded the ball’s position, the cells’ firing moved the paddle, and the authors reported longer rallies within minutes of closed-loop play.
Key takeaways
- DishBrain connected flat (2D) neuronal cultures to a game through a high-density microelectrode array, closing the loop between stimulation and activity.
- Ball position was encoded with place and rate coding on eight sensory electrodes; spikes in two “motor” regions moved the paddle.
- Instead of reward, the cells got predictable stimulation after a hit and unpredictable stimulation after a miss, a design based on the free energy principle.
- The authors reported improved rally length within five minutes of gameplay, which was not seen in control conditions.
- The paper’s use of the word “sentience” drew a published critique (Balci et al., 2023) and a reply from the authors.
What was DishBrain designed to test?
DishBrain asked a simple question: if you give living neurons a body, a world and consequences, will their behavior change in a goal-directed way? Earlier culture experiments had shown that networks of neurons on electrode arrays respond to patterned input, but most did not let the culture’s output change what it received next. DishBrain made that loop the center of the experiment, a design known as closed-loop stimulation.
The work came from Brett Kagan and colleagues at Cortical Labs and collaborating institutions, with Karl Friston among the authors. It was framed as a test of active inference, the idea that a system acts on the world to make its own sensory input more predictable.
How did DishBrain work?
The system had four parts, each of which is a general building block of organoid intelligence experiments.
The cells and the chip
The team tested primary cortical cells from embryonic mice and human cortical neurons derived from induced pluripotent stem cells by two different methods. The cells grew as a flat layer on a commercial CMOS high-density array that reads up to 1,024 of its 26,400 electrodes at 20,000 samples per second, according to the paper. Low-latency software the authors called DishServer linked the chip to the game simulation.
Encoding: telling the cells where the ball is
Eight electrodes formed a sensory area. The ball’s position was encoded two ways at once:
- Place coding: which sensory electrode was stimulated signaled the ball’s vertical position.
- Rate coding: stimulation frequency rose from 4 Hz to 40 Hz as the ball approached the paddle.
This is a concrete example of the encoding half of neural encoding and decoding.
Decoding: letting the cells move the paddle
Other regions of the array were defined as motor regions. Activity in one region moved the paddle up and activity in the other moved it down. Because neurons settle randomly across the chip, this mapping was arbitrary. To play well, the network had to organize its action potentials to match a layout imposed from outside.
Feedback: predictable versus unpredictable
The most distinctive design choice was the feedback. There was no chemical reward and no “pain” signal.
- After a hit, the cells received a short, predictable burst of stimulation across all sensory electrodes, and play continued.
- After a miss, they received unpredictable stimulation at random sites and times for about four seconds, then the ball restarted in a random direction.
Under the free energy principle, a self-organizing system should act to avoid states that lead to unpredictable input. If missing produces disorder and hitting avoids it, a culture that behaves as the theory predicts should hit more often.
What results did DishBrain report?
Kagan and colleagues reported “apparent learning within five minutes of real-time gameplay not observed in control conditions.” In their main comparison, mouse and human cultures showed significantly longer average rallies in the later part of a 20-minute session than in the first five minutes, and outperformed controls by then. Human cultures started worse than controls and then improved. The authors also reported fewer “aces” (missing the first serve) over time and stronger functional connectivity between sensory and motor regions during play than at rest.
Which controls did they use?
The controls are what make the result interpretable.
| Condition | What changed | Purpose |
|---|---|---|
| Media only | No cells on the array | Rule out electronic artifacts |
| Rest | Cells control paddle but get no sensory input | Separate feedback from spontaneous activity |
| In silico | Paddle driven by random noise | Baseline for chance performance |
| Silent feedback | Misses followed by no stimulation instead of random stimulation | Test whether unpredictability matters |
| No feedback | Misses produce no restart or feedback | Test whether closing the loop matters |
In the feedback experiment, only the standard predictable and unpredictable condition showed a significant increase over matched rest controls. The authors concluded that structured, closed-loop feedback was important for the improvement. A step-by-step walkthrough is in our article on how neurons learn in a dish.
Was it learning?
The paper reports a change in behavior driven by feedback, which the authors called learning. Whether that change reflects lasting synaptic plasticity, short-term network adaptation, or both is harder to establish from a single session. Showing that behavior is consistent with the free energy principle also does not prove the principle is the only explanation. Simpler accounts, such as networks reorganizing to reduce strong, disruptive input, could predict similar outcomes. The silent-feedback condition addresses this in part, but the theoretical debate continues.
Why was the word “sentience” controversial?
The paper’s title stated that the neurons “exhibit sentience.” In the text, the authors define sentience in a formal sense: being responsive to sensory impressions through adaptive internal processes. They did not claim consciousness, feelings or experience.
Many readers hear “sentience” as implying subjective experience. In 2023, Neuron published “A response to claims of emergent intelligence and sentience in a dish” by Fuat Balci and a large group of co-authors, who challenged the use of terms such as sentience and intelligence for these results. Kagan and colleagues replied in the same journal with “Scientific communication and the semantics of sentience.” The exchange is now a standard reference point for how biocomputing results should be described. Our glossary entry on sentience and our article on whether brain organoids can become conscious cover the wider question.
Why does DishBrain matter for organoid intelligence?
DishBrain became one of the landmark results cited in the case for organoid intelligence. Smirnova et al. (2023, Frontiers in Science) discuss it in their roadmap for the field. Its importance is less about Pong and more about the template it offers:
- A closed loop in which a culture’s output changes its next input.
- Explicit encoding and decoding rules that can be written down and reproduced.
- Feedback designed from a theory, so the experiment tests a hypothesis rather than just showing activity.
- Controls that separate biology from artifacts and from chance.
The same structure applies to three-dimensional brain organoids, which have richer internal structure than flat cultures but are harder to stimulate and record evenly. It also contrasts with Brainoware, which used an organoid as a reservoir with a trained digital readout rather than giving the tissue direct control of an action.
What are DishBrain’s limitations?
- Simple task. A one-dimensional paddle game is far from general-purpose computation.
- Short timescale. Effects were measured within a session, not over weeks.
- Variability. Cultures differ, and results are averages across many sessions.
- Interpretation. The link to the free energy principle is one explanation among several.
- Language. Terms like “sentience” can outrun what the data show.
Making protocols like DishBrain’s encoding, decoding and feedback rules versioned and rerunnable across cultures is the kind of work we are building toward at Organoid Programming.
Conclusion
DishBrain showed that cultured neurons, embodied in a simple game through a high-density electrode array, changed their behavior under structured closed-loop feedback within minutes. Its careful controls made it influential, and its choice of words made it controversial. For anyone designing experiments in biocomputing, DishBrain remains a clear worked example of how to close the loop between living neurons and a task.
Frequently asked questions
What is DishBrain in simple terms?
DishBrain is a system in which living neurons grown on an electrode chip played a simplified version of Pong. Electrical pulses told the cells where the ball was, and the cells’ own firing moved the paddle. After a hit the cells received predictable stimulation, and after a miss they received random stimulation. Kagan and colleagues reported in Neuron in 2022 that rallies grew longer within minutes.
Who created DishBrain?
DishBrain was described by Brett Kagan and colleagues at Cortical Labs, an Australian biocomputing company, working with collaborating academic institutions. The theoretical neuroscientist Karl Friston was among the authors. The paper, titled In vitro neurons learn and exhibit sentience when embodied in a simulated game-world, was published in the journal Neuron in 2022.
Were the DishBrain neurons a brain organoid?
No. DishBrain used two-dimensional cultures, meaning neurons grown as a flat layer directly on the electrode array. The cells were either primary cortical cells from embryonic mice or human cortical neurons derived from induced pluripotent stem cells. Brain organoids are three-dimensional tissues with more structure. The closed-loop principles DishBrain tested can, however, be applied to organoids.
Did DishBrain show that neurons are sentient?
The authors used sentience in a narrow, formal sense: responding to sensory input through adaptive internal processes. They did not claim the cultures were conscious or had feelings. Many researchers objected to the word, and Balci and colleagues published a critical response in Neuron in 2023. Kagan and colleagues replied in the same journal, discussing how the term should be used.
Why did DishBrain use unpredictable stimulation instead of a reward?
The design was based on the free energy principle, which proposes that self-organizing systems act to keep their sensory input predictable. Hitting the ball produced predictable stimulation and missing produced random stimulation. If the culture behaves as the theory predicts, it should come to favor actions that lead to predictable input, which means hitting the ball more often.
What are the limitations of DishBrain?
The task was very simple, the performance gains were measured over a 20-minute session, and cultures vary a lot from one to another. Behavior consistent with the free energy principle does not rule out other explanations, such as networks adapting to avoid strong disruptive stimulation. Long-term memory, transfer to other tasks and scaling to harder problems were not demonstrated.
Sources
- Kagan et al. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron.
- Balci et al. (2023). A response to claims of emergent intelligence and sentience in a dish. Neuron.
- Kagan et al. (2023). Scientific communication and the semantics of sentience. Neuron.
- Smirnova et al. (2023). Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish. Frontiers in Science.
