Biocomputing examples: what has actually been demonstrated
Living neural cultures have learned to play Pong, classified spoken sounds, and run as a remotely bookable cloud service. Each is a real, published demonstration, and each is narrower than the headlines around it.
The fastest way to misunderstand this field is to read its press releases. The fastest way to understand it is to ask, for any claim, what was measured and by whom. Below is a dated survey of demonstrations that cleared peer review or run as inspectable platforms, with the caveats stated next to the result rather than buried beneath it.
DishBrain: a culture that learned Pong
The most cited demonstration is DishBrain, reported by a Cortical Labs led team in 2022. Cultures of roughly 800,000 human and mouse neurons grown on a microelectrode array were embedded in a simplified Pong environment. Electrode positions encoded the ball; the culture's firing moved the paddle. Predictable stimulation followed a hit, unstructured stimulation followed a miss. Performance improved within a single session.1
It shows goal-directed adaptation from a local learning rule and a feedback signal, with no model and no training corpus. It does not show general intelligence, language, or anything resembling a brain. The paddle was one dimensional and the task was simple. The significance is the mechanism, not the score.
Brainoware: organoid reservoir computing
In 2023 a team at Indiana University reported Brainoware, a system that used a human brain organoid as the reservoir in a reservoir-computing architecture and applied it to spoken vowel classification and a nonlinear prediction task.3 Reservoir computing is a natural fit for living tissue: the messy, high-dimensional, nonlinear dynamics that make a culture hard to control are exactly what a reservoir needs, and only a simple linear readout is trained.
The reported accuracy was well below a conventional classifier, which is the honest and expected result. The contribution is the architecture, a way to extract useful computation from tissue you cannot program directly.
FinalSpark Neuroplatform: biocomputing as a cloud service
The Swiss group FinalSpark operates the Neuroplatform, a remotely accessible system of living brain organoids on microelectrode arrays that external researchers can stimulate and record over the internet.4 It is less a single result than an attempt to turn wetware into shared infrastructure, with the life support, perfusion, and electrophysiology handled centrally.
Treat any specific organoid count, electrode count, or uptime figure as something to confirm against the platform's current documentation before relying on it; these change as the system evolves. The durable point is the model: access without owning an incubator.
The demonstrations side by side
| System | Year | Substrate | Task shown | Access |
|---|---|---|---|---|
| DishBrain | 2022 | Neuron culture on MEA | Simulated Pong | Lab |
| Brainoware | 2023 | Brain organoid reservoir | Vowel classification | Lab |
| Neuroplatform | 2024 | Organoids on MEAs | Open research access | Remote |
| Cortical Labs CL1 | 2026 | Neurons on MEA | Commercial unit | Purchase |
The pattern across all four: real, narrow, and early. The honest framing is that biocomputing has demonstrations, not products that outperform silicon at anything. The per-platform specifications, with their caveats, are tracked on the platform comparison hub.
Frequently asked questions
What is the best known biocomputing example?
DishBrain, a 2022 demonstration in which a culture of about 800,000 neurons learned to play a simulated game of Pong through closed-loop electrical feedback.
Can you access a biocomputer remotely today?
Yes. FinalSpark's Neuroplatform lets external researchers stimulate and record from living brain organoids over the internet, with life support handled centrally.
Do these systems beat conventional computers at their tasks?
No. Every demonstration to date is narrow and below conventional baselines on accuracy. The significance is the mechanism and the substrate, not competitive performance.
What is reservoir computing and why does it suit organoids?
Reservoir computing reads useful output from a fixed, complex dynamical system by training only a simple linear readout. An organoid's uncontrollable nonlinear dynamics make a good reservoir.
References
- Kagan BJ, et al. In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron. 2022;110(23):3952-3969. doi:10.1016/j.neuron.2022.09.001. Accessed 2026-06-12.
- Cai H, et al. Brain organoid reservoir computing for artificial intelligence. Nature Electronics. 2023;6:1032-1039. doi:10.1038/s41928-023-01069-w. Accessed 2026-06-12.
- Jordan FD, et al. Open and remotely accessible Neuroplatform for research in wetware computing. Frontiers in Artificial Intelligence. 2024;7:1376042. doi:10.3389/frai.2024.1376042. Accessed 2026-06-12.