Wulf A. Kaal

· Artificial Intelligence, Governance

Who Vouches for a Dish of Neurons?

Living neurons can now be rented through the cloud. Their operators monitor them. Nobody publishes a record of which units to trust.

What is biological compute?

Biological compute uses living neurons, grown on a chip, to process information. In 2022 the DishBrain study connected cultures of human and mouse neurons to a simple game of Pong. When the paddle hit the ball, the neurons received a brief, predictable signal. When it missed, they received seconds of unpredictable stimulation. Within a short session, their play improved modestly (Kagan et al. 2022).

It is now a market. In March 2026 Cortical Labs launched a cloud of 120 units that customers can rent (Sharwood 2026). FinalSpark runs a remote platform on which more than 1,000 brain organoids have been used (Jordan et al. 2024).

What is it missing?

Operators watch their cultures closely. They record activity, health, lifetime and death. What is missing is a record of which units to trust.

A culture can fail in three ways. It can give poor answers. It can drift, because living tissue keeps rewiring. It can die. A renter needs to know which units have held up, and who answers when one fails.

The dish cannot answer for its mistakes. The law does not recognize artificial intelligence as a legal subject (Fenwick, Kaal, and Vermeulen 2017). For accountability, a culture is in the same position. If someone is to answer, the natural candidate is the lab that operates it.

What can reputation add?

Reputation adds three things that monitoring does not.

  • A judge other than the seller. A monitor is the operator's report on its own unit. A reputation record is written by validators who judge the unit's work: experts stake reputation for or against it, and the winning side splits the losing side's stake (Calcaterra, Kaal, and Andrei 2018).
  • A cost for being wrong. Validators risk their own stake. Without that, a reputation system has nothing at stake, and experts have no investment to protect by policing it (Calcaterra and Kaal 2021).
  • A reason to perform tomorrow. Reputation turns a one-off deal into a repeated one by offering the promise of future business (Calcaterra and Kaal 2021). A lab that expects later rentals to depend on a unit's standing has a reason to keep that unit reliable.

Two rules protect the record. First, a unit's reputation cannot be sold. If it could, money could buy a good name without the skill, and the signal would stop meaning anything (Kaal 2026b). Second, a failing unit cannot simply restart under a new name. Where a decentralized record depends on a lasting identity, it breaks down when a poor performer can start fresh at no cost (Kaal 2026a). The same risk applies to living units. A replacement culture gets a new identity and a verified record of where it came from.

Drift and death go on the record too. A unit answers known test questions at regular intervals. If its answers shift, it is suspended until cleared. When it dies, its standing is retired.

Why does it matter?

Customers can now rent this compute. Without a trust record, a renter has only the operator's own report, and a bad unit can pass as a good one. Repeated runs and comparison groups will still be needed. A trust record would help renters judge the units used in those tests.

Regulation will be slow to fill the gap. Rules written for a new technology can be obsolete before they are final (Kaal 2016). The people building this market can create the record now.

What are the prospects of getting this right?

A basic record can be built now. A lasting identity, a stake, a challenge, and a history someone else can check are familiar parts of a reputation record. Using them on living neurons is the hard part. A fully trustworthy record cannot yet be built, for four reasons.

  • Proving the readings are real. Until the hardware can prove what the neurons did, the record measures the lab's claims, not the culture's work. This is the largest gap.
  • Agreement is not truth. Validators can agree and still be wrong. Where correctness matters, the standard needs ground truth or an independent appeal.
  • A small market. The market is still new. It may be too small, for now, to attract a deep pool of expert validators.
  • Privacy. Human-derived cells raise consent and privacy duties. The record must follow the consent terms of each cell source and keep donor-identifying details off any permanent public record.

One idea is a long shot. In DishBrain, play improved modestly while hits brought predictable signals and misses brought unpredictable ones. Could a unit's settled reputation, fed back to it as a signal, improve its behaviour? Nobody knows. The time delays are large, and a fair test would need careful controls. It is a research question, not a promise.

What are the applications?

  • Choosing and pricing rented units. A record of which culture was used, how old it was, whether it drifted, and whether it died mid-job.
  • Trustworthy research. A published result would name the unit that produced it and its standing at the time.
  • Drug and toxicity testing. Laboratories already test drugs on networks of cultured human neurons (Parodi et al. 2024). Standing can sit beside their existing quality checks as an early warning. It does not replace them.
  • An audit trail for regulators. Where human cells came from, which cultures replaced which, and when each was retired.
  • A test of reputation itself. A culture is unlikely to game the record, although the people around it can. That makes it a cleaner test of whether reputation tracks real reliability.

Biological compute can now be rented. What it lacks is a public record of which units to trust. Monitoring tells you what happened, in the operator's own words. Reputation tells you whom others trusted, and what they risked on it. That record is what this market still needs.

References

Calcaterra, Craig, Wulf A. Kaal, and Vlad Andrei. 2018. "Blockchain Infrastructure for Measuring Domain Specific Reputation in Autonomous Decentralized and Anonymous Systems." SSRN. https://ssrn.com/abstract=3125822

Calcaterra, Craig, and Wulf A. Kaal. 2021. "The Importance of Reputation for the Evolution of Decentralization." SSRN. https://ssrn.com/abstract=3782210

Fenwick, Mark, Wulf A. Kaal, and Erik P. M. Vermeulen. 2017. "Regulation Tomorrow: What Happens When Technology Is Faster Than the Law." SSRN. https://ssrn.com/abstract=2834531

Kaal, Wulf A. 2016. "Dynamic Regulation for Innovation." SSRN. https://ssrn.com/abstract=2831040

Kaal, Wulf A. 2026a. "Evolution of Domain-Specific Reputation Systems: From Binary Validation to Citation-Weighted Knowledge Attribution." SSRN. https://ssrn.com/abstract=6192998

Kaal, Wulf A. 2026b. "AI's Mother's Instinct: Engineered Consequence, Emergent Ethics, and the Institutional Trajectory Toward Agentic Alignment." SSRN. https://ssrn.com/abstract=6244278

Jordan, Fred D., Martin Kutter, Jean-Marc Comby, Flora Brozzi, and Ewelina Kurtys. 2024. "Open and Remotely Accessible Neuroplatform for Research in Wetware Computing." Frontiers in Artificial Intelligence 7. https://doi.org/10.3389/frai.2024.1376042

Kagan, Brett J., et al. 2022. "In Vitro Neurons Learn and Exhibit Sentience When Embodied in a Simulated Game-World." Neuron 110 (23). https://doi.org/10.1016/j.neuron.2022.09.001

Parodi, Giulia, et al. 2024. "In Vitro Electrophysiological Drug Testing on Neuronal Networks Derived from Human Induced Pluripotent Stem Cells." Stem Cell Research & Therapy. https://doi.org/10.1186/s13287-024-04018-2

Sharwood, Simon. 2026. "Inside Datacenter Where Day Starts with Cerebrospinal Fluid." The Register, March 14, 2026. https://www.theregister.com/2026/03/14/cortical_labs_biological_cloud/