Human Neuron Augmented Machine?

A computer can turn the activity of living neurons into movement on a screen. It can also send electrical signals back to those cells as the screen changes. In one experiment, this loop connected a dish of neurons to a simplified version of Pong. In a later commercial demonstration, a different system linked neurons to Doom.

Those examples are easy to mistake for something more familiar: a human brain placed inside a computer, watching and playing a game. The actual systems are much less complete and more revealing. They combine living cells with electrodes, software and conventional computers. Researchers are investigating whether the cells’ ability to change their connections can be useful for computing and for studying the nervous system.

The questions extend beyond performance. What should donors know about the uses of their cells? How should researchers oversee increasingly complex neural cultures? And what evidence would be needed before a laboratory system raised a serious question about subjective experience?

Those questions animate “The Inevitable Horrors of Biocomputing,” a video by Pursuit of Wonder. The video discusses DishBrain, the commercial CL1 system and other projects, while exploring scenarios involving consciousness, pain and brains in vats. Its philosophical concerns are an invitation to examine the research, not evidence that today’s cultures suffer. The video received grant support from the Center for AI Safety.

Cells, organoids and brains

The distinction between a neuron culture and a brain matters because the word neuron describes a building block, not an entire organ.

DishBrain used cells grown as a relatively flat, two-dimensional culture over an array of electrodes. Neurons can form networks in such a dish and produce electrical activity that electrodes detect. The cultures studied in the 2022 DishBrain paper consisted separately of human neurons derived from stem cells or embryonic mouse cortical cells; they were not a single mixed human–mouse culture.

A brain organoid is different. It is a three-dimensional structure grown from cells that can develop some features of neural tissue. Its organization can make it useful for studying aspects of development and disease that are difficult to reproduce in a flat layer. An organoid is nevertheless not a miniature whole brain: it lacks the fully developed, complex circuitry and brain-wide connections of that organ. A 2021 National Academies analysis described those missing features as relevant to assessing consciousness, alongside questions about consent and oversight.

Neither dish cultures nor organoids should be confused with an implanted brain-computer interface. An implanted interface connects electronics to the nervous system of a living person; the cells remain part of that person’s brain and body. In the systems discussed here, researchers grow neural tissue outside a body and connect it to laboratory equipment.

That separation does not make ethical scrutiny unnecessary. It does define the object of scrutiny: a living experimental culture connected to a machine, rather than a person using an implant.

How the loop works

A neuron-linked game begins with an ordinary computer model of the game. Software converts selected information about its changing state into electrical stimulation delivered through electrodes beneath or around the neural culture. The electrodes also record patterns of activity from the cells.

A second translation turns some of that recorded activity into a control signal. The conventional computer uses the signal to move a virtual paddle or select another game action. The game state then changes, and the system sends new stimulation to the culture. Input, neural activity, decoded action and feedback form a closed loop.

The translations are essential. Electrodes do not give cells eyes, and the software does not establish that a culture sees a ball or understands a game. Researchers choose what information to encode, how to stimulate the cells, which signals to read and how those signals affect the simulated environment.

In the DishBrain study, the authors reported apparent improvement in a simplified Pong-like task within about five minutes compared with controls. Missed trials produced an unpredictable pattern of electrical stimulation; successful play involved more predictable stimulation. The authors interpreted the results as evidence of learning in an embodied neural system, though the interpretation is debated. Cortical Labs authors were among the researchers.

The paper’s title used sentience in a narrow sense related to responsiveness to sensory input. That usage does not establish conscious feeling. The electrical feedback does not show that the cells experienced discomfort, had a purpose or perceived the game.

From Pong to Doom

Cortical Labs later presented a separate Doom demonstration using its CL1 platform. The company describes CL1 as a research system combining neurons grown on a silicon chip with a bidirectional electrical interface, software and life support that supplies nutrient-rich medium. It also markets remote research access to the platform. Cortical Labs’ description is evidence of what the company offers and claims, not an independent validation of every performance claim.

The Doom demonstration is not the peer-reviewed Pong experiment repeated with a more elaborate game. A conventional computer runs Doom and the interface; the neural culture contributes activity to a translated feedback loop. The project’s source code describes how game information is encoded for electrodes and neural activity is mapped to control. That is useful evidence about the design, but it is not, by itself, a peer-reviewed benchmark. The public demonstration does not establish that the neurons understood or consciously experienced Doom, or that they achieved independently validated human-level performance.

The distinction matters because a visually recognizable game can make a sparse input-and-output arrangement look more capable than the experiment establishes. The display shows the game to human observers. It does not show what, if anything, a culture experiences.

Other ways to build with tissue

DishBrain and CL1 are not the only approaches. FinalSpark’s Neuroplatform, described in a 2024 research paper, provides remote research access to three-dimensional neural organoids coupled to electrodes and maintenance equipment. Its authors reported working with more than 1,000 organoids over three years, with some lasting more than 100 days. Those are historical platform figures, not a count of 1,000 simultaneous brains.

A separate project, Brainoware, used a human brain organoid alongside conventional electronics and software in an approach called reservoir computing. In simplified terms, the organoid supplied a complex response to inputs, and the surrounding system learned to use that response for constrained tasks. The reported tests included speaker recognition and prediction of a nonlinear equation. Recognizing speakers under experimental conditions is not the same as understanding language, and neither task demonstrates person-like thought.

Even within neuron-chip research, capabilities vary. A 2026 Scientific Reports paper presented methods for routing signals and reading activity on a stem-cell-derived neural platform, while explicitly stating that it did not demonstrate task training, learning or long-term plasticity. A device that can exchange signals with neurons has cleared one engineering hurdle; it has not automatically demonstrated every capability associated with a broader field.

Promise and practical costs

Neural cultures could help researchers study how networks develop and respond to drugs, or investigate features of neurological disease in human-derived tissue. The value would depend on how well a particular model represents the biological process being studied and whether findings translate beyond the dish. Such systems may complement other research models rather than stand in for an entire patient or brain.

Computing is a more speculative application. Living networks can adapt, making them interesting candidates for tasks involving changing patterns. But showing a response or improvement in a controlled experiment is far from establishing a general-purpose computer that is reliable, economical and useful at scale.

The engineering problems are substantial. Cells need stable conditions and nutrients. Cultures differ from one another and change over time. Electrodes offer a limited way to stimulate and observe a complex network, while software must translate between biological signals and digital tasks. Researchers also need meaningful controls and repeatable benchmarks to determine what the living component contributes.

Energy claims require the same care. A comparison between biological neurons and digital hardware is not enough to establish the consumption of a complete system. A fair assessment would include the conventional computer, interface electronics and life-support equipment, measured on comparable tasks and performance targets. Cortical Labs makes claims about energy and performance, but those claims should not be treated as independent, matched-system measurements.

The ethical boundary is still being studied

The available evidence does not establish that current neural cultures or organoids are conscious, experience pain or suffer. That conclusion leaves room for disagreement about how research should prepare for future systems.

Neuroscientist Kenneth Kosik argues that current organoids lack features needed to support operational consciousness, while recognizing that future developments could raise new questions. His 2024 perspective is an expert assessment of present limitations, not a definitive test of every culture or a guarantee about future technology.

Wood and coauthors take a more precautionary approach, considering the possibility that future organoids could develop morally relevant capacities and asking how research ethics should respond. Their argument concerns uncertainty and future risk; it is not evidence that current cultures feel pain. These positions differ in emphasis, but neither turns electrical activity alone into proof of subjective experience.

Oversight also has immediate subjects. Human-derived cells originate with donors, and consent needs to address how cells may be used in research, including uses donors might not anticipate when material is collected. The National Academies’ analysis places donor consent and research oversight alongside questions about increasingly complex neural models. As experimental systems change, review can examine the source and proposed use of cells, the design of the research and whether new evidence changes its ethical significance.

For now, the striking result is that living neural networks can be coupled to digital systems in ways researchers can measure and study. What those systems can ultimately compute, and what moral questions future versions may present, remain open questions whose answers depend on evidence more specific than a game appearing on a screen.

DutyStation News prepared this AI-assisted article using the linked public sources.