Cortical Labs, an Australian biotech company, has pushed the boundaries of biological computing by connecting living human brain cells to video game environments, building on peer-reviewed research that showed neurons in a dish could learn to play a simplified version of Pong. The work has now reportedly extended to more complex game worlds, including the classic first-person shooter Doom, though the foundational science behind these experiments remains rooted in a 2022 study published in the journal Neuron. The progression from a simple paddle game to a maze-filled shooter raises pointed questions about how far biological intelligence can stretch when wired into digital feedback loops.
At the same time, the narrative around these experiments has moved quickly from technical details to sweeping claims about “living computers” and even “sentient” cultures in a dish. That shift makes it essential to separate what has been documented in peer-reviewed literature from what is still at the level of demonstration, prototype, or speculation. Understanding the actual methods used to interface neurons with game environments, and the limits of what those neurons can be said to “know” or “experience,” is key to evaluating both the promise and the risks of this emerging field.
How Neurons in a Dish Learned to Play Pong
The story begins with a system Cortical Labs calls DishBrain, in which human and rodent neurons grown in vitro were placed on multielectrode arrays and connected to a simplified Pong-like game environment. The neurons received electrical signals representing the position of the ball, and their own firing patterns controlled the in-game paddle. Through structured feedback, the cells adapted their behavior over time, effectively learning to return the ball more consistently. This closed-loop setup, where sensory input and motor output flow between living tissue and software, was detailed in a PubMed-indexed study published in Neuron that first brought the DishBrain concept to a wider scientific audience.
The key mechanism was not traditional programming or reward signals of the kind used in machine learning. Instead, the researchers applied structured electrical stimulation that gave the neurons predictable input when they performed the task correctly and random, noisy input when they failed. Over successive sessions, the neurons reorganized their activity to reduce unpredictability in their environment. The associated journal article framed this as evidence that biological neural networks can exhibit goal-directed learning even outside a living organism, a finding that generated significant attention across the scientific community and sparked debate about how to interpret such adaptive behavior in disembodied tissue.
From Pong to Doom: What Changed
After demonstrating that neurons could handle the relatively simple dynamics of Pong, Cortical Labs began exploring whether the same biological systems could cope with far greater complexity. Doom, a game with three-dimensional navigation, enemy encounters, and spatial reasoning demands, represents a dramatic step up. The shift matters because Pong involves a single axis of movement and a predictable ball trajectory, while Doom requires processing visual information across a full environment and responding to threats that appear unpredictably. If neurons can adapt to that kind of input, it suggests that even small clusters of brain cells possess a surprising capacity for flexible, real-time decision-making when embedded in a well-designed feedback loop.
However, a critical distinction separates the Pong work from the Doom claims. The Pong experiment was published in a peer-reviewed journal and subjected to independent scrutiny, including coverage by outlets such as Nature’s news reporting. The Doom extension, by contrast, has not yet appeared in a peer-reviewed primary source based on available reporting. Without published methodology, raw data, or independent replication, the Doom results should be understood as a reported demonstration rather than a validated scientific finding. The underlying biological principles, including multielectrode array interfacing and closed-loop stimulation, remain well-established from the earlier work, but the specific performance metrics and learning dynamics in a Doom environment have not been formally documented in the scientific literature.
Why Structured Feedback Matters More Than the Game
The real scientific contribution of the DishBrain project is not that neurons “played” a video game in any conscious sense. It is that researchers demonstrated a method for giving biological neural networks structured feedback in real time and observing adaptive behavior emerge without traditional training algorithms. As Nature’s coverage emphasized when discussing the original Pong result, the experiment showed neurons adapting to structured tasks, but responsible framing requires distinguishing between adaptive electrical activity and anything resembling awareness or intent. In other words, the language of “playing” or “learning” is a useful shorthand for describing changes in firing patterns, not evidence that the cells possess subjective experiences.
The multielectrode arrays used in these experiments serve a dual purpose: they both stimulate the neurons with input signals and record the neurons’ electrical output. This bidirectional communication channel is what makes the closed-loop system possible. The neurons are not simply reacting to stimuli; they are embedded in a feedback cycle where their own activity shapes what happens next. That architecture, rather than the specific game being played, is the core innovation. Whether the task is bouncing a virtual ball or moving through a virtual corridor, the underlying question is the same: can biological tissue outside a body learn to minimize unpredictability in its environment? The full-text version of the Neuron paper, available through PubMed Central, provides detailed methodology on how the stimulation and recording protocols were designed to test exactly this question.
Biological Computing and Its Practical Limits
One of the most discussed implications of the DishBrain work is whether biological neurons could eventually serve as an alternative to silicon-based processors for certain tasks. Living neurons consume far less energy than conventional chips when performing pattern recognition and adaptive learning. A single human brain operates on roughly the same power as a dim light bulb, while training a large artificial intelligence model can consume electricity equivalent to hundreds of households over weeks. That efficiency gap has led some researchers to speculate that hybrid biological-digital systems could handle specific computational problems, particularly those involving real-time adaptation to unpredictable inputs, more efficiently than purely electronic hardware.
But practical barriers remain steep. Biological neurons are fragile, difficult to maintain outside the body for extended periods, and far harder to scale than transistors on a chip. The DishBrain cultures required carefully controlled laboratory conditions, and the neurons’ performance degraded over time as the networks changed and cells died. There is also no clear path from a dish of cells playing a simplified game to a system that could replace or meaningfully augment existing computing infrastructure. The excitement around the Doom demonstration risks outpacing the actual science, which remains at a very early experimental stage. Researchers have shown that neurons can adapt to structured digital environments, but the distance between that finding and a functional biological computer is vast, involving challenges in stability, reproducibility, manufacturing, and integration with existing hardware.
Ethical Questions That Follow the Science
The DishBrain team’s own published paper used the word “sentience” in its title, a choice that drew both attention and criticism. That language raised concerns that the experiments were being marketed using philosophically loaded terms that go far beyond what the data support. Access to Nature’s online platform, via an authentication gateway, shows how closely discussions of the research have been tied to debates about consciousness, animal welfare, and the moral status of neural tissue maintained in vitro. Critics argue that invoking “sentience” risks confusing the public about what has actually been demonstrated, potentially undermining trust in both neuroscience and emerging neurotechnology.
Beyond rhetoric, there are substantive ethical questions about how far such experiments should go. If future cultures become more complex, with richer inputs and more elaborate feedback, at what point, if any, might they acquire morally relevant properties? Current evidence does not indicate that DishBrain systems possess anything like human or even animal consciousness, but the work forces ethicists and regulators to think ahead. Issues such as consent for the use of human-derived cells, standards for humane treatment of advanced neural cultures, and limits on embedding biological networks in exploitative or harmful virtual environments will need to be addressed as the field advances. For now, the most responsible stance is to treat DishBrain as a powerful tool for probing basic questions about learning and adaptation in neural tissue, while maintaining clear conceptual boundaries between adaptive behavior in a dish and the rich inner lives associated with sentient beings.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
