A small cluster of human brain cells, grown in a petri dish and wired to a computer, has learned to play the 1993 first-person shooter video game Doom. The achievement, reported by researchers at startup Cortical Labs, represents a striking demonstration of biological intelligence operating outside the human body — and raises profound questions about the nature of consciousness, the future of computing, and the ethical boundaries of neuroscience.
The experiment builds on earlier work by the same Melbourne, Australia-based company, which in 2022 made headlines when it showed that a dish of roughly 800,000 brain cells — dubbed “DishBrain” — could learn to play Pong, the simple paddle-and-ball arcade game. Now, as reported by Futurism, the team has dramatically scaled up the complexity, training biological neurons to engage with the far more demanding environment of Doom, a game that requires spatial awareness, movement through corridors, and the ability to engage enemies in a three-dimensional space.
From Pong to Doom: A Leap in Biological Computing
The progression from Pong to Doom may sound like a novelty, but it represents a significant jump in computational demand. Pong is a two-dimensional game with a single moving object and one axis of player control. Doom, by contrast, involves navigating a 3D environment, identifying threats, and making split-second decisions about movement and combat. The fact that a collection of neurons in a dish can process visual input from the game and produce meaningful output — actions that correspond to gameplay — is a testament to the remarkable adaptability of biological neural tissue.
Cortical Labs achieved this by growing human neurons on a multi-electrode array, a device that both stimulates the cells with electrical signals and reads their responses. The neurons receive information about the game state — essentially, what is happening on screen — through patterns of electrical stimulation. Their responses are then translated back into game commands. Over time, the neurons appear to learn, improving their performance in ways that suggest genuine adaptation rather than random firing.
How the Neurons Actually “Learn”
The mechanism by which these neurons improve is rooted in a principle the Cortical Labs team calls the “free energy principle,” a theoretical framework associated with neuroscientist Karl Friston of University College London. Under this model, biological systems — including brains — are fundamentally driven to minimize surprise, or unpredictability, in their environment. When the DishBrain system makes a poor move in the game, the electrodes deliver random, unpredictable stimulation. When it performs well, the stimulation is structured and predictable. The neurons, in effect, learn to play the game in order to make their world more predictable.
This is not the same as conscious decision-making. The neurons are not “thinking” about Doom in any meaningful sense. But they are responding to environmental feedback and adjusting their behavior accordingly — a fundamental hallmark of intelligence, however primitive. Brett Kagan, chief scientific officer at Cortical Labs, has described the system as demonstrating “sentient” behavior, though he uses the term in its narrow, technical sense: the capacity to sense and respond to stimuli, not the broader philosophical meaning that implies subjective experience.
The Ethical Tightrope of Lab-Grown Brains
The experiment inevitably raises ethical questions that the scientific community is only beginning to grapple with. If a cluster of human neurons can learn, adapt, and respond to its environment, at what point does it acquire moral status? The current system is far too simple to warrant concerns about suffering or consciousness — it has roughly the complexity of a small insect’s nervous system, if that. But the trajectory of the research points toward increasingly sophisticated biological computing systems, and ethicists warn that the field needs guardrails before it gets there.
The National Institutes of Health and several international bodies have begun examining the ethics of brain organoid research, which involves growing three-dimensional clusters of brain cells that can develop surprisingly complex structures. Some organoids have been shown to produce electrical activity resembling brain waves. The Cortical Labs work adds another dimension to this debate: not just growing brain tissue, but putting it to work, giving it tasks, and measuring its performance. As Futurism noted, the research forces a reckoning with what it means to use human-derived biological material as a computing substrate.
Biological Computing: A Genuine Alternative to Silicon?
Beyond the philosophical questions, there is a practical dimension to this work that has attracted serious attention from the technology and defense sectors. Biological neurons are extraordinarily energy-efficient compared to silicon-based processors. The human brain operates on roughly 20 watts of power — about the same as a dim light bulb — while performing computations that the most advanced supercomputers struggle to replicate. If biological neural networks could be harnessed for computing tasks, the energy savings alone would be transformative.
Cortical Labs has received funding from various sources, and the potential applications range from drug discovery — where biological neural networks could model the effects of pharmaceuticals more accurately than digital simulations — to autonomous systems and robotics. The Australian government’s defense research agency has also expressed interest, seeing potential in systems that can learn and adapt in ways that conventional AI cannot. Unlike artificial neural networks, which require vast amounts of data and energy to train, biological neurons appear to learn from remarkably small amounts of experience.
How This Compares to Artificial Intelligence
The comparison to modern AI is instructive. Large language models like OpenAI’s GPT-4 and Google’s Gemini are trained on trillions of tokens of text data, consuming enormous quantities of electricity in the process. They produce impressive results but remain fundamentally statistical engines, predicting the next word in a sequence without any understanding of meaning. The DishBrain system, by contrast, operates on entirely different principles — biological ones that evolved over hundreds of millions of years to process information efficiently.
That said, the DishBrain’s current capabilities are extremely limited. Its Doom gameplay is rudimentary at best. The neurons can move through corridors and engage with basic elements of the game, but they are not competing with human players or even competent AI bots. The significance lies not in the quality of the gameplay but in the proof of concept: biological tissue, disconnected from a body and a brain, can still learn to interact with a complex digital environment. The gap between this and practical biological computing remains vast, but the direction of travel is clear.
What Comes After Doom
Cortical Labs has signaled that it intends to continue scaling up the complexity of tasks it assigns to its biological systems. The company is also working on improving the longevity of its neural cultures — currently, the cells survive for several months, a limitation that would need to be addressed for any practical application. Other research groups around the world are pursuing parallel lines of investigation, including teams at Johns Hopkins University, where scientists have been developing “organoid intelligence” as a formal research program.
The Johns Hopkins initiative, led by Thomas Hartung, envisions a future in which biological computers made from brain organoids could perform certain types of computation more efficiently than any silicon chip. Hartung has been quoted describing the potential as a new form of computing that could complement, rather than replace, traditional digital systems. The vision is ambitious, and the timeline is long — likely decades before anything resembling a practical biological computer could be built. But the foundational work, including the Cortical Labs experiments, is laying the groundwork.
A Question Science Cannot Yet Answer
Perhaps the most unsettling aspect of the DishBrain research is what it reveals about how little we understand about intelligence itself. If a few hundred thousand neurons in a dish can learn to play a video game, what does that say about the nature of learning, memory, and cognition? The human brain contains roughly 86 billion neurons connected by trillions of synapses. The DishBrain system is to the human brain what a single transistor is to a modern microprocessor — and yet it can learn.
This suggests that the capacity for learning may be a more fundamental property of neural tissue than previously appreciated, emerging at scales far below what we typically associate with intelligence. It also raises the possibility that consciousness, or something like it, could emerge in biological computing systems as they grow more complex — a prospect that scientists, ethicists, and policymakers will need to confront with far more urgency than they have shown so far. For now, a dish of brain cells is playing Doom. The question is what happens when the game gets harder.
WebProNews is an iEntry Publication