If someone were to tell us what there is live human neurons playing Doom Thinking of a chip, we'd probably think of a science fiction movie. However, that's exactly what the Australian company Cortical Labs has shown: neural cultures connected to a computer system capable of navigating, with varying degrees of clumsiness, one of the most iconic video games of the nineties.
Far from being a classic artificial intelligence, we are talking about real biological tissue interacting with software through a specific interface. A small laboratory “brain”, made up of hundreds of thousands of neurons, receives information from the game environment as electrical patterns and responds with its own neuronal activity, which translates into movements, turns and shots within Doom.
From Pong to Doom: the evolution of a biological computer
The path to this experiment began several years earlier, when Cortical Labs presented a prototype biocomputer capable of playing Pong. That system, based on more than 800.000 cultured human neurons Using a microelectrode array, it took him about 18 months of work to learn how to move the game bar and keep the ball on screen.
In that first project, the neurons demonstrated a real-time adaptive learning It was geared toward a very specific goal: tracking the ball's trajectory. The environment was simple and predictable; if the ball went up, the bar had to go up, and if it went down, the bar had to go down. Even so, it was a milestone because it showed that cells could adjust their behavior to digital stimuli.
The prototype eventually crystallized into a commercial device: the CL1It was presented as the world's first programmable biological computer. This device combines live human neurons grown on silicon with proprietary software, known as bioS, which manages the exchange of electrical signals between the biological tissue and the digital system.
As soon as it became public that the system was capable of playing Pong, the reaction from the tech community was immediate and quite predictable. The same question was repeated on social media and forums: "Okay, but can he play Doom?"Cortical Labs decided to take the challenge seriously and take the test to the next level.
Doom, a chaotic environment for a miniature brain
Doom is hardly a simple video game. This first-person shooter classic, with its three-dimensional mazes, enemies, shooting, and constant choices, presents a chaotic environment difficult to predictThe difference with Pong is enormous: here there is not a single ball to follow, but a world full of stimuli that change at full speed.
According to Brett Kagan, chief scientist at Cortical Labs, the leap from Pong to Doom represents a shift from a nearly linear scenario to one where chaos reigns. While in Pong the relationship between stimulus and response is direct, in Doom neurons have to manage depth, movement, threats and rewards In parallel, something that greatly complicates the learning task.
To meet this challenge, the company did not resort to a single isolated chip, but to a networked platform composed of several CL1 unitsEach module integrates more than 200.000 live human neurons, interconnected and mounted on an electrode array that acts as an intermediary between biology and electronics.
The neurons, however, don't "see" the screen the way a human player would. Instead of images, the system translates the game's state into electrical stimulation patterns which are applied directly to the cell culture. The neurons' responses, also in the form of electrical impulses, are interpreted as actions within Doom: move forward, turn, shoot, move to one side or the other.
Kagan jokes that, for now, his biological “players” behave like a beginner who’s never touched a computer. Their movements are erratic, their aim leaves much to be desired, and they die repeatedly. But even in this chaos, the researchers observe something key: Each defeat provides information, and the cells adjust their behavior accordingly..
A closed loop between neurons and video games
The essence of the experiment lies in the feedback loop generated between the game and the brain's neural activity. An independent developer managed to design an interface that converts Doom's visual information into these electrical patterns, and which in turn translates the neural firing into concrete commands within the game.
In practice, the system works as a closed circuit: The virtual environment sends stimuli to the biological tissueThe neurons respond by altering their activity, and the result of that response (survival, advancement, or elimination) is immediately reflected in the game environment. This dynamic reinforces certain patterns and weakens others, a mechanism very similar to how a living organism learns.
Scientists insist that there are no explicit rules here, unlike in many artificial intelligence algorithms. The biological neural network reorganizes itself, taking advantage of its own plasticity. This is referred to as pure adaptive learning, without predefined complex artificial architectures or millions of training iterations.
One aspect that has caught the attention of the research community is the speed at which changes are being observed. Compared to AI models on silicon that require enormous amounts of data and energy—already research on the cognitive impact of ChatGPT—, these crops showed noticeable improvements in a matter of daysuntil achieving recognizable behavior within the game in less than a week.
Despite these advances, Cortical Labs is urging caution. Its CTO and other executives emphasize that the goal is not to create a "miniature brain" to compete with humans or the leading AI systems on the market, but rather to use neurons as a calculation material with unique properties that silicon cannot replicate.
What does CL1 contribute to modern biocomputing?
CL1 is presented as more than just a one-off experiment. Cortical Labs has positioned it as a programmable biological computing platform, which some even describe as a kind of “wetware-as-a-service”: wet hardware accessible through software, designed so that third parties can carry out their own projects on living neurons.
In practice, the system offers an open API and development tools, so researchers and developers can submit tasks, collect data, and explore new applications. The bioS software manages this exchange of information, translating high-level instructions into precise electrical stimuli and reading the neural responses to return them in a format understandable by traditional programs.
One of the arguments the company uses to defend this approach is energy efficiency. Biological systems, as is often pointed out in neuroscience, are capable of performing complex operations while consuming a fraction of the energy which requires a conventional supercomputer. In a context where the energy cost of AI is increasingly debated, this line of research is gaining appeal.
Furthermore, biocomputing opens up a field where computer science, biology, and medicine converge. The fact that it can observe how real neurons solve problems In controlled environments, it could provide valuable clues both for designing new AI models and for better understanding neurological disorders or testing treatments under very specific conditions.
Cortical Labs themselves insist that Doom is just a flashy demonstration to grab attention, but that the real potential lies in everything that can be built on top of this platform: from studios of Neuronal plasticity up to simulations of biological decision-making processes in complex scenarios.
Europe joins the race: the role of the University of Milan
Interest in these types of systems is not limited to Australia. In Europe, collaborations are also beginning to emerge focused on thoroughly exploring biological computing based on human neurons. One example is the project announced by the Italian technology consultancy Reply together with the Department of Pathophysiology and Transplantation, University of Milan.
In that agreement, researchers from the Milan Polyclinic are working with similar biological computing platforms to study how Active neurons are integrated with digital systemsThe idea is to use these tools to investigate the mechanisms of learning, memory, and neural plasticity from a perspective different from that of traditional animal models.
Professor Stefania Corti, Chair of Neurology and head of the Neuromuscular and Rare Diseases area at the Polyclinic, has emphasized that this type of experiment opens up "unprecedented opportunities" to analyze how connections in real neural networks are reorganized when faced with new tasks.
Instead of simply observing the human brain using imaging techniques or recording activity in laboratory animals, these platforms allow create specific neural configurations and expose them to specific problems, measuring very precisely the changes in their electrical activity.
Beyond Italy, European interest is also focused on potential medical applications: from testing pharmacological compounds on controlled cultures to designing new types of brain-machine interfaces that may one day help patients with paralysis or movement disorders.
Potential applications: from medicine to new AI
While the image of human neurons firing at demons in Doom is striking, the true importance of the experiment lies in what might come next. Many experts see these systems as a kind of testbed for studying biological learning in a highly structured and measurable context.
One clear area of application is research into neurological diseases. By working with stem cell-derived cultures, it is possible to create models of specific pathologies and observe how they respond to different environments or treatments, which could complement, and in some cases reduce, the use of animal models.
Another line of work involves the development of new artificial intelligence architectures directly inspired by biologyObserving how these real neural networks reorganize themselves when solving complex tasks could help design more flexible algorithms, capable of adapting with less data and more efficiently.
Some researchers also point to a connection with brain-machine interfaces. If cultured neurons can navigate smoothly—or at least with some degree of effectiveness—in an unpredictable three-dimensional environment like Doom, that same adaptability could be used to to control robotic prostheses, exoskeletons, or assistive devices in changing real-world scenarios.
All of this is accompanied by ethical and philosophical questions: to what extent should these cultures be considered as mere laboratory resources, what limits should be established on their use, or how to guarantee adequate regulation if biocomputing ends up having massive commercial applications.
For now, what's on the table is a solid proof of conceptA set of human neurons, grown on a chip and connected to a computer system, are able to learn to play Doom with remarkable speed compared to many artificial systems, although still far from an experienced human player.
These kinds of experiments, which a few years ago would have sounded like something out of a futuristic TV series, are becoming a real field of work for laboratories and companies around the world. Between public demonstrations, collaborations with European universities, and the opening of APIs to third parties, the feeling is that the computer science based on living neurons It is beginning to move beyond purely theoretical grounds and seek its place alongside traditional silicon.
Today, these neurons aren't going to dethrone e-sports champions or replace major AI models, but they do mark a turning point: they are tangible proof that some of the computing of the future could rely on them. biological substrates capable of learning on their ownand that a cell culture playing Doom may be just the first step in a much larger paradigm shift.
