In a remarkable demonstration of the intersection between biology and technology, the Australian biotechnology firm Cortical Labs recently showcased living human neurons controlling gameplay on a classic video game. Their neurons, grown on a silicon chip, played the 1993 first-person shooter Doom, navigating corridors, encountering enemies, and firing weapons, albeit clumsily and with frequent in-game deaths. This achievement, while still rudimentary, points toward a new frontier in computing and biomedical research, potentially revolutionizing how we approach artificial intelligence, energy consumption in computing, and neurological drug testing.
### How Living Neurons Play Doom
At the heart of this innovation lies a culture of approximately 200,000 human neurons grown on a microchip. These neurons were not harvested directly from human brains but were derived from stem cells, which themselves originated from accessible human tissues such as blood or skin. By isolating specific cell types and converting them into neural cells, the company created a self-sustaining neural culture housed within a silicon chip. This biocomputing platform keeps the neurons alive for up to six months using a life-support system and interfaces with the cells electrically. Since active neurons generate electrical impulses, the chip both reads these signals and delivers electrical stimuli back to the cells, creating a dynamic feedback loop.
Cortical Labs' chief scientific officer, Brett Kagan, describes the neurons' behavior as "adaptive, real-time goal-directed learning." The neurons are not playing Doom out of preference or consciousness but are responding to an engineered system that rewards predictability and punishes randomness. The company employed a concept known as the free energy principle-developed by neuroscientist Karl Friston-which posits that neural systems strive to predict their environment to maintain stability and function effectively. In this context, the neurons were trained using a feedback loop where incorrect moves in the game produced unpredictable "white noise" signals (akin to punishment), while correct moves generated predictable signals (reward). This mechanism encouraged the neural network to avoid chaos and seek order, gradually learning to navigate the game environment.
### From Pong to Doom: Progress in Neural Computing
This demonstration builds on earlier work published by Cortical Labs in 2022, where they showed that neurons on microchips could learn to play Pong, a much simpler game involving a bouncing square and a paddle. Pong's two-dimensional gameplay is relatively straightforward compared to Doom's three-dimensional corridors, multiple enemies, and complex navigation challenges. The leap from Pong to Doom represents a significant advance in the complexity of tasks the neural cells can adapt to.
To enhance the system's learning efficiency, Cortical Labs collaborated with researchers at Stanford University during a hackathon. Independent researcher Sean Cole paired the living neurons with traditional machine learning algorithms. The resulting hybrid system outperformed the algorithm alone, suggesting that the biological neurons contributed unique computational capabilities that enhanced the learning process.
### Potential Applications: Medical and Computational
Cortical Labs envisions two primary applications for this technology: medical research and computational innovation.
On the medical front, the company aims to improve the notoriously high failure rates of neuropsychiatric drug trials, which can reach 93 to 99 percent. Traditionally, drug testing involves neurons cultured in isolation without meaningful stimulation, which does not accurately reflect the dynamic environment of a functioning brain. Kagan emphasizes that neurons respond differently to drugs and exhibit disease characteristics more realistically when embedded in an interactive environment like a game or virtual world. This approach could lead to more predictive models of neurological diseases and better evaluation of drug effects, potentially reducing costly clinical trial failures.
Computationally, biological neurons represent some of the most powerful information-processing systems known. Unlike silicon transistors, which operate with binary states (0s and 1s), neurons have a far richer complexity. They can maintain multiple interacting dynamic states simultaneously, providing what Kagan calls "third-order complexity" or higher. This inherent complexity enables computations that are difficult or impossible for traditional silicon-based systems to replicate efficiently.
One key advantage of neuronal computing is energy efficiency. The human brain operates on approximately 20 watts of power-less than that of a typical household lightbulb-yet performs feats of perception, learning, and decision-making far beyond current AI systems. To match the brain's computational power using silicon-based AI would require energy millions of times greater. Researchers like Feng Guo, an associate professor at Indiana University Bloomington, highlight this potential for "high-level computing" with biological systems. Guo's own work with three-dimensional brain organoids for computing, published in Nature Electronics in 2023, underscores the promise of biocomputing platforms to dramatically reduce energy consumption in AI and computing applications.
### Challenges and Realistic Expectations
Despite the excitement, Kagan is cautious about overstating the near-term capabilities of neuronal computing. He points out that conventional digital devices still outperform biological systems in many precise and rapid calculations, such as long division on a pocket calculator. However, biological systems excel at tasks requiring flexible learning, adaptation, and interaction with complex environments-skills that current AI struggles to master.
Kagan describes biological computing as a "new tool in the intelligence toolbox," complementary rather than a wholesale replacement for existing silicon technology. While the idea of a personal computer powered by neurons in a vat sounds like science fiction, the field is steadily moving toward practical applications. A few years ago, the only published example of neural gameplay was a simple Pong demonstration. Now, Cortical Labs offers a commercial platform complete with an application programming interface (API) that developers can access, alongside video evidence of neurons navigating the challenging 3D environment of Doom.
The company's work represents a transition from theoretical research toward applied science, as they continue to refine the technology and explore its full potential. The ability to keep neurons alive and functioning in a controlled environment, interface them with software, and harness their adaptive learning abilities opens new avenues for both artificial intelligence and neurological research.
### Broader Implications and Future Directions
The intersection of living neurons and computing hardware raises profound questions about the nature of intelligence, consciousness, and the future of technology. While the neurons on the chip do not possess awareness or subjective experience, their capacity to adapt and learn in real time offers a glimpse into alternative forms of computation that go beyond binary logic.
If further developed, such biocomputing platforms could transform drug discovery, enabling faster and more accurate testing of compounds on human-like neural networks. This could accelerate the development of treatments for neurodegenerative diseases, psychiatric disorders, and other brain-related conditions.
On the computing side, integrating biological neurons could lead to new hybrid systems that leverage the strengths of both living cells and silicon-based processors. These systems might perform certain types of tasks-such as pattern recognition, sensory processing, and decision-making-with far greater efficiency and lower energy consumption than today's AI.
However, realizing these possibilities will require overcoming significant technical challenges, including maintaining neuron viability over long periods, scaling up the number of neurons, and developing robust interfaces between biological and electronic components. Ethical considerations surrounding the use of living human neurons in computing will also need careful attention.
### Conclusion
Cortical Labs' achievement of having living human neurons play Doom marks a milestone in the emerging field of biological computing. By harnessing the adaptive learning capabilities of neural cells grown on silicon chips, the company has demonstrated a novel approach that blends biology with technology in unprecedented ways. This work not only promises advances in low-power, efficient computing but also holds potential to revolutionize how neurological diseases are studied and treated.
While still in its early stages, this fusion of living neural networks with computational platforms represents a significant step forward from simple experiments toward practical applications. As the field progresses, it may well open new horizons in artificial intelligence, medicine, and our understanding of the brain's remarkable computational power.