Human Neurons on a Chip Learn to Play Doom

Cortical Labs, an Australian biotech company, has successfully demonstrated that a cluster of approximately 200,000 living human neurons grown on a silicon chip can play the classic video game Doom. This achievement marks a significant step in the development of what the company describes as the "world's first code deployable biological computer." The neurons, which are not sourced from human brains but rather derived from reprogrammed white blood cells, interact with the game through electrical signals, allowing them to navigate corridors, encounter enemies, and execute actions—albeit with limited proficiency.

Smithsonian MagazineScientific American

The neurons' ability to learn was facilitated by a feedback loop based on the free energy principle, developed by neuroscientist Karl Friston. This principle posits that neural systems are driven to predict their environment. In this setup, correct actions produced predictable signals, while incorrect ones resulted in chaotic noise, effectively rewarding the neurons for successful predictions and punishing them for failures. This method allowed the neurons to adapt and learn to play Doom, showcasing what Cortical Labs' Chief Scientific Officer, Brett Kagan, refers to as "adaptive, real-time goal-directed learning."

Scientific AmericanSmithsonian Magazine

Cortical Labs envisions two primary applications for this technology. The first is in the medical field, where the ability to test drugs on living neurons in a dynamic environment could enhance the understanding of neuropsychiatric disorders. Kagan notes that traditional drug testing methods often fail, with 93 to 99 percent of clinical trials not succeeding. By utilizing neurons in a game-like environment, researchers can observe how these cells respond to various treatments, potentially leading to more effective therapies.

Scientific American

The second application is computational. Kagan argues that biological neurons possess a complexity that far exceeds silicon-based systems, potentially leading to significant energy savings. For instance, the human brain operates on just 20 watts, while silicon-based AI systems require vastly more power. This efficiency could revolutionize computing, as highlighted by Feng Guo, an associate professor at Indiana University Bloomington, who sees the biocomputing platform as capable of high-level computing.

Scientific AmericanSmithsonian Magazine

Despite the promising developments, Kagan cautions against overestimating the immediate capabilities of biological computing. He emphasizes that while biological systems can outperform traditional algorithms in certain tasks, they are not yet ready to replace conventional computing methods. The research is still in its early stages, and Kagan acknowledges the need for further exploration before practical applications can be realized.

Scientific American

• “A pocket calculator will outperform me at long division any day,” — Brett Kagan, Chief Scientific Officer, Cortical Labs

Scientific American

• “People are looking at it from biomedical research angles, for disease modelling,” — Hon Weng Chong, CEO, Cortical Labs

Smithsonian Magazine

• “Right now, the cells play a lot like a beginner who’s never seen a computer—and in all fairness, they haven’t,” — Brett Kagan, Chief Scientific Officer, Cortical Labs

Smithsonian Magazine

Cortical Labs aims to continue refining its biocomputing technology, with aspirations to explore additional applications, including potentially training neurons to play more complex games. The overarching goal remains focused on leveraging this technology for advancements in medical research and drug development.

Scientific American