Biological Chips Master Video Game Mechanics, Signaling a Shift in Computing Efficiency

2026-02-28 general

London, Saturday, 28 February 2026.
In a major biocomputing breakthrough, human neurons on a chip learned to play Doom in just one week, paving the way for ultra-efficient organic hardware to challenge silicon.

From Pong to First-Person Shooters

In a demonstration that bridges the gap between biological potential and digital execution, independent developer Sean Cole has successfully trained a neuron-powered chip to play the classic first-person shooter Doom in approximately one week [1]. This achievement utilizes the CL1, a commercial biological computer developed by Melbourne-based startup Cortical Labs, which houses roughly 800,000 living human neurons derived from stem cells [3][4]. While the neurons currently play at a level described as better than random but inferior to a human, the speed of their learning curve is the critical metric here; unlike the company’s previous work with Pong—which represented years of scientific labor—this integration was achieved in mere days using a newly developed Python interface [1].

The Mechanics of “Wetware” Gaming

The CL1 system operates by growing human neurons directly onto a silicon chip equipped with a high-density multielectrode array (MEA), allowing the hardware to send electrical pulses to the cells and record their firing patterns [4]. The learning process is driven by the “Free Energy Principle,” a theory suggesting that neurons naturally organize to minimize chaos or “surprise” in their environment [4]. In the context of Doom, the game environment is translated into electrical sensory signals; if the neurons perform a beneficial action, they receive a predictable signal, whereas a mistake results in chaotic electrical noise [4]. To avoid this chaotic feedback, the neurons physically reorganize their connections to favor gameplay actions that yield predictable outcomes [4].

Efficiency and Commercial Viability

Beyond the novelty of gaming, this development highlights the extreme energy efficiency of biological computing compared to traditional silicon. The human brain functions on approximately 20 watts of power, whereas training massive silicon-based AI models consumes megawatts [4]. This efficiency is central to the value proposition of the CL1 units, which began shipping in 2025 [3]. A single CL1 unit is priced at $35,000, though the cost drops significantly for enterprise-scale deployments; a unit within a 30-unit server rack costs $20,000, representing a per-unit savings of 15000 [3]. These racks consume between 850 and 1,000 watts, offering a potentially sustainable path for energy-intensive computational tasks [3].

Future Applications and Ethical Frontiers

The ability to program biological chips via standard languages like Python signals that this technology is moving toward broader enterprise adoption [1][2]. Yoshikatsu Hayashi of the University of Reading suggests that mastering Doom is a precursor to complex real-world tasks, such as controlling robotic arms [1]. Furthermore, the technology offers a new paradigm for pharmaceutical research; scientists can use these systems to test how drugs for conditions like dementia affect neural learning without relying on animal or human subjects [4]. This aligns with recent regulatory shifts, such as the FDA’s April 2025 announcement promoting organoids for drug safety testing [6]. However, as the sector grows—evidenced by the launch of the “Cortical Cloud” for developers—ethical questions persist regarding the sourcing of biological material and the long-term implications of “Synthetic Biological Intelligence” [2][3].

Sources

Biotechnology Biocomputing