Crazy: Scientists Compute With Human Brain Cells

Human brain cells can be used to compute with a fraction of the energy consumed by today’s AI systems—about 100,000 times less—yet the field is still largely sidelined compared with flashier approaches like quantum computing. A key example is Corticle Labs’ biological computer, the CL1, which grows roughly 800,000 living human neurons on a chip. The neurons connect to electronics that can stimulate them and record their activity, while a life-support module keeps the cells alive for several months. Launched in 2025 and reportedly costing around $35,000 (with cloud access available), the CL1 is designed less for consumer products and more for research labs exploring how far “wet” computation can go.

Corticle Labs’ earlier breakthrough came in 2022, when researchers trained a cluster of live neurons to play Pong. The setup encoded the ball’s position as a stimulus and required the neural activity to update a hypothetical paddle position. Importantly, the learning mechanism relied on external electronic feedback: the system delivered corrective signals based on how close neuron firing was to the desired outcome, rather than letting the neurons independently discover the task. A similar effort is underway at Switzerland’s Final Spark, which grows smaller brain chunks—about 10,000 neurons each—on electronic chips. Final Spark also offers remote access for as little as $5,000 per month and has reported research showing one neuron chunk can encode Braille with about 80% accuracy.

The promise extends beyond energy savings. Unlike conventional AI, where learning is largely software-driven, these organic systems adapt through changes in their physical connections while they operate. That physical learning loop—embedded in living tissue—could, in theory, point toward more general forms of intelligence. The field also overlaps with neuromorphic chips, which aim to mimic aspects of brain-like computation but remain early-stage.

Still, three practical hurdles loom. First, the neuron cultures don’t last long—typically only a few months. Second, scaling up from small brain chunks to larger, controllable systems is an open engineering problem. Third, reproducibility is weak because organically grown tissue varies from sample to sample. Even the question of whether these systems can learn to “think” on their own remains unsettled; demonstrations so far depend heavily on engineered feedback.

Then come ethical and philosophical concerns. Growing and training human brain organoids raises questions about sentience, suffering, and what researchers should even look for if consciousness emerges. The transcript also challenges the premise of building human-like intelligence from human tissue: if the goal is a new mind, why not start with a baby? For now, the research continues as a high-risk, high-interdisciplinary path—one that may eventually lead to computing systems that feel less like software and more like living substrates, potentially even “cloud” cognition. But widespread use remains far off, and the immediate reality is experimental, fragile, and tightly controlled.