从实验室培养的大脑类器官到硅基计算的替代方案
In a quiet laboratory in Melbourne, a cluster of neurons smaller than a peppercorn is quietly learning to play a simplified version of Pong. These are not cells scraped from a donor's cortex; they are cerebral organoids—lab-grown, three-dimensional aggregates of human brain tissue that have been coaxed from stem cells into rudimentary cortical structures. For years, organoids have served as disease models for Zika, autism, and Alzheimer’s. But a more audacious ambition is now taking root: using these living neural networks as the hardware for a new kind of biocomputer that could one day outpace silicon in efficiency and adaptability.
The logic is deceptively simple. A single human brain, consuming roughly 20 watts, can outperform a supercomputer that draws megawatts. Organoids offer a scaled-down version: tens of thousands of neurons, interconnected through synapses that fire and learn in real time. Researchers at institutions from the University of Tokyo to Johns Hopkins have already demonstrated that organoids can be trained to recognise patterns and control robotic arms through feedback loops of electrical stimulation. Unlike classical AI, which runs on rigid transistors, these mini-brains display plasticity: they rewire themselves without explicit reprogramming, hinting at a form of hardware that evolves as it learns.
Vocabsavvy AI · a Scientific-American-style science communicator · Vocabsavvy Original