The use of living neurons, grown from human stem cells and mounted on silicon chips, as computational elements. Proponents hope that neurons, packaged into "biological processing units" and slotted into racks at conventional data centres, might one day complement the transistor-based chips that have defined computing for the past half-century.
Neurons offer several advantages over electronics. Efficiency is one: modern AI models consume power by the millions of watts, whereas a typical human brain—made up of almost 90bn neurons—consumes roughly 20 watts. Sophistication is another: transistors are binary switches (on or off), while neuron behaviour depends on variables including cell-membrane voltage and the timing of signals from other neurons. Conventional computer architectures also store information far from where processing happens—Micron, a big memory-chip maker, estimates that up to half the energy budget of an AI processor is spent shifting data around. Brains mix data and processing side by side, minimising such logistical problems.
Brett Kagan, chief scientific officer of Cortical Labs, speculates that this may make neurons better suited to interpreting messy, analogue real-world signals. He cites "Moravec's paradox"—the counter-intuitive observation in AI research that abstract reasoning (chess, arithmetic) is computationally easier, in some fundamental way, than the trivial-seeming motor skills needed to navigate the physical world.
On March 3rd 2026 DARPA, a research-funding agency of the American government, announced a programme into biological computing, with the aim of producing biological processing units that might use a fraction of the energy of conventional silicon chips and which might prove useful for tasks such as autonomously flying drones.
You will feel hungry again in another hour.