In a recent study, researchers sent signals from a sensor detecting tactile patterns associated with different Braille letters that were received by brain organoids. Individual organoids were able to correctly decode them 61 percent of the time but three organoids boosted accuracy to 83 percent.
Their biohybrid system used brain organoids less than a millimeter across and placed on multielectrode arrays contained in a system that keeps the organoids alive. Environmental Working Group and Greenpeace don't need to start suing, these are only organoids with only about 10,000 neurons and supporting cells but lack most of the structures of an actual brain. They can't learn over time. The multielectrode arrays allow them to receive electrical signals so researchers applied a sensor that records tactile data to the 26 letters of the Braille alphabet. Those data were converted to electrical stimulation patterns the organoids received, to see if they would produce neural activity a support vector machine would see as the patterns and detect whether given signals corresponded to specific letters.
Credit: Tianyi Liu
The paper also showed that three were better at distinguishing real signals from noise than the decoder working directly on the obscured signals, without involvement of the organoids. They applied various noise filters to the Braille letter signals to compare decoding from the organoid responses with direct decoding of the obscured signals. The machine decoding turned out to be easier to fool than decoding from the organoid responses.
It's still unclear why three would be better than one. Speculation is that it could be due to individual variation - different network connections and developmental states - causing the organoids to complement each other. Before they can build a predictable and standardized system they'd have to figure that out.
Citation: Liu T, Philamore H and Ward-Cherrier B, 'Encoding tactile stimuli for Braille recognition with organoids', Front. Artif. Intell. 9:1800219. doi: 10.3389/frai.2026.1800219
