Meta has an AI for brain typing, but it’s stuck in the lab

Norman likens the device to “an MRI machine tipped on its side and suspended above the user’s head.”

What’s more, says King, the second a subject’s head moves, the signal is lost. “Our effort is not at all toward products,” he says. “In fact, my message is always to say I don’t think there is a path for products because it’s too difficult.”

The typing project was carried out with 35 volunteers at a research site in Spain, the Basque Center on Cognition, Brain, and Language. Each spent around 20 hours inside the scanner typing phrases like “el procesador ejecuta la instrucción" (the processor executes the instruction) while their brain signals were fed into a deep-learning system that Meta is calling Brain2Qwerty, in a reference to the layout of letters on a keyboard.

The job of that deep-learning system is to figure out which brain signals mean someone is typing an a, which mean z, and so on. Eventually, after it sees an individual volunteer type several thousand characters, the model can guess what key people were actually pressing on.

In the first preprint, Meta researchers report that the average error rate was about 32%—or nearly one out of three letters wrong. Still, according to Meta, its results are most accurate yet for brain typing using a full alphabet keyboard and signals collected outside the skull.

Research on brain reading has been advancing quickly, although the most effective approaches use electrodes implanted into the brain, or directly on its surface. These are known as “invasive” brain computer interfaces. Although they require brain surgery, they can very accurately gather electrical information from small groups of neurons.