Norman likens the device to an “MRI machine leaning to the side and suspended above the user’s head.”
Furthermore, the second king whose subject’s head moves says that the signal is lost. “Our efforts are not about the product,” he says. “In fact, my message is that I always say I don’t think there’s a way for products.
The typing project was carried out with 35 volunteers at the Spanish research site, the Basque Centre for Cognition, Brain and Language. Each spends about 20 hours within scanner typing phrases such as “El Proquesadore Ejectara Rhine Struchon” (processor executes instructions), and brain signals are fed into a deep learning system where Meta is calling Brain2Qwerty. Ta. Keyboard letters.
The job of that deep learning system is to understand which brain signals mean that A is entering a. This means Z and so on. Ultimately, after individual volunteers type thousands of characters, the model can infer what important people are actually pushing against them.
In the first preprint, Meta researchers reported that the average error rate was about 32%, with one of the three-character sentences being incorrect. Still, according to Meta, the results are most accurate in brain typing results using a complete alphabetical keyboard and signals collected outside the skull.
The most effective approach is used directly on electrodes embedded in the brain, or on their surfaces, but research into brain reading has progressed rapidly. These are known as “invasive” brain computer interfaces. They require brain surgery, but they can collect electrical information very accurately from small groups of neurons.
