Brain implants and neural networks helped a 65-year-old man paralyzed below the neck to type text messages on a computer at 90 characters per minute. The system’s accuracy was 94%,reported by researchers from Stanford University.
Researchers attached two widgets to the surface of the patient’s brain, inserting electrodes about a millimetre deep. They then asked the patient, over three days, to imagine himself writing 572 sentences. The excerpts contained all the letters of the alphabet, spaces and punctuation marks.
The signals from the electrodes were fed into a recurrent neural network as input data. The model was trained to convert each reading from the patient’s brain into a specific letter.
Patterns of brain waves recorded when thinking about writing the letter ‘a’ by hand differed from those that arose when imagining ‘b’. In this way, the software could be trained to link signals for ‘a’ with the letter ‘a’, and so on. When the subject thought about writing each character in a sentence, the neural network decoded the sequence of brain signals into the required characters.
Using a dataset of 31,472 characters, the machine-learning algorithm learned to decipher the patient’s brain signals for each symbol.
The system has no delete function, so messages could contain errors. However, researchers reduced the likelihood from 6% to 3.4% through autocorrection.
“Taken together, these results suggest that even years after paralysis, the neural representation of handwriting in the motor cortex is likely strong enough to be useful for a brain–machine interface,” the team wrote.
John Ngai, director of the BRAIN Initiative at the US National Institutes of Health, who did not participate directly in the study, called it a ‘milestone’ for neurointerfaces and machine-learning algorithms.
“This knowledge creates a critically important foundation for improving the lives of others with neurological injuries and disorders,” he said.
However, the research team acknowledged that many problems must be overcome before this kind of technology can be used by a much larger number of people.
According to them, at this stage, researchers will need to retrain their model for each individual’s brain signals, and performance may vary.
In April 2021, Sony patented artificial intelligence that tracks a person’s playing style and attempts to replicate it.
In March, researchers from the University of Georgia developed an AI backpack to help people with visual impairment navigate their surroundings.
Subscribe to ForkLog’s Telegram news: ForkLog AI — all the news from the world of AI!