Scientists use AI system to translate brain activity into text



A brand new AI system that converts mind exercise into textual content might rework communication for individuals who can’t converse or sort.
The mind-reading machine makes use of mind implants to trace what neurons do when somebody speaks.
An algorithm then converts the information right into a string of numbers, which one other system then interprets right into a sequence of phrases.
For one participant, simply 3% of sentences wanted to be corrected.
This made the AI extra correct than skilled human transcribers, whose error charges have been recorded at 5%.
[Learn: Gamers’ brains will teach AI how to control military drone swarms]
The system was developed by researchers from the College of California, San Francisco, who revealed their findings in the journal Nature Neuroscience.

Turning mind exercise into textual content

The researchers examined their system on 4 folks with mind implants that monitor their epileptic seizures.
The members have been requested to learn a variety of easy sentences aloud, together with “Tina Turner is a pop singer” and “the oasis was a mirage.”
As they spoke, their mind alerts have been fed into a pc and decoded into phrases.
The system typically struggled to transform the phrases into coherent sentences. For instance, it translated “these musicians harmonize marvelously” into “the spinach was a well-known singer.”
However the AI was nonetheless more accurate than other speech transcription systems.
The researchers pressured that the tech at present solely works when somebody is talking aloud.
Nevertheless, it might be upgraded to translate the ideas of people that can’t talk verbally, equivalent to these with locked-in syndrome, a neurological dysfunction that causes paralysis.
Revealed March 30, 2020 — 17:39 UTC


Source link 

Comments