American researchers have managed to create an artificial intelligence device with unheard of skills. This machine is able to convert brain signals into intelligible words!
Latest experiment shows progress
Researchers from the Zuckerman Institute from the University of Columbia (United States) described the working functions of their artificial intelligence device. Their study was published in the journal Scientific Reports on the 29 January 2019. The researchers newly developed device is able to convert electric brain signals into intelligible words.
Their results were gathered during an ultimate experiment which follows similar experiments including the piloting of a drone using only thoughts or even the collaboration of three people on a section of Tetris. However in each of these experiments, the system relied on simplistic commands which required a certain amount of communication. This however was not the case in this latest experiment.
What was the approach of the experiment?
The system is developed on the basis of words heard by the subject and not by the vocalisation of their thoughts. In any case, the experiment involved the transformation of these signals into sound. Researchers have therefore focused on the auditory cortex. This is the part of the brain which receives and analyses information from heard sounds. Deep learning has been integrated into the Artificial Intelligence device to train it to recognize words and analyze their own recordings.
Signal recordings have been carried out using a system of invasive electrodes which aims to measure the brain’s activity. Researchers have benefited from authorization as part of neurosurgical intervention on patients with epilepsy. Sufferers were then set up with electrodes that aimed to measure intracranial electroencephalography. The recordings were obtained while the suffers heard stories that were read by four people: two men and two women.
Very encouraging results
Other tests have been carried out, particularly on objective and subjective intelligibility. Researchers used for example no less than 40 recording of numbers from zero to nine with the same people. These numbers were recited by four different people from those that recounted the stories. Signal recordings were then placed under artificial intelligence device which reconstructed the sounds, which 11 other volunteers heard. These volunteers were then asked to indicate the figure that they heard, the quality of the sound as well as the way the person formulated the figure.
According to the results the identification rate of the number was at 75% and the grade given by volunteers on the sound quality was on average 3.4 out of 5. The success rate for those that distinguished the sex of the speaker was 80%.
Research is currently still being carried out as scientists are now trying to reconstruct the sound of more complex phrases and words. Zones of the brain responsible for speech need to be targeted and a less invasive method needs to be developed. If scientists are successful, it would then be possible to control complex objects by thought or even to give a voice back to people that are now incapable of speaking.