Covid-19
Advertisement

Scientists Create AI That Turns Thoughts into Text

The end result could prove hugely useful for people who are unable to talk or type.

Scientists Create AI That Turns Thoughts into Text
Highwaystarz-photography/iStock

Mind reading may soon be possible: a team of scientists has developed artificial intelligence (AI) that can translate someone's brain activity into text.

The creation from the University of California, San Francisco (UCSF) team may prove very useful for people who are unable to talk or type, for instance, those who suffer from Locked-in syndrome

Their study was published in Nature Neuroscience.

SEE ALSO: HOW MACHINE LEARNING AND AI WILL IMPACT ENGINEERING

Not there yet

"We are not there yet but we think this could be the basis of a speech prosthesis," said co-author on the paper Dr. Joseph Makin from UCSF. 

In order to get their results, the team enrolled four participants who had electrode implants placed in their brains to monitor epileptic seizures. Then the participants were asked to repeat 50 different sentences aloud while their brain activity was being tracked by the monitor. 

Once the data from the 50 spoken sentences was collected it was placed into a machine-learning algorithm, which in turn changed them to a string of numbers and compared them to the audio recording. After a while, the system was able to convert the numbers into English sentences. 

Initially, these sentences made little to no sense, but as the machine kept comparing each sequence of words with the sentences that were actually said, it started learning and improving. The end result still isn't perfect, with sentences such as "Those musicians harmonize marvelously" being translated as "The spinach was a famous singer", or "A roll of wire lay near the wall" turned into "Will robin wear a yellow lily".

However, the researchers found that the accuracy of their system was far higher than previous approaches. The error rate was as low as 3% on average, when compared with the average score of 5% by professional transcribers this is a positive jump down.

All in all the system took just 40 minutes to learn, proving just how fast and dependable it could be. 

One major point that still needs to be addressed, however, is that in this study Makin and his colleagues depended on brain activity provided by spoken speech, rather than inner thoughts — something that wouldn't be useful to patients suffering from Locked-in syndrome for example. 

Furthermore, as it currently stands the system requires electrode arrays to be implantedinto the brain, which isn't something everyone would be willing to accept. 

All that being said, the team is still working on its project and keeps progressing towards a much-needed technology. 

Advertisement
Follow Us on

Stay on top of the latest engineering news

Just enter your email and we’ll take care of the rest:

By subscribing, you agree to our Terms of Use and Privacy Policy. You may unsubscribe at any time.

Advertisement