JVTech News Meta is designing an AI that can read your mind: a historic breakthrough?
Artificial intelligence is taking more and more space in our lives, and it’s not Meta who is going to say the opposite. An AI that can interpret people’s thoughts already exists, and gives more than surprising results.
An AI, which reads minds, really? Yes
The ability to decode language with brain activity is a very important issue concerning for neuroscience but especially on the health side. Many patients who have difficulty expressing themselves, communicate or who are in a vegetative state could benefit from a technological advance such as this. The solutions that exist today are risky and invasive, requiring an operation to implant electrodes.
At Meta, we’re not just busy creating dubious virtual worlds or finding new ways to peck users’ personal data. Other teams seem determined to make the world a better place with the development of artificial intelligence to interpret brain activity and thus decode letters and words.
“But it’s science fiction after all!”, are you going to tell me. No way. And the process of coming to understand the human brain is very interesting, you’ll see. Jean-Rémi King is a CNRS researcher and works in the department of cognitive studies at the École Normale Supérieure in Paris. It is also part ofa research unit in the company Meta, with whom he created an extraordinary artificial intelligence.
First, King and his team trained the “machine” to detect and understand words and phrases across 56,000 hours of recordings in 53 languages. This tool has become the basis for a study conducted on 169 participants to analyze their brain activity. During the experiments, the volunteers simply listened to different stories, and their brains were scanned in order to determine the magnetic and electrical reactions linked to such and such a syllable, or such and such a word.
With this data collected, it was now time to put the AI to the test. After taking into account the physical differences between each brain, the goal was to guess what the participants were hearing, simply by picking up brain activity. Thereby, among 1000 possibilities, the correct answer was present in the top 10 of the machine 73% of the time. A result that surprised the scientific community, including Giovanni Di Liberto, computer engineer at Trinity College Dublin:
The performance of artificial intelligence has exceeded what many thought possible at that time.
Even without taking anything away from the research team (this is a big step towards an upcoming revolution), everything is not yet ready to really read people’s minds in a simple and practical way.
For now, the AI is able to decode words and letters from a predefined list. Gold, human language has an infinity of words, of different ways of saying it and of changing expressions. For this to become practical on a daily basis, it will still take a lot of work.
Finally, in addition to making the machine lighter and cheaper to manufacture, the end goal to help people who are unable to communicate is to decipher the intention, whether it is to say yes or no, to express a desire, a request… One day, perhaps.