Brain scans and AI could transcribe what people are thinking

Scientists have announced a breakthrough in the decoding of human thoughts, utilising a combination of brain scans and artificial intelligence (AI) models. The technology uncovers “the gist” of an individual’s thoughts and may significantly benefit those who have lost their ability to communicate. Although concerns about potential impacts on “mental privacy” were raised, tests were conducted to show that the language decoder could only function on those who permitted extensive training on their brain activity.

Rather than invasive brain implants traditionally used for “brain-computer interfaces,” the research team created a decoder that operates “at the level of ideas, of semantics, of meaning,” according to neuroscientist Alexander Huth from the University of Texas at Austin. This marks the first development of such a system that does not require implantation.

For the study, three participants listened to spoken narrative stories inside a functional magnetic resonance imaging (MRI) machine for a total of 16 hours. Through this, the researchers were able to identify the language-processing centres of the brain and patterns associated with spoken words, phrases, and meanings. The information was then utilised by an AI-driven neural network language model, using GPT-1, trained to predict brain responses to speech.

The model’s accuracy was then tested by having the participants listen to new stories while still in the MRI machine. The decoder successfully captured the broader sentiment of phrases but struggled with more specific elements such as pronouns.

Even during participant experimentation with silent movies or the creation of their own stories, it demonstrated the ability to recognise the general idea. This is because the MRI scans cannot capture individual words due to the speed of the language, but they can aggregate “information over a few seconds,” Huth stated.

David Rodriguez-Arias Vailhen, a bioethicist, highlighted the potential of machines “able to read minds and transcribe thought,” though he cautioned that there exists the risk of this taking place against individuals’ will. Despite these concerns, the researchers pointed out that the decoder relies on detailed training on a person’s brain activity, and the participants were able to easily disrupt the decoder by using various tactics to distract themselves.

The research team aims to improve the process further by speeding up the decoding of brain scans in real time, reported Bangkok Post. They also emphasised the need for regulations to ensure the protection of mental privacy as this technology progresses.