Scientists have made a breakthrough in the realm of Artificial Intelligence (AI) and their ability to foretell human thoughts. A new AI model called a semantic decoder can read and interpret brain activity scans, with the assistance of models similar to the legendary OpenAI ChatGPT and Google’s Bard, in order to find out “the gist” of someone’s thought. This advancement can prove greatly beneficial to help those who, for medical or injury-related reasons, cannot physically speak.
The study was published in the journal Nature Neuroscience and scientists from The University of Texas at Austin, US have stated that this new AI system is a leap forward from what has been achieved to date. With no requirement for surgical implants, the process is non-invasive. Brain activity is measured by an fMRI scanner. After completing extensive training, people are made to listen to podcasts, stories, or imagine narrating a story and the AI will then generate the corresponding text based on the brain activity alone.
The output of the AI will not be a direct transcript of what the subject is thinking but instead will provide an accurate “gist” of the idea being formed. In the experiments that have taken place, the AI was successful in producing text that was very close to the intended content and, in some cases, was exactly matched. An example given was when a subject was listening to a sentence, “I don’t have my driver’s license yet”, the AI generated the sentence “She has not even started to learn to drive yet”.
The AI technology has potential for misuse, like intrusion of citizen privacy by government authorities. To address this, scientists have devised a system where cooperative participants are required, who can be trained as part of a process that can last up to 15 hours in an MRI scanner. Furthermore, if an unwilling individual wishes to be protected from having their thoughts being decoded, they can imagine animals or start to compose a story of their own, thus foiling the AI system.
The AI system also currently has no practical applications outside the lab environment as it still needs an fMRI machine to work. To protect individuals and their sense of privacy, the researchers are calling on the implementation of laws in order to dictate the usage of such AI technology.
The research was led by Jerry Tang and Alex Huth, whose combined efforts on this project have been well recognized in the field of AI and brain studies. Tang is a researcher at the University of Texas at Austin, US and Huth is a postdoctoral researcher at San Francisco State University, US.
