If there’s one thing I wish I could do, was read people’s minds. Some may say it’s a boon; for others, it’s a bane. Reading people’s minds is something that’s close to being magical.
Thousands of people claim they can read minds, but AI has won the race this time.
Surprisingly, an AI-based decoder has been discovered that can be able to read a person’s mind. Being one of a kind, the system translates activity that occurs in the brain in the form of text, reading thoughts non-invasively.
The new state-of-art system reconstructs speech with efficient accuracy, while the person listened to a story or even imagined one. This is done using the fMRI scan data.
The research suggests that surgical implants were required for previous language decoding systems. The most recent breakthrough offers the possibility of novel approaches to regaining speech for individuals who have difficulty communicating as a result of a stroke or motor neuron disease.
Lead Research Neuroscientist Dr. Alexander Huth from the University of Texas at Austin says, “We were kind of shocked that it works as well as it does. I’ve been working on this for 15 years … so it was shocking and exciting when it finally did work.”
The accomplishment addresses a fundamental constraint of fMRI, namely, that despite its exceptional ability to pinpoint brain activity to a precise location, there is an inherent delay that prevents real-time tracking of activity.
Huth says, “It’s this noisy, sluggish proxy for neural activity.”
As reported by the Guardian, the delay is present in fMRI scans because they measure the blood flow response to brain activity, which reaches its peak and returns to baseline over approximately 10 seconds. As a result, even the most advanced scanner cannot enhance this response time.
To add more, this hard constraint has hampered the AI’s ability to interpret brain activity in response to natural speech, causing a spread of a “mishmash of information” for a few seconds.
Large language models, however, are bringing the new way in. One such system is the OpenAI’s ChatGPT that are able to provide the semantic meaning of speech to number, which allows scientists to look at patterns of neuronal activity. Meaning, these models allow researchers to identify patterns of neuronal activity that align with specific meanings of word sequences instead of trying to decipher activity word by word.
The rigorous training process required three participants to undergo 16 hours of podcast listening each inside a scanner. The decoder was educated to associate brain activity with meaning by employing GPT-1, an earlier version of ChatGPT.
Be aware, AI can get into your head!