AI language models could help diagnose schizophrenia, study says

Artificial intelligence (AI) language models could help in diagnosing schizophrenia, according to the findings of a new research.

Researchers at University College London in the UK developed new tools, based on AI language models, that can characterise subtle signatures in the speech of patients diagnosed with schizophrenia, which is characterised by distortions of reality and disturbances of thought and language.

An AI language model is a natural language model which generates probable sequences of words based on the text dataset it was trained on. While the most popular ones include ChatGPT and Bard among others, in this study, the researchers used Facebook AI Research's (FAIR) fastText.

They then used their new AI tools in verbal fluency tasks that 52 study participants - 26 having schizophrenia and 26 control - were asked to perform. The findings are published in the journal Proceedings of the National Academy of Sciences.

In these tasks, the participants were asked to name as many words as they could either belonging to the category 'animals' (category fluency) or starting with the letter 'p' (letter fluency) in 5 minutes.

The researchers then tested their AI tools to see if they could predict the words that the participants recalled and spoke out spontaneously.

The answers of the control group were more easily predicted by the AI model than those of the schizophrenic patients, the researchers found, adding that the difference grew bigger with more severe symptoms.

The researchers think that this difference might have to do with the way the brain learns relationships between memories and ideas, and stores this information in so called 'cognitive maps'.

Schizophrenia is linked to abnormalities in neural processes that support these cognitive map representations, thus disrupting their associative cognition, which involves linking concepts from memory.

'We test a neurocognitive hypothesis where we propose that differences in verbal fluency relate to cognitive and neural processes that underpin associative (relational) cognition,' the researchers write in their study.

They found support for this theory in a second part of the same study where the authors used scans to measure activity in parts of the brain involved in learning and storing these 'cognitive maps'.

'Until very recently, the automatic analysis of language has been out of reach of doctors and scientists. However, with the advent of AI language models such as ChatGPT, this situation is changing.

'This work shows the potential of applying AI language models to psychiatry - a medical field intimately related to language and meaning,' said lead author Matthew Nour, a psychiatrist and a neuroscientist.

Currently, psychiatric diagnosis is based almost entirely on talking with patients and those close to them, with only a minimal role for tests such as blood tests and brain scans.

However, by combining AI language models and brain scanning technology, how the brain constructs meaning is being uncovered and how this meaning construction goes awry in psychiatric disorders, the researchers said.

'There is enormous interest in using AI language models in medicine. If these tools prove safe and robust, I expect they will begin to be deployed in the clinic within the next decade,' said Nour.