Researchers say the artificial intelligence behind the chatbot program ChatGPT, made famous for its ability to create human-like responses when prompted, could help doctors detect the early stages of Alzheimer's disease.

A study from the Philadelphia-based Drexel University School of Biomedical Engineering, Science and Health Systems found Open AI's GPT-3 program was 80 per cent accurate in predicting early signs of dementia.

The study's authors write in their peer-reviewed paper, published on Dec. 22 in the journal PLOS Digital Health, that the program offers a "promising approach" for assessing Alzheimer's disease.

"Our proof-of-concept shows that this could be a simple, accessible and adequately sensitive tool for community-based testing," Hualou Liang, a Drexel professor and co-author of the study, said in a news release.

"This could be very useful for early screening and risk assessment before a clinical diagnosis."

OpenAI is a non-profit research organization founded by Silicon Valley investor Sam Altman and billionaire Elon Musk, whose funders include venture capitalist Peter Thiel.

ChatGPT gained widespread attention after OpenAI offered public testing of the AI-powered chatbot, surprising users with its ability to provide responses to a range of questions.

Researchers behind the GPT-3 study say because language impairment affects between 60 and 80 per cent of dementia patients, programs that can detect "subtle clues," such as grammar and pronunciation mistakes, hesitation, and forgetting the meaning of words, could help determine whether a patient needs a full examination.

Although there is no cure for Alzheimer's, the researchers say therapeutic options exist if the disease is spotted early.

"The most commonly used tests for early detection of Alzheimer's look at acoustic features, such as pausing, articulation and vocal quality, in addition to tests of cognition," Liang said. "But we believe the improvement of natural language processing programs provide another path to support early identification of Alzheimer's."

With help from the U.S National Institutes of Health, researchers say they trained the program using transcripts from speech recordings, gathered specifically to test the ability of natural language processing programs to predict dementia.

They say the program was able to produce a profile of Alzheimer's speech, also known as "embedding," which the researchers then used to re-train the program to pick out transcripts produced by someone developing Alzheimer's.

The program also performed better in predicting the severity of a person's dementia compared to an "acoustic" analysis that looked at features such as pauses, voice strength and slurring.

Felix Agbavor, a doctoral researcher and the paper's lead author, said training GPT-3 with a "massive dataset of interviews," including some with Alzheimer's patients, would provide the information needed to detect speech patterns that could then identify future patients.

Moving forward, the researchers plan to develop a web app that could be used as a pre-screening tool for Alzheimer's at home or in a doctor's office.

With files from CNN and Reuters


This story has been corrected to say the study focused on the artificial intelligence behind ChatGPT, known as GPT-3.