The mind-blowing Artificial Intelligence-powered ChatGPT, which can produce a human-like response to any task that involves language, has been making headlines since its recent release because of its remarkable capabilities.
It can generate convincing articles, reports and much else on any subject. Even scientists have been fooled, unable to tell the difference between AI-generated abstracts and original human-written scientific abstracts.
Of course, there are pitfalls from the use of this technology that concern many. These include epic language fails in articles and reports; however, there’s one very clear health benefit: AI is, without a doubt, accurately predicting the early stages of Alzheimer’s disease.
We’ve talked in this newsletter before about how language impairment is a symptom in 60 to 80 percent of Alzheimer’s patients and often occurs many years before the disease is diagnosed. In fact, language impairment can be one of the very first symptoms of the disease.
Previous research, some of which we’ve told you about here, has shown valuable clinical information can be gleaned from these linguistic changes. What’s more, identifying them early on using technology means you’ve got the potential to provide a quick, cheap, accurate and non-invasive diagnosis of Alzheimer’s disease and begin an intervention.
Programs already exist that can pick up on subtle clues such as hesitation and mistakes in grammar and pronunciation. But ChatGPT, and its even more powerful language model cousin, GPT-3, are seen as a step up from existing programs.
To discover if this was true, researchers from Drexel University, Philadelphia, put the AI technology through its paces.
AI Taught To Recognize Alzheimer’s Patients’ Speech Patterns
The Drexel team collected 237 audio recordings of Alzheimer’s patients and healthy volunteers. The recordings were then converted to text using a pre-trained speech recognition model.
The program captured meaningful characteristics of the word-use, sentence structure and meaning from the text to produce what researchers call an “embedding” – a characteristic profile of Alzheimer’s patients’ speech patterns. The researchers then used the embedding to re-train the program — turning it into an Alzheimer’s screening machine.
Predicts Alzheimer’s With 80 Percent Accuracy
To test the program, the researchers asked it to review dozens of transcripts and decide whether each one was produced by someone with or without Alzheimer’s.
The model was able to predict the disease with a suprising accuracy of 80.3 percent. This was significantly better than the 74.6 percent accuracy achieved when they applied a more conventional approach that relied on acoustic features painstakingly identified by human experts.
Surpasses Two Other Tests
The Drexel team also compared GPT-3 with three other approaches using large language models. The program matched the top model’s performance and outshone the other two.
In a final test, GPT-3 was able to accurately infer, solely based on speech data, the subject’s cognitive testing score on the Mini-Mental State Exam, a common test for predicting the severity of dementia.
Hualou Liang, co-author of the study published in the journal PLOS Digital Health in December, explained, saying, “[T]his could be a simple, accessible and adequately sensitive tool for community-based testing [and] be very useful for early screening and risk assessment before a clinical diagnosis.”
His fellow author Felix Agbavor added: “GPT3’s systemic approach to language analysis and production makes it a promising candidate for identifying the subtle speech characteristics that may predict the onset of dementia.”
My Takeaway
GPT-3 is still relatively new technology and far from flawless. It was created in 2020 and was the culmination of extensive research by leading technology companies Microsoft, Google, and Facebook.
It’s still too early to tell how mainstream medicine will practically use technology like this to detect Alzheimer’s, but any tool that’s non-invasive and reliable for early disease detection is a win in my book.
