Due to its remarkable capacity to produce textual comments that resemble those of real people in response to some of the most original questions, ChatGPT has already gained notoriety as a chatbot. It has now demonstrated another another amazing capability: in the future, the AI algorithms powering it will help doctors identify the early stages of Alzheimer’s disease!
A recent study from Drexel University’s School of Biomedical Engineering Science and Health Systems showed that the GPT-3 programme of OpenAI can recognise cues from spontaneous speech and is 80% accurate in detecting the early stages of dementia.
Although there is currently no cure for Alzheimer’s disease, the standard diagnosis entails a number of extensive physical and neurological tests as well as a study of a patient’s medical history. Early detection can therefore help patients get ready for treatment and support. Researchers are concentrating on programmes that can detect subtle clues like hesitation in speaking, grammar, and pronunciation errors through simple tests that can determine whether the person needs to be fully examined or not. Language impairment is a noticeable symptom in almost 80% of dementia patients.
Since GPT-3 excels at “zero-data learning,” it can even respond to inquiries that call for outside knowledge that is not supplied. Additionally, GPT-3 has received sufficient training to evaluate and modify itself in order to produce the required information.
By running a training programme with transcripts from a speech recording dataset to evaluate an NLP algorithm’s capacity to predict dementia, Drexel researchers were able to confirm their theory. In the trial, two of the best NLP tools were used, and it was discovered that GPT-3 outperformed the others in accurately distinguishing between examples with and without Alzheimer’s disease.
The Mini-Mental State Exam, a typical test to gauge the severity of dementia, was used as the second test (MMSE). Here, the textual analysis of the GPT-3 was utilised to forecast the scores of a large number of patients in the sample. GPT-3 is almost 20% accurate in predicting the MMSE score, as shown by a comparison between the standard analysis and the prediction. The researchers came to the conclusion that the text embedding produced by GPT-3 might be successfully applied to distinguish Alzheimer’s patients from healthy controls. The article recommends text embedding with GPT-3 as a successful strategy for the early identification of dementia.
