According to researchers, the artificial intelligence that powers the chatbot program ChatGPT—famous for its capacity to produce human-like responses when instructed—could aid in the early detection of Alzheimer’s.
Recent research from Drexel University’s School of Biomedical Engineering, Science, and Health Systems has revealed that the GPT-3 program from OpenAI can recognize cues from spontaneous speech that are 80% accurate in identifying the early stages of dementia.
The Drexel study is the most recent in a line of efforts to demonstrate the efficacy of natural language processing programs for Alzheimer’s early detection by drawing on recent findings that language impairment may serve as an early sign of neurodegenerative diseases.
Detecting an Early Sign
Doctors usually perform a thorough assessment of medical history and a battery of physical and neurological examinations and tests as part of the standard procedure for diagnosing Alzheimer’s today. The illness still has no known cure, but early discovery can give patients more therapeutic and support options.
Language deterioration is a symptom in 60-80% of people with dementia. Therefore, researchers have focused on programs that can detect subtle clues, such as hesitation, grammar and pronunciation errors, and forgetting the meaning of words, as a quick test that could imply whether or not a patient should undertake a complete examination.
Hualou Liang, Ph.D., a co-author of the research, stated that in addition to cognitive tests, the most commonly used tests for early detection of Alzheimer’s look at acoustic features such as pausing, articulation, and vocal quality. The researchers believe that advancements in natural language processing programs may provide another avenue for supporting early Alzheimer’s detection.
GPT-3: A Program that Listens and Learns
The third generation of OpenAI’s General Pretrained Transformer (GPT), GPT-3, employs a deep learning algorithm trained by analyzing extensive amounts of internet data with an emphasis on word usage and linguistic construction. Through this training, it can produce a human-like response to any language-related task, including answering straightforward questions and crafting poetry or essays.
GPT-3 excels in “zero-data learning,” the ability to respond to queries without any need for external knowledge that is usually required. For instance, asking the program to write “Cliff’s Notes” on a text would necessitate an explanation that this means a summary. However, GPT-3 has received enough training to understand the reference and adjust itself to generate the expected response.
According to Felix Agbavor, the lead author of the study, because of its systematic approach to language analysis and production, GPT3 is a good contender for figuring out the minute speech cues that could indicate the beginning of dementia. Training GPT-3 with a colossal dataset of interviews, some of which are with Alzheimer’s patients, would give it the information it needs to extract speech patterns, which could help identify markers in future patients.
Looking for Speech Cues
The researchers tested their theoretical hypothesis by feeding the algorithm a collection of transcripts from a sample of a dataset of voice recordings compiled to evaluate natural language processing programs’ capacity to predict dementia. The computer tool extracted significant word usage, sentence construction, and meaning traits from the text to create what academics refer to as an “embedding”—a distinctive profile of Alzheimer’s speech.
They then retrained the software using the embedding, converting it into a device for diagnosing Alzheimer’s. To test it, they instructed the program to examine dozens of transcripts from the dataset and determine whether or not each one was written by an individual who was developing Alzheimer’s.
The team tested two of the best natural language processing tools side by side and discovered that GPT-3 outperformed both in terms of accurately identifying Alzheimer’s examples, identifying Alzheimer’s non-examples, and with rarer missed cases than both programs.
A second test used textual analysis from the GPT-3 to predict patients’ results on the Mini-Mental State Exam, a standard test for assessing the degree of dementia (MMSE).
The research team then compared the accuracy of the GPT-3 forecast to that of an analysis that predicted the MMSE score only based on the acoustic characteristics of the recordings, such as voice strength, pauses, and slurring. GPT-3 was nearly 20% more accurate in predicting MMSE scores in patients.
What do the results imply?
According to the researchers, the results showed that the text embedding produced by GPT-3 could be consistently utilized to distinguish between people who have Alzheimer’s and healthy controls and to infer the subject’s cognitive assessment score, both exclusively based on speech data.
They also demonstrated that text embedding performs better than the traditional acoustic feature-based method and even competes with tuned models. These findings collectively imply that GPT-3 based text embedding is a promising method for assessing Alzheimer’s and has the potential to enhance early diagnosis of dementia.
The researchers intend to build on these encouraging findings by creating a web application that people might use as a pre-screening tool at home or a doctor’s office.
- Agbavor, F. and Liang, H., 2022. Predicting dementia from spontaneous speech using large language models. PLOS Digital Health, 1(12), p.e0000168. https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000168.
- Can the AI Driving ChatGPT Help to Detect Early Signs of Alzheimer’s Disease?. Science Daily. https://www.sciencedaily.com/releases/2022/12/221222162415.htm. Accessed: 6th Jan, 2023.
- Can the AI Driving ChatGPT Help to Detect Early Signs of Alzheimer’s Disease?. Neuroscience.com. https://neurosciencenews.com/chatgpt-dementia-ai-22133/. Accessed: 6th Jan, 23.