Exploiting ChatGPT for Diagnosing Autism-Associated Language Disorders and Identifying Distinct Features

Area of Science:

• Neuroscience
• Artificial Intelligence
• Linguistics

Background:

• Diagnosing autism-associated language disorders is challenging due to subjective and slow traditional methods.
• Current assessments require significant human effort, leading to delayed interventions.
• There is a need for faster, more specific, and accurate diagnostic tools.

Purpose of the Study:

• To evaluate ChatGPT's efficacy in diagnosing autism-associated language disorders.
• To compare ChatGPT's performance against conventional supervised learning models like BERT.
• To identify specific linguistic features indicative of autism.

Main Methods:

• Utilized ChatGPT, a large language model, for analyzing linguistic features.
• Compared ChatGPT's zero-shot learning performance with BERT.
• Identified and analyzed ten distinct features of autism-associated language disorders.

Main Results:

• ChatGPT demonstrated a significant improvement, exceeding conventional models by over 13% in accuracy and F1-score.
• ChatGPT outperformed BERT in a zero-shot learning configuration for diagnosing language disorders.
• Ten key linguistic features, including echolalia and pronoun reversal, were identified as crucial for diagnosis.

Conclusions:

• ChatGPT shows substantial potential as a superior tool for neurological diagnostics.
• The findings support the adoption of advanced AI like ChatGPT in clinical settings for developmental disorder assessment.
• This AI-driven approach enhances diagnostic precision and supports personalized medicine for autism evaluation.