Paper published: Can ChatGPT detect obese dogs?

Published on

Image of tablet being used with abstract computer background

Obesity is a major health and welfare concern in dogs and cats. Relevant clinical information is often recorded in large volumes of unstructured text that are part of a patient’s health record. Recently, large language models (LLMs) such as GPT3.5 (which underpins ChatGPT) have become available providing an exciting opportunity for automated data extraction.

Here we compared the performance of a validated rule-base system using regular expressions (RegexT) to that of a prompt-based approach using ChatGPT to identify a dogs recorded body condition score (BCS).

ChatGPT efficiently identified most overweight animals, with or without a BCS. However, occasional false positives were identified; this behaviour might be avoidable through more subtle prompt engineering.

The full, open access preprint paper (not peer reviewed) is available here.

Our research infographic:

Quote from Dr PJ Noble, AI lead on veterinary data at University of Liverpool.

"AI is becoming the key instrument for data mining and analysis in all walks of life. Here at Liverpool we are designing AI tools to explore veterinary data. These studies are essential to ensure that the veterinary profession learns to understand and leverage these technologies wisely in its aims to improve animal welfare".

This work has been kindly supported by a Canine Welfare Grant from DogsTrust.