Good evidence confuses ChatGPT when used for health information 

ChatGPT for health information
Photo: estradaanton 123rf

A world-first study has found that when asked a health-related question, the more evidence given to ChatGPT the less reliable it becomes—reducing the accuracy of its responses to as low as 28 per cent. 

As large language models (LLMs) like ChatGPT explode in popularity, they pose a potential risk to the growing number of people using online tools for key health information. 

Scientists from CSIRO, Australia’s national science agency, and The University of Queensland (UQ) explored a hypothetical scenario of an average person (non-professional health consumer) asking ChatGPT if ‘X’ treatment has a positive effect on condition ‘Y’.  

The 100 questions presented ranged from, ‘Can zinc help treat the common cold?’ to ‘Will drinking vinegar dissolve a stuck fish bone?’ ChatGPT’s response was compared to the known correct response, or ‘ground truth’, based on existing medical knowledge.  

CSIRO principal research scientist and associate professor at UQ, Dr Bevan Koopman said even though the risks of searching for health information online are well documented, people continue to seek health information online, and increasingly via tools such as ChatGPT.  

“The widespread popularity of using LLMs online for answers on people’s health is why we need continued research to inform the public about risks and to help them optimise the accuracy of their answers,” Dr Koopman said. 

“While LLMs have the potential to greatly improve the way people access information, we need more research to understand where they are effective and where they are not.” 

The study—recently presented at Empirical Methods in Natural Language Processing, a premier Natural Language Processing conference in the field—looked at two question formats. The first was a question only. The second was a question biased with supporting or contrary evidence.  

Results revealed that ChatGPT was quite good at giving accurate answers in a question-only format, with an 80 per cent accuracy in this scenario.  

However, when the language model was given an evidence-biased prompt, accuracy reduces to 63 per cent. Accuracy was reduced again to 28 per cent when an ‘unsure’ answer was allowed. This finding is contrary to popular belief that prompting with evidence improves accuracy. 

“We’re not sure why this happens. But given this occurs whether the evidence given is correct or not, perhaps the evidence adds too much noise, thus lowering accuracy,” Dr Koopman said. 

Previous articleMeet special needs dentist Dr Lydia See
Next articleResearchers determine oral bacteria accelerate pancreatic cancer development in mice


Please enter your comment!
Please enter your name here