Dear Margaret,
You make important points:
==
"This study of the use of ChatGPT for snakebite <https://pmc.ncbi.nlm.nih.gov/articles/PMC10339276/> states it "effectively addresses widespread myths and misconceptions associated with snakebites, helping to reduce misinformation. It offers evidence-based
guidance based on current guidelines and practices, ensuring the advice is reliable and accurate [7,9,13]. Additionally, the response emphasizes the importance of seeking immediate medical attention and following the guidance of healthcare professionals, which is crucial for the proper evaluation, treatment, and management of snakebites."
However, it also states, "there are some limitations to the response, such as limited depth on toxicology, as it does not delve into the specific toxicological aspects of venomous snakebites, like the type of venom, how it affects the body or the administration of antivenom. Furthermore, *the reply does not address the variation in venom toxicity and severity of symptoms between different snake species, which could be important information depending on the region or type of snake involved*." [emphasis added]
For AI to be useful to patients without immediate access to medical care who are dealing with snake bite, it would be essential to have information specific to the region and type of snake, so the individual knows whether the person who was bitten is in mortal danger or whether simple supportive care is sufficient. Such capability apparently has been developed <https://www.sciencedirect.com/science/article/abs/pii/S1386505623000412> (study is behind a paywall) but was not part of ChatGPT, for example...
Google AI has lots of documented problems discussed here <https://www.cbsnews.com/amp/news/software-developers-want-ai-to-give-med...
and here <https://www.cbsnews.com/news/google-ai-overview/>, for example.
==
With regard to the need for locally relevant information for snakebite that is specific to the type of snake, I think we shall inevitably see the evolution of AI responses that take into account the location of the user. If the type of snake is entered into the question, AI would be expected to provide specific advice on that species. It might also be able to help identify the species. In addition, it is quite possible that responses could be tailored for other characteristics, such as health literacy and of course language.
One item caught my eye in your reference https://www.cbsnews.com/amp/news/software-developers-want-ai-to-give-med...
'The World Health Organization is one of the companies dipping its toes into the AI waters. The organization's chatbot, Sarah, pulls information from the WHO's site and its trusted partners, making the answers less prone to factual errors. When asked how to limit the risk of a heart attack, Sarah gave information about managing stress, sleeping well and focusing on a healthy lifestyle.'
If that is all that 'Sarah' had do say on the issue, then this is an example of a gross failure to highlight key points such as quitting smoking. This would be such a large omission that I wonder if we have been misled by the journalist's anecdote.
I have tried WHO's 'Sarah' myself just now: She says: "To prevent a heart attack, focus on maintaining a healthy lifestyle. This includes eating a balanced diet rich in fruits, vegetables, whole grains, and lean proteins. Regular physical activity, managing stress, avoiding tobacco, and limiting alcohol intake are also key..."
HIFA profile: Neil Pakenham-Walsh is coordinator of HIFA (Healthcare Information For All), a global health community that brings all stakeholders together around the shared goal of universal access to reliable healthcare information. HIFA has 20,000 members in 180 countries, interacting in four languages and representing all parts of the global evidence ecosystem. HIFA is administered by Global Healthcare Information Network, a UK-based nonprofit in official relations with the World Health Organization. Email: neil@hifa.org