Can you trust AI for health information?

The internet is an amazing tool. You can video call your friends and family from across the globe. You can buy something and have it delivered to your door within a day. You can even access everything we know about the world in just a few clicks.

Increasingly, we turn to the internet for help and advice on all sorts of topics, including our health.

But with the rise of artificial intelligence (AI) models, like ChatGPT, it is becoming more difficult to tell what information you can trust on the internet.

AI summaries

Nowadays, when you search the internet with tools like Google, the first search result is often an AI summary. These AI summaries may be useful, as they explain the answer to your question in a detailed but conversational tone.

However, these summaries are often imperfect.

You should always check the information they tell you. This is because these types of AI models are capable of ‘hallucinating’, meaning they can create answers which are factually incorrect.

Some humorous examples of ‘hallucinations’ are claims that you should eat at least one small rock a day, or that Leo Varadkar, an Irish politician, is in fact more than 2,716 feet tall.

These hallucinations are not just funny, but obviously incorrect. This is not always the case, though.

Don’t mistake confidence for competence

Often, hallucinations are harder to spot. AI summaries are often ‘confidently wrong’, so their advice can still sound correct even when it is not.

Sometimes, rather than giving completely made-up answers, an AI summary will instead give you an outdated answer which was correct at some point.

In addition, AI summaries often have a ‘US bias’ due to how they are created. For example, when we searched “new glaucoma treatment”, Google described the Glaukos iDose® TR. This is a medical implant which releases travoprost over a few years but is only currently available in the US.

Check your sources carefully

Checking that the information is correct, up to date and not US biased is very important for healthcare questions.

The wrong advice could cause potentially severe health issues. Always use trusted and reliable sources of healthcare information, such as the NHS or Glaucoma UK websites, for advice to avoid being tricked by a well-meaning but imperfect AI model.

This article was originally published in our membership magazine, Insight. To receive the magazine quarterly, consider becoming a Glaucoma UK member.