Google has removed some of its artificial intelligence health summaries after a Guardian investigation found that false and misleading information was putting people at risk of harm.
The company has said that its AI overviews, which use generative AI to provide snapshots of essential information about a topic or question, are “helpful” And “reliable,
But some summaries, which appear at the top of search results, provide inaccurate health information, putting users at risk of harm.
In a case that experts described as “alarming” and “alarming”, Google provided fake information about vital liver function tests, which could mislead people with severe liver disease into thinking they were healthy.
The Guardian found that typing “what is the normal range for a liver blood test” produced large numbers of numbers, had little context and did not account for patients’ nationality, gender, ethnicity or age.
What Google’s AI overview said was normal may actually differ significantly from what is considered normal, experts said. Summary may cause seriously ill patients to mistakenly think that their test results are normal, and they do not bother to attend follow-up health care meetings.
After investigation, the company has removed AI observations for the search terms “what is the normal range for a liver blood test” and “what is the normal range for a liver function test.”
A Google spokesperson said: “We do not comment on individual removals within Search. In cases where some context is missed in an AI overview, we work to make comprehensive improvements, and we also take action under our policies where appropriate.”
Vanessa Hebditch, director of communications and policy at the British Liver Trust, a liver health charity, said: “This is excellent news, and we are pleased to remove Google AI observations in these cases.
“However, if the question is asked differently, potentially misleading AI observations may still be given and we are concerned that other health information produced by AI may be inaccurate and confusing.”
The Guardian found that typing slight variations of the original queries into Google, such as “LFT reference range” or “LFT test reference range”, prompted an AI overview. That was a big concern, Hebditch said.
“Liver function tests or LFTs are a collection of different blood tests. Understanding the results and what to do next is complex and involves much more than comparing a set of numbers.
“But AI Overview presents a list of tests in bold, making it very easy for readers to forget that these numbers may not even be true for their test.
“Furthermore, the AI observation fails to warn that someone may get normal results on these tests when they have severe liver disease and need further medical care. This false reassurance could be very harmful.”
Google, Which has 91% share of the global search engine marketsaid it was reviewing the new examples provided by the Guardian.
Hebditch said: “Our big concern with all of this is that it’s messing up a single search result and Google can turn off AI observations for this but it’s not dealing with the bigger issue of AI observations for health.”
Sue Farrington, president of the Patient Information Forum, which promotes evidence-based health information to patients, the public and health care professionals, welcomed the removal of the summaries but said she still had concerns.
“This is a good result but it is a necessary first step to maintain trust in Google’s health-related search results. There are still too many examples of Google AI observations giving people inaccurate health information.”
Farrington said millions of adults around the world already struggle to access reliable health information. “That’s why it’s so important that Google informs people about the robust, researched health information and care they offer from trusted health organizations.”
AI observations still pop up for other examples that the Guardian originally highlighted on Google. They include a summary of information about cancer and mental health that experts have described as “completely inaccurate” and “really dangerous”.
When asked why these AI observations were not removed, Google said they are linked to well-known and reputable sources, and inform people when it is important to seek expert advice.
A spokesperson said: “Our internal team of practitioners reviewed what was shared with us and found that in many cases, the information was not inaccurate and was also supported by high quality websites.”
Victor Tangerman, senior editor at technology website Futurism, said the results of the Guardian’s investigation showed that Google had work to do.To make sure its AI tools aren’t spreading dangerous health misinformation,
quick guide
Contact Andrew Gregory about this story
show
If you have something to share about this story, you can contact Andrew using one of the following methods.
The Guardian app has a tool for sending suggestions about stories. Messages are end-to-end encrypted and hidden in the routine activity performed by each Guardian mobile app. This prevents the observer from knowing that you are communicating with us at all, let alone what is being said.
If you don’t already have the Guardian app, download it (iOS,Android) and go to menu. Select ‘Secure Messaging’.
Email (not secure)
If you do not require a higher level of security or privacy you can email andrew.gregory@theguardian.com
SecureDrop and other secure methods
If you can safely use the Tor network without being observed or monitored you can send messages and documents to the Guardian through our SecureDrop platform.
Finally, our guide on theguardian.com/tips lists several ways to contact us securely, and discusses the advantages and disadvantages of each.
Google said the AI overview only appears on queries where it has full confidence in the quality of the responses. It says the company continuously measures and reviews the quality of its summaries across many different categories of information.
in an article for search engine journalSenior author Matt Southern said: “AI observations appear above ranked results. When the topic is health, errors are more significant.”
