Google is putting people at risk of harm by downplaying safety warnings that its AI-generated medical advice may be wrong.<\/p>\n
When answering queries about sensitive topics such as health, the company says its AI Overviews, which appear above search results, prompt users to seek professional help, rather than relying solely on its summaries. \u201cAI Overviews will inform people when it\u2019s important to seek out expert advice or to verify the information presented,\u201d Google has said<\/a>.<\/p>\n But the Guardian found the company does not include any such disclaimers when users are first presented with medical advice.<\/p>\n Google only issues a warning if users choose to request additional health information and click on a button called \u201cShow more\u201d. Even then, safety labels only appear below all of the extra medical advice assembled using generative AI, and in a smaller, lighter font.<\/p>\n \u201cThis is for informational purposes only,\u201d the disclaimer tells users who click through for further details after seeing the initial summary, and navigate their way to the very end of the AI Overview. \u201cFor medical advice or a diagnosis, consult a professional. AI responses may include mistakes.\u201d<\/p>\n Google did not deny its disclaimers fail to appear when users are first served medical advice, or that they appear below AI Overviews and in a smaller, lighter font. AI Overviews \u201cencourage people to seek professional medical advice\u201d, and frequently mention seeking medical attention within the summary itself \u201cwhen appropriate\u201d, a spokesperson said.<\/p>\n AI experts and patient advocates presented with the Guardian\u2019s findings said they were concerned. Disclaimers serve a vital purpose, they said, and should appear prominently when users are first provided with medical advice.<\/p>\n \u201cThe absence of disclaimers when users are initially served medical information creates several critical dangers,\u201d said Pat Pataranutaporn, an assistant professor, technologist and researcher at the Massachusetts Institute of Technology (MIT) and a world-renowned expert in AI and human-computer interaction.<\/p>\n \u201cFirst, even the most advanced AI models today still hallucinate misinformation or exhibit sycophantic behaviour, prioritising user satisfaction over accuracy. In healthcare contexts, this can be genuinely dangerous.<\/p>\n \u201cSecond, the issue isn\u2019t just about AI limitations \u2013 it\u2019s about the human side of the equation. Users may not provide all necessary context or may ask the wrong questions by misobserving their symptoms.<\/p>\n \u201cDisclaimers serve as a crucial intervention point. They disrupt this automatic trust and prompt users to engage more critically with the information they receive.\u201d<\/p>\n Gina Neff, a professor of responsible AI at Queen Mary University of London, said the \u201cproblem with bad AI Overviews is by design\u201d and Google was to blame. \u201cAI Overviews are designed for speed, not accuracy, and that leads to mistakes in health information, which can be dangerous.\u201d<\/p>\n In January, a Guardian investigation<\/a> revealed people were being put at risk of harm by false and misleading health information in Google AI Overviews.<\/p>\n Neff said the investigation\u2019s findings<\/a> showed why prominent disclaimers were essential. \u201cGoogle makes people click through before they find any disclaimer,\u201d she said. \u201cPeople reading quickly may think the information they get from AI Overviews is better than what it is, but we know it can make serious mistakes.\u201d<\/p>\n