Miriam González, an engineer from Murcia, Spain, has a rare breast tumor. Her experience illustrates the complexity of using this tool effectively — one that tends to fail on basic medical questions.
In 2021, Miriam González, then 35, went to her doctor because of bleeding from her breast. She was reassured that everything was normal. But in 2024 she was diagnosed with breast cancer — and shortly afterward discovered it was metastatic, stage four. ‘At first I thought the diagnosis was an immediate death sentence, that I had days or weeks left,’ González explained in a message exchange. But that was not the case: ‘I started hearing words like chronicity and quality of life, and I saw that the picture today is different. That mental transition — from ‘I am dying now’ to ‘I am going to live with this’ — was hard, and I needed to understand the terrain I was moving in.’
To understand that terrain, she turned to Perplexity, an AI-powered search engine. Her engineering background then took over: ‘breaking down the problem.’ Her tumor had neuroendocrine differentiation — a subtype so uncommon that standard clinical guidelines simply do not address it. AI helped her understand and ‘organize that complexity and turn an abstract diagnosis into concrete decisions.’
Millions of people now use AI as a jargon translator, medication consultant, or even a substitute doctor. However, Mark Succi, Director of Healthcare Innovation at Mass General Brigham and Associate Professor at Harvard, urges caution: ‘AI seems more useful in the later, more narrowly defined stages of diagnosis — narrowing in on an answer once the case is already structured — and less useful in generating an initial diagnostic framework that is conscious of uncertainty.’
A recently published study analyzing five of the most popular models found that half of the health information provided lacked scientific rigor, a level of inaccuracy that puts patient safety at risk. Nevertheless, a new US survey reveals that one in four Americans uses chatbots for health questions — primarily for quick answers or supplementary information. A significant group, however, uses it instead of a doctor, particularly those with low incomes who cannot afford a medical visit.
González’s case was different. She enlisted the help of Javi López, an AI specialist and co-founder of Magnific, who used the most advanced systems available — ChatGPT Pro+ Extended and Claude Opus 4.6 MAX. He converted her entire medical history into a text document, then created a nearly 2,000-word AI-generated prompt designating the model as a ‘multidisciplinary tumor committee composed of the world’s leading specialists.’ He then passed the response to a second model to check for flaws. ‘This kind of adversarial model has always worked. It’s like having two research teams working in parallel and sharing their findings.’
Oncologist Oriol Mirallas of MD Anderson’s Phase 1 Experimental Therapies Unit acknowledges this is inevitable but delicate: ‘We increasingly see people who come with ChatGPT or clinicaltrials.gov printouts. AI can help — but with expert guidance. If it helps the patient understand their pathology and diagnosis, that’s great. But searching for feasible and optimal treatments, in a field that changes daily, is complicated.’
Researcher Arya Rao of Harvard University is optimistic: ‘It’s exciting that we have more tools to empower and educate patients. Instead of discouraging patients from using these tools, doctors should open the conversation: ask what they’ve searched for, what AI told them, and what questions they have.’
Mark Succi offers a key takeaway: ‘These systems can sound confident even when their reasoning is weak, especially in complex cases. That is why doctors should help patients use AI results as a starting point, not a diagnostic conclusion.’
FUENTE: EL PAIS.COM
