Physicians noticed something unusual in the late 2000s: Patients were coming to appointments armed with sometimes-dubious medical information they had gleaned online from \u201cDr. Google,\u201d according to Adam Rodman, an internist and AI researcher.<\/p>\n
Today, about 68 percent of adults<\/a> have turned to a search engine for medical advice in the past. But Dr. Google has a competitor. About 32 percent of adults<\/a>, approximately half of those who sought advice online, turned to AI chatbots for help.\u00a0<\/p>\n Rodman thinks such resources, used appropriately, are an overall net good. In op-eds<\/a> and online courses<\/a>, Rodman<\/a>, a Harvard Medical School assistant professor of medicine at Beth Israel Deaconess Medical Center, has shared advice for how to best employ Dr. Chat.\u00a0<\/p>\n In this interview, edited for length and clarity, Rodman offers a stoplight system to figure out when it\u2019s safe to ask a chatbot, and when you should really just ask your doctor.<\/p>\n How were doctors thinking about online medical information before the age of AI?\u00a0<\/p>\n The early literature refers to this as the internet-informed patient. In the early 2000s, doctors noticed people would come into their appointments with articles they found online, but it was still only among really tech-savvy people. It certainly wasn\u2019t a normal interaction.\u00a0<\/p>\n Then in the late 2000s, search engines started to take advantage of neural network technology, and they were able to serve up more relevant health information. They figure out what you\u2019re going to want to read next, and they give it to you.<\/p>\n That\u2019s when we first got the phrase \u201cDr. Google,\u201d often used as a pejorative, from doctors who saw patients coming in with a level of confidence that may or may not have been earned.<\/p>\n Of course, there are patients who know a lot about their health and are very well informed, but we also saw a lot of patients misinformed.\u00a0<\/p>\n That\u2019s where we get this concept of cyberchondria. It\u2019s related to hypochondria: this idea that search engines can drive people to more and more extreme places until you go from googling your headache to reading about glioblastoma multiforme \u2014 and research has shown<\/a> that it\u2019s a real phenomenon.\u00a0<\/p>\n \u201cBoth Google and AI companies are now very aware that people are using their tools for health information and are trying to build in safety mechanisms.\u201d<\/p>\n We all have understandable and reasonable anxieties about our health. Seeking out information is something fundamental about humanity.<\/p>\n The problem is when that starts to interact with these recommendation algorithms that are optimized for engagement, and for showing you what you want to see even if it\u2019s incorrect.\u00a0<\/p>\n Now let\u2019s bring AI into the mix. Is it any different to ask a chatbot about symptoms versus googling them?\u00a0<\/p>\n It\u2019s nuanced. In one sense, LLMs do exactly what Google does: They serve you up the things you unconsciously want to hear, even if those things make you anxious.\u00a0<\/p>\n On the other hand, unlike with a Google search, some people feel they have a relationship with an LLM. LLMs speak with extreme authority and confidence no matter what they say. It\u2019s under-explored the extent to which that could make cyberchondria worse.\u00a0<\/p>\n Both Google and AI companies are now very aware that people are using their tools for health information and are trying to build in safety mechanisms. The bots will tell you to go to the emergency room or call your doctor, those sorts of things.<\/p>\n But at least theoretically, language models are much, much better than Google, especially the more modern reasoning models, when it comes to identifying medical conditions.<\/p>\n What do you mean by \u201ctheoretically\u201d?<\/p>\n