“The machines today are like toddlers learning how to walk. They stumble. They’re not very good. They make mistakes.” — Anmol Madan
One in three American adults are already using AI for health information — with or without their doctors’ permission. The machines, in other words, are already seeing us. This will chill some Americans in our summer of luddite discontent. For others, however, like the San Francisco-based medical tech entrepreneur Anmol Madan, the appearance of doctor AI in our lives is mostly good news.
In his new book, The Machines Will See You Now, Madan lays out what he calls a “human roadmap” to “autonomous health care.” I’m not entirely sure what he means by “human” in a Blade Runner-style world where machine and man are quickly merging. But by “autonomous,” which I suspect is a euphemism, Madan means artificial intelligence.
What worries Madan about our current moment is the broken dialogue over AI. On the one hand, we have tech utopians promising a totally automated healthcare industry. Then we have an anxious establishment struggling to change America’s archaic and often dysfunctional medical system. And, of course, American consumers (if that’s the right word) who are already using ChatGPT or, more troublingly, TikTok, as their doctor.
America spends $4.7 trillion on a healthcare industry which compares poorly with the medical systems of other wealthy countries. Madan’s “parallel universe,” he promises, would cost half as much and double access. The machines will see you now. Like it or not, Dr AI is our all-too-human future.
Five Takeaways
• One in Three Already Ask the Machines. The KFF poll behind the episode: a third of American adults already use AI for health information, with or without their doctors’ blessing — and if they’re not asking ChatGPT, they’re asking TikTok. Madan’s frustration is the broken dialogue between AI exuberance (“we’re going to automate healthcare”) and medical anxiety, when the right answer is in between: new technology is no excuse for throwing out the guardrails of medical devices and clinical safety. The stakes are the brutal arithmetic of American medicine — $4.7 trillion spent for the worst outcomes among peer nations — and his “parallel universe”: half the cost, twice the access, delivered by AI systems that are tested, safe, and rigorous. Consumer appetite is already there; the safeguards are not.
• The Waymo Scale for Medicine. The book’s organizing framework applies the self-driving industry’s levels of autonomy to healthcare. Level zero is today: 99 percent of decisions — diagnosis, treatment, prescription — made by humans. Level one is doctor assistance, already deployed: radiologists who once hauled backpacks of reports now guided by machines to where to focus. Levels two and three ease machines into the lowest-stakes treatment and diagnostic decisions with human review — and the fully autonomous end state, Madan concedes, may never arrive. To Andrew’s objection that Waymo’s scale ended with the drivers removed, Madan offers the transatlantic pilot: autopilot flies ninety percent of the route, and nobody thinks the pilot doesn’t matter. The structure, he argues, is what lets regulators, doctors, and builders finally talk about the same thing.
• Toddlers Learning to Walk. Madan’s metaphor for today’s medical AI: toddlers — stumbling, error-prone, needing layers of protection, and badly underestimated at their peril and ours. The mistake, he argues, is concluding that because the systems aren’t perfect we shouldn’t try them at all, when a constrained system is starving for capacity. Pressed by Andrew on what the machines will never do (“toddlers grow up”), his list is candid: human perception — reading body language, assessing the person who walks into the clinic — improves far more slowly than diagnostic reasoning and may never reach human capacity; and the relationships people form with chatbots, loneliness epidemic notwithstanding, have no literature yet showing they produce clinical outcomes like a human therapist’s thirty minutes. Plus one asterisk on the machines’ famous exam results: we grade them on human benchmarks, and machines fail differently than humans do.
• The Behavior-Change Trillion. Of America’s $4.7 trillion, a full trillion is the behavior-change problem — the eating, drinking, smoking, and sitting that no pamphlet has ever fixed. Andrew’s objection: nobody needs an MIT degree to know they should exercise, so why would a machine’s nagging beat a doctor’s? Madan’s answer is the space between visits: patients leave the office motivated and fall off within days, and personalized systems — like those he built for tens of millions at his previous companies — learn what actually moves each individual: this one walks more but won’t change diet, that one needs the stress addressed first, another needs the language and cultural register matched. Not one-size-fits-all “eat healthy,” but friction removed person by person, programmatically, at a scale no human workforce could staff. Andrew’s rejoinder: in an age of ubiquitous AI slop, the exercise reminder may be deleted as exactly that.
• Priceless. On trust, Andrew invoked the neighbors: Slippery Sam Altman, the mistrusted AI giants, and entrepreneurs — RadiantGraph included — getting rich while asking for our bloodwork. Madan’s counsel is unexpected caution: in the AI era all data matters, healthcare data most of all, and it belongs on healthcare-specific, HIPAA-bound platforms — not pasted into general chatbots. He concedes San Francisco’s mansions-and-homelessness inequity (he advises New York City’s Department of Public Health and wants to help at home), but defends the Bay Area as the place where the PC, the iPhone, the web, and now AI actually happened — Jeff Dean announced his next act the morning they recorded. And the bot test: how would he prove he’s human? “I’m quite flawed, and I’m often wrong. An AI avatar of Anmol probably wouldn’t be as wrong as I sometimes am.” Fittingly priceless — which is what Anmol means in Sanskrit.
About the Guest
Anmol Madan is a healthcare entrepreneur and computer scientist, and the founder and CEO of RadiantGraph, an AI platform helping health plans and healthcare organizations deliver proactive care. An MIT Media Lab PhD who worked on the first generation of wearables, he founded Ginger, the pioneering AI tele-psychiatry company later acquired by Headspace, and served as Chief Data Scientist at Livongo and as a data and AI executive at Teladoc Health. He advises the New York City Department of Public Health and lives in San Francisco. The Machines Will See You Now: A Human Roadmap to Autonomous Health Care (Johns Hopkins University Press, September 1, 2026) is his first book.
References:
• The Machines Will See You Now: A Human Roadmap to Autonomous Health Care by Anmol Madan (Johns Hopkins University Press, September 1, 2026). Alex Pentland, MIT: “A critical read about AI in he...
