DocBots: Will AI Reinvent Healthcare?

AI could make medical care more accurate, personalized, efficient and accessible. In some areas it already rivals or surpasses human experts. The implications reach beyond how we diagnose and treat disease to the relationship between patients, physicians and the healthcare system itself.

It may soon seem close to malpractice for a physician not to consult AI on a difficult diagnosis. At the very least, AI “co-bots” could prevent needless deaths from medical error and cut waste in a notoriously expensive, fragmented industry.

But medicine is not just pattern recognition. It is a human endeavor built on trust, empathy, judgment and hard ethical choices. As AI grows more capable, we must ask not only what it does better than we do, but what we should keep for ourselves.

DocBots salon dinner in San Francisco during the J.P. Morgan Healthcare Conference.

The Rise of the DocBot

Medicine has always depended on a physician’s ability to absorb information, recognize patterns and make judgments under uncertainty. AI is well suited to much of this.

Machine learning can find subtle patterns in vast amounts of medical data. It already helps detect tumors and fractures in radiology and retinal disease in ophthalmology, and similar methods may catch early signs of cancer, neurological disorders and heart disease.

The opportunity is to augment physicians, not replace them. A DocBot can act as a second opinion: raising overlooked possibilities, suggesting tests and flagging errors before they cause harm. It never tires, draws on the entire medical literature and weighs many diagnoses at once.

AI is not infallible, though. It can miss context or be confidently wrong, and in medicine those mistakes can be fatal. The best model is a partnership that pairs computational power with clinical experience, judgment and empathy. In time, we may expect doctors to use AI as routinely as any other diagnostic tool.

From Averages to Individuals

Most treatments are developed and tested across large populations. A drug that works for the average patient may fail, or cause harm, in someone whose biology differs.

By analyzing genetics, medical history and lifestyle, AI could predict which treatments will work for a particular patient, anticipate adverse reactions and optimize dosing. That matters most in diseases like cancer, where similar diagnoses can respond very differently to the same therapy.

AI could also make medicine proactive. Wearables and remote monitoring can track heart rhythm, blood glucose, sleep and more, flagging early warning signs before a patient becomes seriously ill. The goal is a system that anticipates and prevents disease, not just one that treats it.

Digital Twins and Drug Discovery

Drug development is slow, expensive and uncertain, and most candidates fail along the way. AI can help by uncovering hidden biological relationships and predicting which compounds are most promising.

One intriguing approach is the biological “digital twin”: a computational model of a biological system built from genetic, molecular and clinical data. Some use causal AI to model the mechanisms that drive disease rather than mere correlations. Aitia, for example, is building Gemini Digital Twins and has worked with Servier on pancreatic cancer, Parkinson’s disease and brain cancer. Other efforts aim to model individual cells and predict the effects of genetic changes.

These tools remain experimental, and a simulation is not proof that a treatment works in people. But they point toward drug discovery guided by models of biology rather than trial and error, which could open new paths for diseases that have resisted decades of research.

Expanding Access

Healthcare is also a logistical and economic problem. Clinicians spend enormous time on documentation, scheduling, billing and insurance approvals. Automating that work could free them for patients and lower costs.

The bigger opportunity is spreading medical expertise. Picture a community health worker in a rural area with a smartphone, basic equipment and an AI that can assess symptoms, read images and recommend next steps under clinical supervision. AI could also help patients manage chronic conditions at home and steer people toward providers that deliver better outcomes at lower cost.

That would expand the reach of expertise without a matching increase in specialists. The risk is that wealthy patients and institutions benefit first. Making sure AI narrows disparities rather than widening them must be a priority.

Risks and Overreliance

AI introduces real risks. It can give plausible but wrong answers, and its reasoning can be hard to explain or audit. It learns from data that reflect existing inequities, so it may perform worse for underrepresented groups. It depends on deeply sensitive data, from genetics to continuous monitoring, which demands strong privacy protections and patient control.

Regulation is hard too, because some AI systems keep changing after approval. And accountability is unclear: when an AI-assisted diagnosis is wrong, who is responsible, the physician, the hospital or the developer? These questions grow sharper as AI moves from advising to acting. We need governance that keeps humans meaningfully in charge of consequential decisions.

A subtler danger is automation bias. The more reliable a system seems, the less likely clinicians are to catch its rare mistakes, and over time their own diagnostic skills may erode. The answer is not to resist the technology but to design it for collaboration: a good DocBot shows its reasoning, offers alternatives and flags uncertainty, widening a physician’s thinking rather than narrowing it to one answer.

Keeping Medicine Human

Patients face fear, pain and decisions rooted in personal values. Physicians explain hard choices, offer reassurance and sometimes help patients face mortality.

AI may get very good at simulating empathy, and its patience and availability can help. But recognizing suffering is not the same as understanding it. Questions like how to weigh longevity against quality of life, or when to move from aggressive treatment to palliative care, are human questions.

If AI takes on more of the analytical and administrative load, physicians may have more time to listen and build relationships with patients. In that sense, AI could make medicine more human.

Toward Amplified Healthcare

AI’s impact on healthcare is likely to come in three stages: optimization, making today’s medicine more accurate and efficient; transformation, changing how care is delivered through personalized medicine, AI-assisted physicians and new drug discovery; and reinvention, reshaping the economics and structure of healthcare itself.

We are still early. But for centuries, healthcare has been limited by scarce human expertise, and AI could lift that constraint. The opportunity is not just to automate today’s system, but to make healthcare more proactive, personalized and accessible.

The promise of DocBots is not that machines will replace physicians, but that AI will amplify them. Success will be measured not by how smart our medical machines become, but by how much healthier and longer they help us live.

This article is based on a January 16, 2025 salon dinner, “DocBots: AI in Healthcare,” hosted by Matthew Stepka in San Francisco during the J.P. Morgan Healthcare Conference as part of Conspiracy of Love, a salon dinner series he founded with John Perry Barlow nearly a decade earlier. The gathering brought together entrepreneurs, scientists, physicians, investors and other thinkers to explore the promise and challenges of AI in healthcare. This article was co-authored by AI.