Before the doctor walks in, something else has already been looking at you.
It has scanned your imaging. It has flagged a pattern in your chest X-ray so subtle that human eyes, even excellent ones, might have passed over it. It has cross-referenced your electronic health records — every lab result, every medication, every visit stretching back years — and quietly updated its picture of your risk. By the time a physician pulls up your chart, an artificial intelligence system has already been thinking about your case.
This is not a vision of 2035. This is what is happening right now, inside some of New York City’s most celebrated hospitals. And it is changing medicine faster than most patients realize.
A New Kind of Second Opinion
At Mount Sinai, the shift is already measurable. By March 2025, the hospital had performed more than 100,000 AI-assisted mammograms. The system doesn’t replace the radiologist — it reads the image alongside them, highlighting areas that deserve a second look, surfacing the subtle shadows that might otherwise go unnoticed for another year.
“Artificial intelligence is a phenomenal tool,” said Laurie Margolies, MD, Chief of Breast Imaging at Mount Sinai’s Dubin Breast Center. “It does not replace the expertise of our radiologists — it enhances it. It gives us an added set of eyes.”
That framing — AI as an enhancer, not a replacement — has become the working philosophy across New York’s leading health systems. But the technology is advancing quickly, and the capabilities are expanding into territory that would have seemed implausible just a few years ago.
Researchers at the Icahn School of Medicine at Mount Sinai and Memorial Sloan Kettering conducted a first-of-its-kind “silent trial” in pathology, in which an AI analyzed live patient samples in real time without its results being visible to clinicians. The system was able to reliably detect EGFR mutations — a specific genetic marker critical for matching lung cancer patients to the right treatments — and showed the potential to reduce the need for rapid genetic tests by more than 40 percent. The implications are significant: faster results, lower costs, and a potential lifeline for patients who currently wait days for answers that could determine their treatment path.
Reading the Record Like No Human Can
One of the most underappreciated applications of AI in medicine isn’t imaging at all — it’s the electronic health record. A patient’s chart is, in theory, one of the richest sources of diagnostic information available. In practice, it’s often a chaotic accumulation of clinical notes, lab values, medication lists, and timestamps that no human has the time or bandwidth to synthesize completely.
Mount Sinai’s InfEHR system approaches this differently. Rather than applying a standardized diagnostic lens to every patient, it builds an individual network of medical events specific to that person — mapping how their health has evolved over time, what has changed, what connections exist between events that might not be obvious. Tested across health record datasets from both Mount Sinai and UC Irvine, the system brings what researchers describe as personalized diagnostics within reach.
The ambition here is not just efficiency. It’s catching things that would otherwise stay hidden — rare genetic conditions, brewing chronic diseases, subtle deteriorations — before they become crises.
The Five-Year Horizon
At NYU Langone, the conversation has moved beyond pilots and prototypes. Health informatics leaders there have said publicly that fully autonomous clinical AI is coming within the next five years — systems capable of managing routine tasks like blood pressure medication titration and diabetic retinopathy screening without requiring a human to sign off on every step.
That is a provocative claim in a field that has historically been cautious about automation. But NYU Langone frames it not as a convenience, but as a necessity — an answer to the workforce shortages and system inefficiencies that are already straining hospitals. When demand for care exceeds the supply of clinicians, autonomous AI that handles the routine frees human expertise for the complex.
The hospital has built infrastructure to match the ambition. Its Division of Applied AI Technologies vets commercial tools, runs a managed generative AI program for research and innovation, and maintains strict data governance — including explicit prohibitions on using patient health information in public AI models.
What the Machine Can’t Do
None of New York’s leading hospitals are rushing to remove the human from the equation entirely, and for good reason. A 2025 Mount Sinai study found that AI, like humans, can jump to conclusions — performing poorly in situations that require ethical sensitivity, nuanced judgment, or what the researchers called emotional intelligence.
“Our findings don’t suggest that AI has no place in medical practice,” said the study’s co-author, “but they do highlight the need for thoughtful human oversight.”
That tension — extraordinary capability sitting alongside real limitation — defines where New York medicine is right now. The algorithm in the room is getting smarter, faster. It is catching cancers earlier, matching patients to clinical trials they would never have found, and reading the hidden language buried in years of health records.
But it still needs the doctor to walk through the door. For now.