Karan Singhal, a researcher at OpenAI, is working on a way to shrink the 17-year average lag between new medical evidence and updated clinical guidelines down to hours. The effort, which Singhal has described publicly, aims to use AI to accelerate the guideline-writing process that currently takes years to complete. Faster updates could mean patients receive care based on the latest evidence far sooner than they do today.
The 17-year gap between evidence and practice
Medical guidelines are supposed to reflect the best available science. But the process of reviewing new studies, convening expert panels, drafting recommendations, and pushing them out to clinicians is slow. On average, it takes 17 years for research findings to make their way into routine clinical practice, according to the well-known estimate that Singhal cites. That delay isn't just an academic inconvenience. It means treatments that have been shown to work can sit on the shelf for nearly two decades while patients receive older, less effective care.
The bottleneck isn't a lack of evidence. It's the manual, time-consuming work of synthesizing that evidence into clear, actionable guidelines. Singhal's work at OpenAI is focused on using AI to speed up that synthesis — not to replace expert judgment, but to handle the heavy lifting of scanning literature, drafting summaries, and flagging where recommendations need to change.
What Singhal is actually building
Details of the system are still emerging, but the core idea is straightforward: use large language models to process the flood of new medical research and produce draft guideline updates that human experts can review and refine. Instead of a panel starting from a blank page every few years, they'd start from an AI-generated draft that already incorporates the latest studies. The goal is to cut the timeline from years to hours — or at least to something far shorter than 17 years.
Singhal hasn't announced a product or a partnership with a specific medical body. The work appears to be at the research and prototyping stage. But the direction is clear: apply the same technology that powers chatbots to the unglamorous but high-stakes work of keeping medical guidance current.
Why faster guidelines matter for patients
The implications for patient care are direct. When guidelines lag, clinicians fall back on habit, outdated training, or inconsistent local practices. A faster update cycle could reduce unwarranted variation in care and get proven treatments to patients sooner. It could also help health systems in low-resource settings, where access to the latest guidance is often even more limited.
There's also a cost angle. Delayed adoption of best practices wastes money on ineffective treatments and preventable complications. Speeding up guideline updates won't solve every problem in healthcare, but it addresses a root cause of inefficiency that rarely gets attention because it's invisible to most patients.
The hard part: trust and verification
AI-generated medical guidance raises obvious questions. Who checks the AI's work? How do you ensure it doesn't miss a crucial study or misread a nuanced finding? Singhal's approach seems to keep humans in the loop — the AI drafts, experts decide. That's a sensible safeguard, but it also means the speed gain depends on how quickly experts can review the drafts. If review remains slow, the 17-year gap won't close as dramatically as hoped.
There's also the matter of trust from the medical community. Guidelines carry weight because they come from respected bodies like specialty societies and government agencies. An AI-assisted process will need to earn the same credibility, which means transparency about how recommendations are generated and validated.
What happens next
Singhal hasn't given a timeline for when this work might reach real-world guidelines. For now, it's a research effort with a compelling target. The next step would likely be a pilot with a medical organization willing to test AI-drafted updates against the traditional process. Until then, the 17-year gap remains the standard — and patients are waiting.
