At Ochsner Health, AI is moving from experimental project to core clinical capability, and Jason Hill, MD, Innovation Officer, says the physicians and health systems that master it first will gain a decisive advantage in care quality and efficiency.

Leading an innovation division inside an $8 billion, 46-hospital integrated delivery network that also operates its own health plan, Hill draws on a decade spent wiring a single Epic EHR, serving as associate CMIO for inpatient care and directing informatics efforts to bridge the gap between clinicians and technology teams.

For Hill, the opportunity is to use technology to redesign care models across a broad Gulf South footprint while keeping ethics and safety at the center of deployment; as he put it, “the very interesting part about AI is that it is ultimately a democratization of human knowledge.”

That capability is already visible in tools such as OpenEvidence, which he uses on the hospital floor and which can score at or near perfect on licensing exams, matching or surpassing the test-taking performance of many practicing physicians.

He sees that level of cognitive capacity as a fundamental shift in knowledge work, including medicine and software development, and believes clinicians who adopt AI as a partner will outcompete those who do not, saying that “doctors are not going to replace AI, but doctors who use AI will most certainly replace those that do not.”



Balancing Governance With Rapid Experimentation

Hill spends much of his time thinking about what happens after AI models are introduced into clinical workflows and begin to learn from new data and interactions, a behavior pattern that differs sharply from fixed-function software.

He noted that generative models can drift in subtle ways as they are tuned and retrained, producing responses that may no longer align with original expectations, and that some research groups have documented behaviors such as deception, sycophancy and other forms of misalignment that raise novel safety questions for healthcare.

Traditional monitoring approaches that compare model predictions with outcomes are difficult to apply to narrative or conversational outputs, so evaluation must instead focus on the quality, consistency and reliability of large volumes of free-text responses. In Hill’s view, that scale challenge leaves organizations little choice but to enlist more automation in the oversight process; in his words, “realistically, the only way to do that is with another AI, so you then have AIs judging AIs.”

That dynamic, he suggested, will force health systems to design governance structures that treat advanced models less like conventional applications and more like complex, semi-autonomous systems whose behavior must be continuously observed, logged and reviewed.

From Ambient Listening to Augmented Intelligence

On the clinical front, Hill sees the fastest gains coming from tools that remove clerical burden while leaving final decisions with clinicians, such as ambient documentation systems that listen to visits and generate notes directly in the EHR.

These services are already widely adopted across Ochsner physicians and many peers nationally, he said, easing burnout and freeing clinicians to focus attention on patients instead of keyboards.

He expects the next wave of capability to assemble a much richer context around each encounter, drawing from the EMR, prior imaging, genomics, external records and even social factors to give physicians a consolidated view before they act.

Hill described a future state in which AI helps transform an open-ended clinical question into a small set of likely diagnoses or management options that a human can quickly assess, turning complex differential diagnosis into a workflow that resembles responding to a well-designed multiple-choice exam.

In that model, augmented intelligence becomes the goal in that...