Balancing Human Oversight and Automation in AI-Driven Trials

In a recent video interview with Applied Clinical Trials, Michelle Longmire, MD, co-founder and CEO of Medable, discussed the role of agentic artificial intelligence (AI) in clinical trials. She highlighted that 45% of clinical development time is wasted due to delays and siloed data. Agentic AI aims to reduce this by leveraging technology in areas previously untouched, freeing up human resources for strategic work. The human-in-the-loop concept ensures that AI supports, rather than replaces, human decision-making. Integration of AI across legacy systems is facilitated by model context protocols, reducing manual effort. Automation handles routine tasks, while critical decisions remain human-controlled. No-code platforms for AI tools require clear job descriptions and validation similar to human training.

ACT: How can human-in-the-loop oversight be preserved when introducing AI-driven tools for critical clinical trial functions like monitoring, planning, and data review?

Longmire: The human in the loop part of agentic AI is critical in any high stakes environment, and certainly in clinical development. With Agent Studio—we knew this going in—we're able to essentially imagine kind of like a lever that you can pull, full human in the loop all the way to fully autonomous. When you look at specific actions, specific goals, specific decisions, you want those to be highly human in the loop, and maybe human in the loop at a very frequent and regular basis, maybe human in the loop for the final decision. Ultimately, that's up to what goal and task you're out to bring agentic AI into. I think having that as a dial up and dial down is really, really important. I'd say beyond just having a single human in the loop, we want to be able to use AI to even augment the way that humans are making those final decisions. You imagine today, you're monitoring a site, and you're looking across data from 13 different systems, and you're trying to make a critical decision, but not only are you looking at disparate data, that decision in the optionality is limited to your own mind. With agentic AI, not only can we have human in the loop, but we can actually surface potential options for those decisions that might not be those that are the only ones that come to the mind of the human.