Humans in the loop.
That term has become the industry shorthand for assuring people nervous about AI’s increasing foothold in healthcare. There’s always someone watching over what AI does. We are there to make sure nothing goes awry. But the problem is that sometimes we may be in the way. So now, humans are faced with an interesting dilemma. To unleash the power of AI, sometimes we need to cede space. But then that may also be dangerous – what if something goes wrong?
Despite this Catch 22 vibe, healthcare companies like Merck are venturing forth, adopting generative AI and agentic AI tools — where the software has agency/power to take actions instead of simply surfacing information on which humans act — to tackle diseases. In a recent interview, Asheesh Chhabra, vice president and chief data officer at Merck, cast the role of humans in a new fashion.
“Let’s take the human out of the unnecessary middle there, but put them in the right room,” Chhabra declared. “The human will never go. How do we put him in the right room?”
He shared an example of what the unnecessary middle is: the publication of clinical trial results based on all the data that has been gathered. That process can take up to six months. But now with the generative AI capabilities, that output takes days, and a Merck employee with the right expertise can simply oversee the final output and approve it.
Clinical trials reform through AI is, of course, something that all pharma companies are exploring, as is AI drug and target discovery. Chhabra mentioned how TEDDY — the large language model developed by Merck and external collaborators — is being used in the biologics large molecule space, focusing on oncology. There’s another AI language model being leveraged for small molecules too. He added that Merck has published data on both these models.
“So we’re not really ring-fencing them into just one therapeutic area,” he said. “But we’re going to start in a set of target identifications that our subject matter experts are working on and prove that out ….”
In the field of agentic AI, Merck is looking at a few different areas where they could be deployed. One involving communications with healthcare providers around market access of drugs. This would be on the commercial side of the business. Another area would be in boosting productivity and efficiency in engineering and manufacturing. And finally, agentic AIs could be used in digital twins. Digital twin technology allows pharma companies to create digital avatars of actual humans that may someday in the distant future reduce the need to enroll actual people in clinical trials.
Enrollment is one problem but keeping people engaged in clincial trials over time is also challenging. This is where Chhabra sees an opportunity for digital twins.
“I think with that, we see a strong opportunity with digital twins to maintain what I call the behavioral aspect of the patient,” Chhabra said.
In other words, after the initial enrollment, even if people drop off during the clinical trial, there is enough information gathered that the digital twin technology will be able to predict how that “person” would behave had they completed the clinical trial.
“So there are a couple of examples that we’re looking at driving our pharma value chain,” Chhabra explained. “It doesn’t mean there’ll be an agent engaging with the patient. There are agents that we use that enable that business process that we can optimize and take a human out of the loop and put that human on more complex tasks.”
The success of the AI or human in that new, more complex role, for that matter, will largely depend on how strong the foundational data is.
“Data is truly the fuel for your AI,” Chhabra said. “I cannot go create a simple UI model using ChatGPT. I now need to get access to strong data.”
Photo: ipopba, Getty Images