From Promise to Practice: Leading Science in the Age of AI

A few years ago, artificial intelligence (AI) in drug discovery felt like a future-facing idea—interesting, exciting, but largely theoretical. Today, “someday” has become “now.” AI is no longer just a tool we imagine using; it’s something we rely on every day to make better, smarter decisions in science.

As a physician-scientist, I’ve watched that shift happen in real time. And what’s struck me most isn’t just how powerful the technology has become, but also how much it depends on the people behind it.

At Bristol Myers Squibb, we’ve learned that realizing AI’s potential takes more than hype or horsepower. It requires three things: good data, clear scientific questions, and, above all, human creativity and judgment. AI doesn’t replace the researcher’s role; it elevates it. It pushes us to think more critically, design more rigorous experiments, and ask better questions.

Applying AI to the Foundations of Research

AI now plays a role across our entire R&D continuum—from early discovery all the way through development—to get transformational medicines to patients faster. Personally, I’ve seen a profound impact at the front end, where foundational decisions take shape. Here, AI helps us work smarter across the first three principles of our R&D strategy.

The first is selecting targets with strong causal human biology. For decades, identifying the right target has been one of the most difficult—and most pivotal—steps in drug discovery and development. A wrong turn here can lead to years of development with little to show. AI helps us make these early, critical decisions with more clarity. By analyzing vast population-scale datasets like UK Biobank and FinnGen, and integrating findings from genetic consortia and clinical data, AI tools complement other computational tools to help researchers pinpoint associations between genetic variants, biomarkers and disease. But more importantly, they help us generate stronger hypotheses by identifying patterns that suggest causality—insights that might otherwise be hidden in the noise. These tools don’t replace our hypotheses; they help us focus them by pointing to the most promising directions for investigation.

Second, once we’ve identified a target, we must determine how best to address it. That means matching the right modality—small molecule, antibody, CAR T cell therapy, or beyond—to the biological mechanism of action in play. This is not always straightforward. Here, AI models assist by simulating protein structures and predicting  pharmaceutical properties informing the long-term success of small and large molecules. These insights help us iterate faster, rule out dead-ends earlier and design therapeutic approaches that are more likely to succeed.

Finally, AI is improving how we bridge from research to development with a clearer path to clinical proof of concept. Traditionally, this transition has carried substantial risk at high financial cost. Promising preclinical results don’t always translate to patient benefit. But AI is helping us reduce that uncertainty. For example, we’re using machine learning to optimize patient selection, organize patient groups based on predicted responses and identify biomarkers that may provide early signs of efficacy.

To see how these principles are applied more broadly across teams at BMS, check out this video and article from our Science Firsthand series. I also shared a deeper dive on this topic, including how we’re building an AI-first culture, on my personal blog PlengeGen.

A New Era of Collaborative Hybrid Intelligence

We’re now working on a model I think of as collaborative hybrid intelligence. Scientists and algorithms are working together, each sharpening the other’s insights. This shift is reshaping how teams interact, how decisions are made and how knowledge is built.

Computational and experimental scientists now collaborate earlier and more often. Where these teams once operated in parallel, they’re now sharing questions, iterating on findings and co-creating strategy in real time. We’ve built shared languages—whether in modeling platforms, scoring systems or data visualization tools—that help bridge what used to be very different ways of thinking.

It’s also changed how we interpret results. AI can help identify patterns and generate predictions at a scale humans can’t match. But the value of those predictions depends on the scientist’s ability to assess, challenge and refine them. Our teams are learning how to trust the outputs while still applying skepticism. We’re asking: Does this align with what we know biologically? What assumptions went into the model? How could we validate it experimentally?

That intersection of AI with insight and intuition is where the most exciting work happens. AI doesn’t reduce the need for human thought—it demands a stronger version of it, one that’s informed not just by experience, but also by the ability to evaluate new forms of evidence and uncertainty.

These are cultural shifts as much as scientific ones. We’re not automating our way to better outcomes. We’re empowering researchers to ask bigger, better questions, and giving them the tools to find sharper answers.

The True Value of AI in Medicine

What excites me most about AI isn’t the technology itself but rather the transformation in how we work and what we can achieve together. Used thoughtfully, AI doesn’t just make science faster. AI makes science better. It helps us see farther, reason more clearly and connect ideas across silos and disciplines.

And most importantly, it keeps us focused on what matters most: making meaningful progress for patients.

This AI Appreciation Day, I’m reflecting on that evolution. The value of AI in medicine is not just its algorithms, but how it enables the people who use them. It helps us ask better questions, make smarter decisions, and ultimately bring more transformative therapies to those who need them most.

How have you seen AI shape drug discovery and development so far, and where do you see the biggest opportunity for AI to impact the future of the field? Let me know in the comments.