A few weeks ago, over coffee with the head of R&D at a mid-sized pharma company, I asked what he made of the industry’s headlong rush into AI-enabled drug discovery. He looked reflective. “We’ve signed three partnerships this year,” he said. “But I can’t say we’re doing anything different from or better than our competitors. We’re all talking to the same few AI companies, about the same targets, with similar data. It feels like we’re all converging on the same point.”

At this, my Darwinian instincts kicked in. I’d just returned from Ethiopia, where I’d seen the famous fossil ‘Lucy’ and read about Homo sapiens’ spread out of Africa and its consequences for human diversity. Stick with me for a moment and I’ll show you how our industry’s latest trends and that ancient migration combine to illuminate the future of AI in the life sciences.

My ancestors, and those of all non-African humans, migrated from Africa to the Middle East about 60,000 years ago, along with a few thousand of their relatives. That small group formed a bottleneck in our genetic history. It’s why there is more genetic variation among the 1.5 billion Africans alive today than among the other 6.5 billion people who are descended from that founding population. Sewall Wright called this the founder effect, when a small group becomes the baseline for everything that follows and later generations inherit its quirks, biases and blind spots. The founder effect isn’t just a curiosity of human history. It’s a reminder of how early conditions shape long-term outcomes, often in ways that are invisible at the time.

Perhaps you can already see why my coffee conversation got me thinking. My R&D friend was describing a small founding population of AI companies including Recursion, Insilico and Exscientia. These companies have the most mature pipelines, the most visible platforms and the densest web of pharma partnerships. I don’t pretend to know the fine detail of their methods, but their dominance suggests that much of the industry is basing its discovery efforts on a relatively narrow range of approaches. Just as Homo sapiens passed through a genetic bottleneck on leaving Africa, pharma may be passing through an AI bottleneck today. A handful of AI discovery companies are becoming the de facto ‘founders’ of the industry’s AI genome. Their architectures, training data and modelling assumptions are rapidly becoming the industry’s shared inheritance. This is not inherently bad. But it is inherently consequential.

This biological analogue suggests three structural risks when many companies rely on the same small set of AI platforms. First, early biases may become industry wide. If a model over-represents certain chemical spaces or under-represents certain biological mechanisms, that bias propagates across every partner pipeline. What begins as a quirk of one model becomes a systemic blind spot.

Second, exploration might narrow. AI is supposed to expand the search space. But if everyone uses the same tools, the search space may actually contract. Third, innovation may become path dependent. Future discoveries may be constrained by the assumptions embedded in the founding models. If the ‘founders’ favour certain modalities, targets or data types, those preferences may echo through the industry for years.

In evolutionary terms, the industry is undergoing a genetic bottleneck at the very moment it believes it is entering an era of unprecedented diversity.

The danger isn’t that AI-driven discovery will fail. It’s that it will succeed too narrowly. A world in which every company uses AI is not necessarily a world of diverse innovation. It may be a world of inadvertently optimised convergence.

But the founder effect also creates opportunities. It opens ecological niches for smaller AI companies exploring different spaces. And it invites ‘gene flow’ – ideas, methods and architectures from other industries that aren’t part of pharma’s bottlenecked genome. Variation could come from unexpected places – academic labs, open source models or cross-disciplinary approaches borrowed from fields like materials science or climate modelling.

No one knows how AI drug discovery will play out. But Darwin taught us that evolution is powerful only when variation is preserved. Without it, even the most promising new species can become evolutionary dead ends. Darwinian thinking implies that pharma’s AI future will be shaped not just by the brilliance of its algorithms but by the diversity of its foundations. And that, as Darwin might say, is a view worth taking.