AI medical breakthroughs might happen sooner than you think

 

       We tend to tell ourselves comforting stories about medicine. But anyone who has spent time near drug development knows that the reason breakthroughs are slow is that medicine is a search problem inside a maze, with unusually high stakes and unusually high friction. Biology is complex in ways that resist simplification. Regulation is essential, but slow. And the cost of failure is measured in human lives.

AI is good at search and pattern-finding in messy, high-dimensional spaces. That’s why it maps so naturally onto discovery problems where you’re rarely looking for a single “right answer,” but instead trying to navigate an enormous possibility space and get better, faster, at choosing what to test next. This is also why AI tools are uniquely good at coding, for example.

Cells and proteins and chemical interactions aren’t as malleable as software, but they’re not inert matter either. They have structure and rules that can be modeled imperfectly but increasingly well.

This is the terrain where AI can create real leverage.

For example: It’s plausible that “AI-discovered” drugs—drugs whose origin story includes machine learning as a core component of discovery—arrive sooner than most people expect.

Drug discovery is an enormous combinatorial problem. The space of possible molecules to treat a disease  is vast. The number of ways a molecule can interact with a biological system (intended and unintended) is vast. The traditional approach has been slow not because scientists lack intelligence, but because the search is hard and the feedback loops are expensive.

AI helps because it can become a better guide through that search space. It can propose candidates, rank hypotheses, identify patterns across disparate data, and help us design the next experiment more intelligently. It can reduce wasted motion. It can increase the number of credible shots on goal. And because biomed development is so costly, even incremental improvements can shift the economics of what is worth attempting.

When people say “AI will cure most diseases in 20 years,” I interpret it as an expression of direction rather than a literal forecast. Some diseases may prove structurally resistant. New diseases will emerge. And the definition of “cure” is itself slippery. But the broader claim, the idea that AI can materially accelerate the process by which we discover and develop therapies, seems not only plausible, but likely.

It's also important to note that as medicine and drug discovery using AI evolve, regulation should too. In other words, the limiting factor may become less “can we propose plausible therapies?” and more “can we validate them responsibly at the pace the discovery engine can generate them?” This is not an argument for no regulation or a criticism of regulation in medicine. In fact, the FDA has done a great job of protecting hundreds of millions of Americans while also delivering them the cures and treatments they need. But it does mean we should think about the clinical and regulatory infrastructure as part of the innovation surface when we’re talking about how AI will reshape the space.

It’s worth remembering how much of modern human prosperity rests on biomedical progress we now take for granted. Life expectancy a century ago was radically different. Vaccines, antibiotics, public health infrastructure, diagnostics changed what it means to live a long, healthy, and purposeful life.

If we get this right—and we must get it right—the story will be that we built a better engine for discovery, and then we did the hard institutional work to translate discovery into outcomes. Unfortunately for those trying to paint a dramatic story of AI, this kind of progress doesn't arrive with one giant breakthrough. Instead, it arrives through compounding intelligence, fewer dead ends, faster cycles, better therapies, and, over time, a less barbaric relationship with disease.

Listen: https://play.megaphone.fm/bt6khyejrz-fg1i4pwk0cg

Watch: https://youtu.be/38D7SPQO81Q

Read: https://www.possible.fm/podcasts/riffs046/

You can catch and subscribe to more Possible here: https://www.possible.fm/