The Proximate and the Possible

Biomedicine has never lacked for miracles. The harder question is who they're for.


There is a particular kind of humility embedded in one of Reinhold Niebuhr's most durable observations, democracy is not in the business of solving problems, but of finding proximate solutions to problems that are at their root insoluble. He was writing about politics, about the management of human conflict and competing interests, about the stubborn persistence of injustice even in the most earnest of societies. But he might just as easily have been writing about medicine.

Disease, after all, is the original insoluble problem. The body ages. Cells misfire. Systems fail. Evolution, indifferent to human flourishing, has left us exquisitely vulnerable to the very complexity that makes us remarkable. No political will, however fierce, and no scientific ingenuity, however brilliant, will ever fully resolve this condition. What medicine does is find proximate solutions. Antibiotics do not eliminate infection; they buy time. Statins do not stop hearts from failing; they slow the process. Insulin does not cure diabetes; it manages it, year after year, injection after injection, in the countless small negotiations that constitute a life with chronic illness. Medicine, practiced honestly, is the art of the proximate.

And yet. Something is shifting in the architecture of what is proximate and what is possible. This shift has a name, though it is so overused as to have become almost meaningless: artificial intelligence. To invoke AI in 2026 is to risk being dismissed as either a naïf or a huckster. The hype cycle has been so aggressive, the promises so extravagant, that a kind of weary skepticism (or worse) has set in even among those who understand the technology best. And so it is worth slowing down, resisting both the breathless optimism and the reflexive cynicism, and asking a harder question: not whether AI will transform medicine, but how, and for whom, and at what cost, and governed by what principles.

*****

Begin with what is genuinely new. For most of the history of medicine, the bottleneck was knowledge. Physicians accumulated experience over careers; institutions accumulated knowledge over generations; the literature grew slowly in the pages of journals that fewer and fewer practicing clinicians had time to read. The gap between what medicine knew and what medicine did was staggering. Studies have suggested with depressing consistency that it takes an average of seventeen years for a research finding to be routinely incorporated into clinical practice. Seventeen years. Entire cohorts of patients lived and died in that interval, untreated or undertreated, not because the knowledge didn't exist, but because the systems for deploying it were too slow, too siloed, too human.

AI compresses time in ways that are genuinely difficult to overstate. AlphaFold's prediction of protein structures—a problem biologists had labored over for half a century—was functionally solved in a matter of months. The implications for drug discovery are not incremental; they are structural. Targets that were previously “undruggable” are being reconsidered. Diseases that were poorly understood at the molecular level are becoming legible. In the language of the possible, we are encountering something rare: a tool that does not merely extend existing capabilities but creates new ones. Consider the impossible—not as a motivational injunction but as a genuine intellectual instruction. Consider what was, until recently, genuinely beyond reach: predicting how a molecule will fold, simulating how it will interact with a receptor, screening billions of compounds in silico before a single experiment is run. These things are happening now; they are not science fiction.

And yet, and yet. History offers a useful corrective to technological optimism. Every great leap in biomedical capability has been followed with dispiriting regularity by a gap between what medicine can do and what most people actually receive. The discovery of penicillin did not immediately reach all the patients who needed it. The development of antiretrovirals transformed HIV from a death sentence into a manageable condition, but only for those in wealthy countries with robust healthcare systems. The mRNA platform that produced two highly effective COVID vaccines in under a year also exposed in real time the gulf between scientific achievement and equitable distribution. Capability is not delivery. Discovery is not access.

This is the domain where Niebuhr's insight cuts deepest. The question of who benefits from a medical breakthrough is not a scientific question. It is a political one, a moral one, an institutional one—the kind of messy, contested, never-fully-resolved question that democratic societies are supposedly organized to address. And democratic societies, to put it plainly, have not covered themselves in glory. The American health system, the most expensive in the world, leaves tens of millions underinsured and millions more in medical debt. Drug pricing in the United States bears no systematic relationship to therapeutic value; it reflects, instead, the negotiating power of pharmaceutical manufacturers in a market deliberately structured to favor them. The pipeline of new medicines is distorted by profit incentives toward conditions that are common, chronic, and prevalent among the wealthy. This is not because researchers are venal, but because the system that funds them is. Rare diseases, neglected tropical diseases, the infectious killers of the global poor: these remain systematically underfunded, not because they are scientifically intractable, but because the people who suffer from them cannot constitute a profitable market.

Into this landscape, AI arrives. And the question, the genuinely difficult, genuinely important question, is whether it will be deployed in ways that replicate and amplify existing inequities, or whether it might, under the right conditions, begin to address them.

*****

There are reasons for cautious optimism. The computational democratization that AI represents could, in principle, reduce some of the structural advantages that large pharmaceutical companies hold over smaller, more nimble, more mission-driven innovators. The cost of early-stage drug discovery, historically a barrier that kept academic researchers and nonprofits from competing with industry, is falling. Open-source AI models for protein structure prediction, molecular generation, and bioactivity screening are proliferating. Researchers at universities in Lagos and Nairobi and Dhaka now have access, in principle, to tools that were five years ago the exclusive province of well-capitalized Western laboratories. The inputs to biomedical innovation are becoming, however fitfully, more distributed.

There is also something important in the nature of the problems AI is best suited to solve. Large language models and multimodal AI systems are exceptionally good at pattern recognition across vast datasets which makes them valuable not only in drug discovery but also in diagnostics, particularly in settings where specialist physicians are scarce. An AI system that can identify diabetic retinopathy from a photograph, or flag early signs of sepsis in an electronic health record, or assist an overworked rural clinician in a differential diagnosis—these applications are not glamorous, but they are consequential. They represent a genuine expansion of the proximate: things we could not reliably do before, that we now can, that save lives that would otherwise be lost.

But the cautions are real. AI systems trained on data drawn predominantly from wealthy, white, Western populations will perform worse for patients who do not resemble that training population, which unfortunately the available evidence suggests is precisely what is happening. Dermatology algorithms trained largely on lighter skin tones perform measurably less well on darker ones. Cardiac risk models calibrated to one demographic show differential accuracy across others. Algorithmic bias is not a hypothetical; it is a documented feature of deployed systems, generating inequities that are invisible to the institutions that benefit from them and highly visible to the patients who are harmed.

There is, too, a deeper structural concern. The development of the most powerful AI systems is concentrated in a handful of companies with enormous market capitalizations and very particular incentives. These companies are not, in the main, oriented toward the health needs of low-income countries or underserved domestic populations. They are oriented toward the same markets that have historically captured the attention of pharmaceutical firms: large, affluent, insured. The risk is not that AI will fail to produce biomedical breakthroughs. The risk is that it will produce breakthroughs that are proprietary, expensive, and inaccessible to the populations that carry the greatest burden of disease.

*****

In Niebuhr's estimation, democracy finds proximate solutions. The proximate solution is rarely satisfying; it is rarely just; it is almost always contested. But it is what is available to imperfect institutions populated by imperfect human beings operating under genuine uncertainty. The task, then, is not to wait for a perfect solution to AI governance, to drug pricing, to health system design but to build the proximate one, with urgency and rigor and a clear-eyed understanding of what is at stake.

This means several things, concretely. It means investing in open-access biomedical AI infrastructure that is not subject to the proprietary restrictions of commercial developers—building the scientific commons that the market will not build for itself. It means establishing governance frameworks for algorithmic health tools that treat equitable performance as a baseline requirement, not a bonus feature. It means funding the development of AI applications oriented toward neglected diseases and underserved populations, through the kind of mission-driven, blended-finance vehicles that have begun to emerge in global health and that deserve to be scaled. And it means insisting that the question of distribution is as urgent as the question of discovery. It is not enough to produce a cure. The cure must reach people.

None of this is easy. None of it will be fully achieved. The problems at the intersection of medicine, technology, and equity are, in Niebuhr's sense, genuinely insoluble—rooted in the conflicting interests of human beings and institutions that will never be fully reconciled. But the proximate solution is within reach. Not in the visionary future of science fiction, but in the harder, duller, more necessary work of institutional design, public investment, and political will.

Consider the impossible. And then consider what it would take in the most practical terms imaginable to make it merely difficult. That is the work. That is, in the end, what democracy is for.

‍