Faster, Toward What? Innovation and the Direction of Technical Change

The idea that innovation has a direction, and that the direction is itself economically determined, is older than the growth models it complicates. John Hicks raised it in 1932, observing that a change in the relative price of the factors of production (e.g., labor becoming more costly relative to capital) would induce invention aimed specifically at economizing on the factor that had grown expensive. Invention, in other words, follows prices.

It is a measure of how unsettling this idea was at the time that Hicks himself came to doubt it; he pulled his book containing the idea from print for decades, telling readers in a later edition that his views had shifted too far for him to stand behind it. And there was in fact a serious objection to answer which was pressed most sharply by W.E.G. Salter in 1960:  a firm bent on profit should want to economize on all of its costs, not merely the most expensive ones, so that whether an advance happens to save labor or capital is beside the point. The concept of “directed” in relation to innovation was therefore shunted to the side.

In 1969, Nathan Rosenberg conceded a la Salter that firms are largely agnostic about which factors to save on. Rosenberg argued, though, that the day-to-day constraints of production act as “focusing devices,” concentrating innovative activity on some problems rather than others. It was a partial, and informal, answer about the direction of new ideas but one that seemed to call for more work. Unfortunately, this intuition ended up sitting mostly dormant for years, a hypothesis without a machine to run it.

The machine finally arrived in 2002 with Daron Acemoglu’s formalization of “directed technical change.” Acemoglu identified two forces that govern which direction innovation flows. The first is the price effect: there is more incentive to develop technologies that serve expensive goods because a higher price per unit means a higher return on the innovation that improves the good. The second is the market-size effect: there is more incentive to develop technologies that serve large markets because the fixed cost of the expensive first copy is spread over more units, and non-rivalry means each additional unit sold is nearly pure return. These two forces can pull in the same direction or against each other, and which one dominates turns on how substitutable the relevant goods are for one another. When substitutability is high, the market-size effect wins. Innovation flows toward the biggest market, not necessarily the biggest problem.

In a directed-technical-change world, the size of a market is not merely a fact about who will buy a product once it exists. It is a force acting backward in time on which products come to exist at all. The market does not simply serve demand; it recruits invention toward demand. And “demand,” in the only sense the economic model can register, means demand backed by the ability to pay.

‍

One might treat all of this as an elegant academic abstraction were it not for the fact that biomedicine offers an unusually clean real-world test of it. Acemoglu and his co-author Joshua Linn asked in a 2004 paper entitled Market Size in Innovation: Theory and Evidence from the Pharmaceutical Industry whether the entry of new drugs into a therapeutic category could be predicted solely by the size of the market for that category. And, to avoid the obvious circularity of successful drugs creating their own demand, they controlled market size using the predictable aging of the American population. The result was decisive. Anticipated increases in market size driven purely by demographic growth accurately predicted the subsequent arrival of new drugs. Innovation measurably just followed the money.

Now compare that finding against the global distribution of disease. The conditions that impose the heaviest burden on the world's poor (diarrheal and parasitic diseases, tuberculosis, neglected tropical diseases) represent enormous need but almost no market size in the only currency the growth model understands. Their sufferers cannot constitute a market, not because their suffering is small but because their purchasing power is. And so the model makes a prediction, cold and precise: these conditions will be systematically under-innovated.

The prediction is confirmed with striking regularity. The Global Forum for Health Research gave the phenomenon its enduring shorthand, the 10/90 gap, the observation that only a small fraction of health-research resources address the conditions responsible for the overwhelming majority of the global disease burden. Trouiller and colleagues, surveying the pipeline across the last quarter of the twentieth century, found that of 1,393 new chemical entities brought to market between 1975 and 1999, just 16 were for tropical diseases and tuberculosis, the afflictions of the poorest billions. The pattern did not break with the new century. A follow-up assessment by Pedrique and colleagues of products registered between 2000 and 2011 found only four new chemical entities for neglected diseases among more than three hundred total NCEs.

The neglected-disease gap is not evidence that the market is broken. It is evidence that the market is running its program correctly. The 10/90 gap is not the deviation from the model. It is the model: directed technical change, aiming the idea-engine at market size, producing precisely the pattern the economics says it will produce.

‍

In March, these Notes followed Kenneth Arrow's 1963 argument that the market for medical care is structurally unlike other markets in ways that make the competitive-efficiency case simply inapplicable (i.e., irregular demand, unverifiable quality, asymmetry of information). That essay, and the tradition it draws on, frames the trouble as market failure: the market is a mechanism that is supposed to produce efficient outcomes and, in healthcare, fails to because the conditions for efficiency are absent. The policy instinct that follows from market-failure framing is repair. Find the broken part—the information asymmetry, the externality, the under-provided public good—and patch it. Regulate the monopoly, subsidize the vaccine, mandate the disclosure.

Directed technical change points at something adjacent but importantly different, and the difference is not academic. It says: with respect to the direction of innovation, the market is not failing at all. It is succeeding. It is allocating inventive effort toward purchasing power with an efficiency that is, in its own terms, almost admirable. The neglected diseases are not neglected because a mechanism designed to serve them broke down. They are neglected because no mechanism was ever pointed at them, and the mechanism that exists is pointed, correctly and by construction, somewhere else. To call this “failure” is to misdiagnose it, and misdiagnosis leads to the wrong treatment. You cannot repair your way out of a direction problem. A perfectly functioning price system, repaired to perfect efficiency, will aim innovation at purchasing power more precisely than a broken one. The task is not to fix the signal. It is to change what the signal is made of.

And lest this direction issue be filed away as just a problem of the poor or of somewhere else, it is worth seeing that the same distortion operates in the wealthiest corners of the market where no one is too poor to buy. Budish, Roin, and Williams demonstrated in 2015 that the structure of the patent system itself bends the direction of cancer research. Because the patent clock effectively runs from the moment of filing rather than the moment of approval, a therapy that requires a long clinical trial to demonstrate its value, for example, a prevention agent, or a treatment for early-stage disease, whose benefit can only be shown by following patients for years, enjoys a shorter period of effective market exclusivity than a drug for late-stage disease whose trials end quickly because, grimly, the patients do. The result is a measurable tilt of private research effort toward late-stage treatment and away from prevention and early intervention, driven not by the science and not by the burden of disease but by an institutional artifact (patent law) of how exclusivity is timed. Here is direction distortion in the richest market imaginable, produced by a design choice no one made on purpose. It is the thesis of this essay in miniature: the vector of innovation is set by the architecture of incentives (in this case, patent-driven exclusivity), and that architecture is largely invisible to the people operating inside it.

None of this by the way is to retract the market-failure account. Arrow's asymmetries are real and continue to do their damage alongside the direction problem; the argument here is in addition to, not instead of. But the two diagnoses call for different remedies. Market failure asks us to repair a mechanism. Directed technical change asks us to redesign a signal.

‍

If direction is endogenous, that is, if it follows market size and prices rather than emerging from the nature of the science, then the direction of ideas is not fate. It is a variable. And a variable can be re-directed.
‍

If innovation flows toward market size, then the way to redirect it is to manufacture market size where there is need but no money. This is the logic beneath a family of instruments that are usually discussed separately and are better understood as variations on a single idea: the deliberate construction of demand signals that the price system will read and obey. Michael Kremer and Rachel Glennerster gave the clearest articulation of it in the advance market commitment, a binding promise by donors to purchase, at a pre-specified price, a vaccine that does not yet exist for a disease the market has ignored. The commitment does not repair a broken market. It builds a market that was never there, so that the direction of innovation, following its usual rule, turns toward it. The pneumococcal advance market commitment launched in 2009 was the first large-scale test of exactly this proposition: pay to relocate the vector.

The priority review voucher works the same lever from a different angle; it attaches a tradable, monetizable asset to the act of developing a neglected-disease therapy, converting a market the developer cannot profit from into one it can. Push mechanisms (grants, subsidized research) and pull mechanisms (prizes, commitments, vouchers) differ in what they reward, and in the distortions they introduce (for it must be admitted that each intervention mode can manufacture its own new failure modes even as it corrects the direction issue). But they share a common structure: they are all attempts to supply the idea-engine with a heading that need, rather than purchasing power, has chosen.

There is an institutional layer here too, and it connects directly to our essay on commons governance that appeared in April. Mission-driven vehicles (e.g., product-development partnerships, disease focused foundations that take equity in the therapies they fund, the whole apparatus of blended finance in global health) are most often described in the vocabulary of charity, as generous efforts to fill a gap the market left behind. The directed-technical-change lens offers a less sentimental and more accurate description. These institutions are not filling a gap in a mechanism that is otherwise working. They exist to reset the direction signal itself—to introduce, deliberately and by design, a heading that the price system would never have generated by itself. When the Cystic Fibrosis Foundation took equity positions and controlled a research agenda, or when a product-development partnership commits patient capital to a tuberculosis regimen, what is happening is not that a market failure is being patched. It is that a community is seizing the direction signal away from the price system and pointing it somewhere the price system would not have looked. Elinor Ostrom's design principles for the commons, which the April essay took up, turn out to describe not only how a community governs a shared resource but who gets to aim the innovation that resource makes possible. Ostrom's “rules crafted by the users themselves” apply as much to the what of research as to the how of its governance. Direction, it turns out, is a commons that can be captured for good ends as easily as for bad ones.

‍

There can be found a quiet misunderstanding in the word equilibrium. It takes a state of affairs that was produced by choices, such as institutional design, the distribution of purchasing power, decisions about what to reward, and re-describes it as the resting point of a system, a place things naturally settle, as morally meaningful as a ball at the bottom of a bowl. “The market directs innovation toward purchasing power” has the cadence of a law of physics, but it is nothing of the kind. It is a sentence about a set of arrangements we have built and could rebuild, laundered through the grammar of science until it sounds like something that happened to us rather than something we did.

In May, drawing on Reinhold Niebuhr, these Notes argued that medicine is the art of the proximate—that disease is the original insoluble problem, and that the honest work is not the perfect solution but the achievable one, pursued with urgency and without illusion. The direction of innovation belongs to exactly this register. We will not devise a mechanism that aims the idea-engine at human need with perfect justice; every re-direction instrument carries its own distortions, its own opportunities for capture, its own new inequities waiting to be discovered by the people they harm. The advance market commitment can be gamed. The priority voucher can be hoarded and traded by parties the scheme never meant to enrich. The mission-driven vehicle can drift toward the fundable rather than the needed. There is no equilibrium of pure justice to be reached and then rested in.

But the proximate solution is within reach, and the refusal to mistake the current arrangement for a natural fact is the first move toward it. Niebuhr's point was never that injustice is permanent and effort futile; it was that the work is never finished, that every settlement is provisional, and that the temptation to declare a contingent arrangement final is itself a kind of moral failure. The direction of technical change is not a solved problem awaiting a clever mechanism. It is a standing responsibility, something a society must choose, and keep choosing, and never be permitted to forget it is choosing. To call the neglected-disease gap an equilibrium is to forget.

‍

These Notes have returned on more than one occasion to the question of what agentic artificial intelligence does to the institutions of biomedicine. Last month, the argument was that agents compress the transactional layer of coordination (search, drafting, monitoring) while leaving the governance layer (the question of whose interests get represented and what counts as a fair allocation) largely untouched and, by contrast, more visible than before. Directed technical change lets us close that loop, because it names precisely what lives in the governance layer that agents will not resolve: the heading. The vector. The choice of which problems the now-cheaper idea engine is aimed at solving.

Consider what a lower first-copy cost does to the market-size calculation. In Acemoglu's framework, a market must clear a certain minimum size before the fixed cost of innovation is worth bearing. If agentic systems genuinely lower that fixed cost (e.g., the cost of the expensive first copy of a new therapeutic idea) then the minimum market size that justifies innovation falls, and problems that were previously too small to attract inventive effort will clear the bar. On this reading, cheaper idea production is democratizing: the direction gap narrows, not because anyone redirected the vector but because more small-market problems become economically worth solving. This is the hopeful case, and it is not naïve; it follows directly from the mathematics.

But the model gives us the other case with equal rigor too. Suppose the fixed cost falls everywhere, across every market at once. Then the relative market-size gradient, the thing that actually determines direction, is unchanged. Everything gets cheaper to develop, the large markets remain proportionally more attractive than the small ones, and if the productivity gains are captured disproportionately by the incumbents already serving the affluent, the equity gap holds, or widens, only now faster and at lower cost in both directions. Cheaper drift is still drift. A more efficient engine, pointed the same way, arrives at the same potentially unjust destination sooner and for less money. This is the same lesson the automation literature learned about labor: a technology does not have a direction of its own to discover; it has a direction that institutions and incentives assign to it.

Which case obtains, democratizing or amplifying, is therefore not a prophecy to be awaited but a design question to be answered. And it is the same design question this essay has sought to  press from the start. Romer taught us to make the technical engine faster. He was right, and the achievement was enormous. But velocity was never the whole of the problem. The hard part is the vector: the direction, chosen and re-chosen, that decides whether all that speed carries us toward the frontier of human need or merely faster toward the same familiar markets. The engine is getting faster. The only question that finally matters is where we decide to point it.♦


‍

‍