Published Original at the Financial Times, on Jan 31, 2026
In July 2011, the Space Shuttle Atlantis landed in Florida for the final time. Back then, the decision to end the shuttle programme was framed as a short-term transition, with a new generation of spacecraft soon taking its place. But for nearly a decade, the US had to rely on Russian Soyuz capsules to send its astronauts to the International Space Station.
That quiet admission of dependence points to something deeper. Knowledge that societies assume they have acquired for good can also disappear. As the US prepares to send astronauts to the Moon once again after several decades with the Artemis programme, the end of the shuttle reminds us that even the most advanced economies can lose capabilities, and that recovering them is harder than we expect.
When I presented my book The Infinite Alphabet in London last year, I was struck by an unexpected reaction. Audiences were less interested in stories about how knowledge grows than in stories about how it fades. They were anxious to hear about the loss of capabilities. Many felt that even in a wealthy and technologically advanced economy such as Britain, tasks once performed routinely — such as home repairs or railway construction — had become slow, expensive and hard to complete.
Knowledge loss is counterintuitive because we live in a world overflowing with learning. That pushes us to think of knowledge as infrastructure: a building that, once erected, will stand for ages. But knowledge is both alive and fragile. It is embodied in people, and the teams and communities they form, and is transmitted through repeated, almost ritualistic practices. Imagine trying to train a new generation of surgeons in a world where no surgeons remain. While that is a particularly dire example, it is reasonable to say that knowledge is like a shark: it must keep moving to stay alive.
Retiring the shuttle was not just grounding a spacecraft. It meant disbanding the team required to sustain a complex capability. Manuals and blueprints are poor substitutes for hands-on experience. To restore human launch capability, the US had to invest billions in contracts with SpaceX and Boeing in an effort to rebuild that capacity and those teams. An effort that, according to a 2019 Nasa audit, was running roughly two years behind.
But the loss of knowledge is not a story of public versus private investment. Corporate history is filled with similar warnings. The ability to produce film for Polaroid cameras, a product that felt settled by the 1970s, nearly vanished in the late 2000s.
Polaroid was the brainchild of Edwin Land, an American entrepreneur who pursued projects that were “manifestly important and nearly impossible”. But when the company’s last film plant in the Netherlands prepared to close in 2008, the teams and routines that had sustained that expertise began grinding to a halt.
The operation was rescued by Florian Kaps, an Austrian entrepreneur who had built an online business selling instant film. Kaps acquired the plant and kept it running with a skeleton crew of experienced staff. But even with the original equipment, the recovery was slow. It took the company named “The Impossible Project” years to regain the capability to manufacture high-quality colour film.
But are these just isolated anecdotes? Or is the loss of knowledge a more widespread phenomenon?
A clever way to measure forgetting is to use the same data scholars use to study learning. In industrial settings, learning is often measured by how productivity increases with experience. For example, if you have made a lot of pizzas in the past you can make pizzas more quickly today.
But what does “in the past” mean? If it makes little difference whether your experience is recent or distant, then you’re not forgetting much. If recent experience is much more valuable than experience years ago, forgetting matters a lot. By estimating the rate at which experience fades, researchers are estimating forgetting — even when productivity is increasing.
Economist Peter Thompson applied this logic to the production of Liberty Ships during the second world war to estimate that shipyards lost between 3 and 6 per cent of their knowledge every month. That amounts to a loss of roughly 40 to 50 per cent a year. This was an estimate for shipyards that were becoming increasingly productive, showing that it is possible to estimate a rate of forgetting even when the net effect is one of learning. In fact, the study of Liberty Ships was motivated by the observation that during the war these shipyards were experiencing an “average annual increase [in productivity] of about 40 per cent”.
But there are also cases in which forgetting outstrips learning. One such case is the L-1011, a commercial midsize aircraft introduced by Lockheed in the 1970s.
The L-1011 TriStar was designed to be the plane of the future. It had wide aisles, reduced noise in the cabin, and an autopilot system that allowed it to complete the first autonomous cross-country flight in the US on May 25 1972.
But its commercial performance was not as impressive. The aircraft received relatively few orders, and the team behind it changed constantly. This translated into costs increasing with each unit, instead of decreasing, as one would expect normally in manufacturing. Between 1975 and 1982, the cost of producing an L-1011 grew from $20mn to $29mn.
Still, there are situations in which forgetting can be beneficial. Especially in situations that require a good deal of unlearning. This is well exemplified in a story involving IBM in the 1980s.
Back then, IBM was known for producing mainframe machines with impressive-sounding names such as the Datamaster. But as personal computers burst on to the scene, some executives became worried about disruption. This motivated Bill Lowe to convene Project Chess. This was a fast-paced project involving a team of 12 engineers who would work out of IBM’s offices in Boca Raton, Florida, instead of the headquarters in Armonk, New York. This team was also authorised to source components from outside vendors; a bold move for a company used to working with internal knowledge. But the gamble succeeded. In January 1981, the team of engineers revealed a working personal computer — something that would have been hard to accomplish for a team involved in the daily worries of Armonk, New York.
We are fortunate enough to live in a world where learning often outstrips forgetting. But that does not mean we should ignore the weaknesses of our collective memory.
Sustaining an economy’s level of expertise is not an effortless endeavour. It demands continuous investment. In this analogy, knowledge is no longer a shark, but a muscle that atrophies as soon as you stop lifting weights.
So retaining an economy’s ability to build large or complex infrastructure projects requires constantly exercising that muscle. That means advanced economies must treat critical infrastructure, not as exceptional undertakings, but as routine forms of knowledge maintenance.
Knowledge economies are not ladders we climb once, but treadmills that will knock us down if we stop running. For policymakers, this means that resilience is not achieved through stockpiles or one-off investments. Preserving critical knowledge means never fully stopping to build metros, power plants or aircraft. You only get to keep the knowledge you use. The cost of maintaining knowledge may seem high, but the cost of losing it may be much higher.
Knowledge does not vanish because it is obsolete. It vanishes when it is not used.
César A Hidalgo is a professor at the Toulouse School of Economics and the director of the Center for Collective Learning, an international research laboratory with offices in France and Hungary. His most recent book is ‘The Infinite Alphabet’