Selection by failure:
a methodical critique of the linear model of innovation
Empirical Risk & Innovation Studies · field notes, not peer-reviewed
Summary: Taleb's central idea, which he formulates in dozens of different ways throughout his work in order to hammer the point home and eliminate any possibility of misunderstanding, is that real innovation does not come from the conscious intention to discover something specific, but from repeated and systematic exposure to chance, to accepted error, and to relentless selection by the real world.
1The author and his method
Nassim Nicholas Taleb, in his book Antifragile published in 2012, spent several hundred pages methodically demolishing the illusion of the linear model we've just described. This intellectual violence is not gratuitous. It stems from the fact that Taleb is neither an academic economist sheltered in a university, nor a consultant who makes a living perpetuating reassuring myths. He's a former derivatives trader who spent twenty years betting real money on hunches he didn't always fully understand at the moment he formed them, and who therefore developed a visceral allergy to fine theoretical speeches that collapse at the first contact with reality.
2The central hypothesis
This idea is not merely counterintuitive. It is directly opposed to everything taught in management schools, to everything told in the hagiographic biographies of successful entrepreneurs.
And Taleb adds a crucial nuance: in the vast majority of rigorously documented historical cases, it is not the theorists, the academic researchers, the credentialed experts who produce the major innovative breakthroughs. It's the practitioners, the tinkerers, the artisans, the entrepreneurs who try something without institutional permission, who fail quickly and cheaply, who try again with variations, who gradually refine through trial and error until something finally works in a reproducible way.
3The role of theory
Theory, in this scheme, is not useless. But it comes after, not before. It observes what has survived the natural selection of the market, it formalizes the observed regularities, it rationalizes the process in hindsight to make it transmissible and teachable. But it does not create from nothing.
4Historical case studies
Take the example of the jet engine in aviation. We've all learned, or at least we all think we know, that the development of jet engines follows logically from advances in thermodynamics, fluid mechanics, and materials science, in short, that scientific theory preceded and enabled the technical application.
The technology historian Philip Scranton1 of Rutgers University has shown, with archival documents to back it up, that this is rigorously false. We built and used jet engines for decades in an entirely experimental way, through successive adjustments, the transmission of tacit know-how between senior and junior engineers, trial and error on real prototypes, without anyone truly understanding the complete theory governing their operation.
The engineers who worked at Rolls-Royce, General Electric, or Pratt & Whitney in the 1940s and 1950s knew how to bend certain parts, adjust the airflows, calibrate the combustion temperatures to get the desired thrust without the whole thing exploding in flight. They passed this knowledge on in an almost artisanal way, through direct apprenticeship, observation, imitation, and rules of thumb like "if you see such-and-such vibration, then tighten that bolt a quarter turn." The formalized theory came much later, to satisfy the institutional need for academic rationalization.
The Gothic cathedrals of medieval Europe offer perhaps the most spectacular illustration of this principle. These structures seem geometrically impossible, feats of engineering that should logically require sophisticated mathematical calculations. And yet they were built by master builders who, for the vast majority of them, knew no formal mathematics.
The French historian Guy Beaujouan2, a specialist in medieval science, established that before the 13th century, there were probably no more than five people in all of Western Europe capable of performing a simple arithmetic division. So how did they build structures that are still standing eight hundred years later?
Through the accumulation of rules of thumb, the oral and visual transmission of proven techniques, learning by direct observation and imitation, and safety margins built in out of caution rather than calculated optimization. Villard de Honnecourt, a 13th-century French architect, left notebooks filled with practical instructions: "to visualize a triangle, think of a horse's head" or "for this vault, use such-and-such wooden template that you make this way." No equations. No formal geometric proof. Just gestures that work because they were tested, refined over dozens of building sites, because the ones that didn't work produced memorable collapses that everyone learned from.
5A rule of institutional engineering
The general rule that Taleb invites us to internalize, and that should guide any innovation policy worthy of the name, is this: massively tolerate local errors (many acceptable small losses, distributed across many independent actors), but absolutely prevent systemic errors (a single large centralized error that destroys the whole system).
max ∑i εi local subject to ε systemic → 0 (1)
| Regime | Allocation | Diagnosis |
|---|---|---|
| Decentralized | A thousand startups that fail quickly after burning through 50,000 euros each. | Healthy |
| Centralized | A national plan that concentrates 500 million euros on a single project. | Catastrophic |
France and Europe have historically excelled at the second approach (large centralized programs, national champions selected by committees, massive funding for a few projects deemed strategic by experts) and chronically underinvest in the first (a proliferation of small decentralized attempts, social acceptance of quick failure, patient but demanding capital distributed widely).
1 P. Scranton, Rutgers University, archival analysis of the empirical development of turbojets (1940s-1950s).
2 G. Beaujouan, history of medieval mathematics; notebooks of Villard de Honnecourt, 13th century.