Epistemic standards for “Why did it take so long to invent X?”
by Jason Crawford · March 1, 2020 · 6 min read
In seeking to understand the history of progress, I keep running across intriguing cases of “ideas behind their time”—inventions that seem to have come along much later than they could have, such as the cotton gin or the bicycle. I’ve started collecting a list here, and will update that page with new analyses as I find them.
Debates on these questions sometimes oddly devolve into arguments, with people fruitlessly talking past each other (although if people on the Internet can argue over how many days are in a week, I guess we can argue about anything). So I want to comment on how we think about such cases and what the standards for evidence are.
To start, there is a need for precision. For example, take the steam engine. There was a thing in antiquity called an engine, that used steam: “Hero’s engine”, also known as the aeolipile. Some people see this and conclude that “steam engines existed in the 1st century” or that there’s no reason ancient Rome couldn’t have used this widely in industrial applications.
This is a mistake. The aeolipile is nothing like the steam engines of the 18th century and later: it’s a turbine, which means it is rotary, rather than using the reciprocating (back-and-forth) motion of a piston, as in Newcomen’s engine. Why does this matter? Because the aeolipile doesn’t generate enough torque for practical applications—one analysis says that Watt’s engine generated a quarter of a million times more torque.
Even before finding that analysis, I had an hunch it would be the case—indeed, that’s how I knew to search for “aeolipile torque”, which led me to that link on the first page of Google results. My intuition was based on a few things. First, if a simple, primitive turbine like the aeolipile could be put to practical use, why didn’t anyone reinvent it in the 18th or even 17th century? Why did Newcomen, Watt, and others focus on much more complicated piston engines? They were smart people and were obviously working hard on the problem, it seems impossible that such a simple solution would have escaped all of them. Second, the aeolipile is small—Hero’s sketch above shows it sitting on a table—but Newcomen’s engine was very large, to the point where a separate shed would be built to house one. Why did the engine have to be that large if a tiny one would do?
Again, a precise understanding of each invention will uncover relevant details like this. A concept, such as “an engine (of any type) that uses steam (in any way)”, is not enough.
This example also illustrates a second principle: practicality matters. A device that works in theory, but is too underpowered, inefficient, expensive, or unreliable, might as well not exist for practical purposes. It must work not only for a demonstration, but for real, human, economic needs, in the context of consumers’ lives, industrial processes, or business operations. Because of this, a difference in degree can become a difference in kind, when an invention crosses a threshold of practicality.
As a side note, this is why it’s perfectly accurate to say that Edison’s lab invented the light bulb, even though there were other light bulbs before it: they were too expensive (e.g., using platinum filaments), or they burned out quickly and thus needed to be replaced too often. In my opinion, it’s redundant to say that someone invented “the first practical X”—this is the same as saying they invented X. To invent something is to invent a practical version of that thing. If your “invention” is impractical, it’s just a demo or prototype. This can be useful to test ideas or to communicate possibilities, but it’s the practical inventions—the ones that actually remove all the obstacles to widespread use—that move history.
Another example of this is the computer. The computer was invented by J. Presper Eckert and John Mauchly at the University of Pennsylvania; their first model was the ENIAC, completed in 1945. It was a breakthrough because it was the first fully electronic computer, and this made it much faster than previous attempts, such as the IBM Automatic Sequence Controlled Calculator (ASSC, aka the Harvard Mark I), which was electromechanical, using magnetic relays. Based on the speeds for these machines given by Wikipedia, the ENIAC was about 600 times faster than the ASSC at division and over 2,000 times faster at multiplication. (The ENIAC was also over 2,000 times faster than a human using a mechanical calculator at calculating a ballistic trajectory, implying that the ASSC was probably not much faster than a human.) The ASSC was an interesting demo that got some press; the ENIAC was the machine that ignited the computer revolution. Again, a difference in degree becomes a difference in kind. (Going even further back, other predecessors such as the Atanasoff–Berry computer or Konrad Zuse’s Z3 were also much slower than the ENIAC, and had other practical limitations. And Babbage’s “computer” was only a concept with an unfinished design that could never have been built with the technology of the day—which is why, despite my respect for his genius, I cannot regard Babbage as the inventor of the computer, any more than da Vinci was the inventor of the helicopter.)
If you want to argue that something could not have been invented before a particular “gate”, I regard this kind of history as the epistemic gold standard: strong economic motivation (and in the case of computers, military motivation in the context of WW2); multiple prior attempts, including some completed projects that were working and reasonably well-publicized; a measurable difference in a key practical dimension (in this case, speed); and an enabling technology that made a significant difference along that dimension (in this case, approximately three orders of magnitude). For this reason I confidently say that the computer—again, it’s redundant to say “practical computer”—could not have existed before the invention of the vacuum tube amplifier in 1907. (It’s plausible, but less certain to me, that it could not have existed until even later, when improved, more reliable vacuum tubes were invented. Plausible, because the ENIAC used over 17,000 tubes, and reliability was a concern among the engineers; less certain, because I don’t know of any failed attempt at building a fully electronic computer with less-reliable tubes, and because some statements from engineers indicate that reliability was less of a problem than feared, particularly if the tubes were operated continuously, to avoid thermal stress. So it might be that, after 1907, all that was holding back the engineering was a key insight.)
On the other hand, if you want to argue that something could have been invented much earlier than it was, you have to do better than glancing at its high-level concepts or components. You need to rigorously examine every part, material, and manufacturing process, and rule them all out as gating technologies. Any detail, even a minor one, can become crucial—especially when we remember that inventions need not only to work but to be practical, which includes performance, reliability and cost. As an example, in my analysis of the bicycle, I described the first proto-bicycle, known as the “draisine” or “Laufmaschine”, as being made of wood with iron tires—both ancient technologies. However, Nick Szabo pointed out to me that it reportedly used brass ball bearings, a much newer technology, which might have been essential to reduce friction.
A related question: how surprised should we be that it took X years for invention Y after enabling technology Z? Inventions do not spring forth immediately upon becoming possible: ideas and information take time to spread, experiments are required, funding must be secured, laboratories organized, materials obtained; and at end of the day all this is performed not by automata or some clockwork mechanism, but by unpredictable individuals with their own vision, inspiration, hopes and fears, operating in complex network of teams, contracts, partnerships, and other social structures. Even in the best of circumstances, a gap of a decade or more from a key enabling technology to the commercial release of an invention is not surprising; if the enabler is a scientific discovery, two or three decades does not surprise me. And chance can intervene—the path to an invention can be derailed by a sudden disease, a financial panic, a war.
In general, I think we should be more surprised at long gaps for inventions that have obvious, predictable impact on major industries. For this reason, the cotton gin and the flying shuttle are more compelling gaps to me than the wheeled suitcase, role-playing games, or the bicycle, which merely offer convenience or entertainment. I think we should also expect longer gaps in places and times that had lower population, less education, less economic surplus (to fund R&D), fewer or less effective financing mechanisms (such as venture capital), less political stability, etc.
My model for this is that innovation, at a societal level, is a stochastic process, with some parameters set by the environment and others by the particular invention in question. The more “pressure” there is to solve a problem (economic motivation), and the more opportunities there are for it to get solved (educated, inventive individuals or organizations, with time, space, materials, and funding, in a context of good legal institutions and political stability), the sooner you expect the leap to be made and the shorter a gap. In the limit, you get simultaneous invention, of which there are many stories (although some are overplayed, in part for the reasons discussed above regarding what counts as an “invention”). Some economics grad student could probably get a PhD thesis out of formalizing this model and fitting the parameters to data—both to quantify the “inventiveness” of a given place and time, and to identify outlier inventions that were truly, measurably, “behind their time.”
