Back in August, I cavalierly said that AI couldn’t design a car if it hadn’t seen one first, and I alluded to Henry Ford’s apocryphal statement “If I had asked people what they wanted, they would have said faster horses.”
I’m not backing down on any of that, but the history of technology is always richer than we imagine. Daimler and Benz get credit for the first automobile, but we forget that the “steam engine welded to a tricycle” was invented in 1769, over a hundred years earlier. Assembly lines arguably go back to the 12th century AD. The more you unpack the history, the more interesting it gets. That’s what I’d like to do: unpack it—and ask what would have happened if the inventors had access to AI.
If Nicolas-Joseph Cugnot, who created a device for transporting artillery over roads by welding a steam engine to a giant tricycle, had an AI, what would it have told him? Would it have suggested this combination? Maybe, but maybe not. Perhaps it would have realized that it was a poor idea—after all, this proto-automobile could only travel at 2.25 miles per hour, and only for 15 minutes at a time. Teams of horses would do a better job. But there was something in this idea—even though it appears to have died out—that stuck.
During the final years of the 19th century, Daimler and Benz made many innovations on the way to the first machine generally recognized as an automobile: a high-speed internal combustion engine, the four-stroke engine, the two-cylinder engine, double-pivot steering, a differential, and even a transmission. Several of these innovations had appeared earlier. Planetary gears go back to the Greek Antikythera mechanism; double-pivot steering (putting the joints at the wheels rather than turning the entire axle) had appeared and disappeared twice in the 19th century—Karl Benz rediscovered it in a trade journal. The differential goes back to 1827 at least, but it arguably appears in the Antikythera. We can learn a lot from this: It’s easy to think in terms of single innovations and innovators, but it’s rarely that simple. The early Daimler-Benz cars combined a lot of newer technologies and repurposed many older technologies in ways that hadn’t been anticipated.
Could a hypothetical AI have helped with these inventions? It might have been able to resurrect double-pivot steering from “steering winter.” It’s something that had been done before and that could be done again. But that would require Daimler and Benz to get the right prompt. Could AI have invented a primitive transmission, given that clockmakers knew about planetary gears? Again, prompting probably would be the hard part, as it is now. But the important question wasn’t “How do I build a better steering system?” but “What do I need to make a practical automobile?” And they would have to come up with that prompt without the words “automobile,” “horseless carriage,” or their German equivalents, since those words were just coming into being.
Now let’s look ahead two decades, to the Model T and to Henry Ford’s well-known quote “If I had asked people what they wanted, they would have said faster horses” (whether or not he actually said it): What is he asking? And what does that mean? By Ford’s time, automobiles, as such, already existed. Some of them still looked like horse-drawn buggies with engines attached; others looked recognizably like modern cars. They were faster than horses. So Ford didn’t invent either the automobile or faster horses—but we all know that.
What did he invent that people didn’t know they wanted? The first Daimler-Benz auto (still in a modified buggy format) preceded the Model T by 23 years; its price was $1,000. That’s a lot of money for 1885. The Model T appeared in 1908; it cost roughly $850, and its competitors were significantly more expensive ($2,000 to $3,000). And when Ford’s assembly line went into production a few years later (1913), he was able to drop the price farther, eventually getting it down to $260 by 1925. That’s the answer. What people wanted that they didn’t know they wanted was a car that they could afford. Automobiles had been firmly established as luxury items. People may have known that they wanted one, but they didn’t know that they could ask for it. They didn’t know that it could be affordable.
That’s really what Henry Ford invented: affordability. Not the assembly line, which made its first appearance early in the 12th century, when the Venetian Arsenal built ships by lining them up in a canal and moving them downstream as each stage of their manufacture was completed. Not even the automotive assembly line, which Olds used (and patented) in 1901. Ford’s innovation was producing affordable cars at a scale that was previously inconceivable. In 1913, when Ford’s assembly line went into production, the time it took to produce one Model T dropped from 13 hours to roughly 90 minutes. But what’s important isn’t the elapsed time to build one car; it’s the rate at which they could be produced. A Model T could roll off the assembly line every three minutes. That’s scale. Ford’s “any color, as long as it’s black” didn’t reflect the need to reduce options or cut costs. Black paint dried more quickly than any other color, so it helped to optimize the assembly line’s speed and maximize scale.
The assembly line wasn’t the only innovation, of course: Spare parts for the Model T were easily available, and the car could be repaired with tools most people at the time already had. The engine and other significant subassemblies were greatly simplified and more reliable than competitors’. Materials were better too: the Model T made use of vanadium steel, which was quite exotic in the early 20th century.
I’ve been careful, however, not to credit Ford with any of these innovations. He deserves credit for the biggest of pictures: affordability and scale. As Charles Sorenson, one of Ford’s assistant managers, said: “Henry Ford is generally regarded as the father of mass production. He was not. He was the sponsor of it.”1 Ford deserves credit for understanding what people really wanted and coming up with a solution to the problem. He deserves credit for realizing that the problems were cost and scale, and that those could be solved with the assembly line. He deserves credit for putting together the teams that did all the engineering for the assembly line and the cars themselves.
So now it’s time to ask: If AI had existed in the years before 1913, when the assembly line was being designed (and before 1908, when the Model T was being designed), could it have answered Ford’s hypothetical question about what people wanted? The answer has to be “no.” I’m sure Ford’s engineers could have put modern AI to tremendous use designing parts, designing the process, and optimizing the work flow along the line. Most of the technologies had already been invented, and some were well-known. “How do I improve on the design of a carburetor?” is a question that an AI could easily have answered.
But the big question—What do people really want?—isn’t. I don’t believe that an AI could look at the American public and say, “People want affordable cars, and that will require making cars at scale and a price that’s not currently conceivable.” A language model is built on all the text that can be scraped together, and, in many respects, its output represents a statistical averaging. I’d be willing to bet that a 1900s-era language model would have access to a lot of information about horse maintenance: care, disease, diet, performance. There would be a lot of information about trains and streetcars, the latter frequently being horse-powered. There would be some information about automobiles, primarily in high-end publications. And I imagine there would be some “wish I could afford one” sentiment among the rising middle class (particularly if we allow hypothetical blogs to go with our hypothetical AI). But if the hypothetical AI were asked a question about what people wanted for personal transportation, the answer would be about horses. Generative AI predicts the most likely response, not the most innovative, visionary, or insightful. It’s amazing what it can do—but we have to recognize its limits too.
What does innovation mean? It certainly includes combining existing ideas in unlikely ways. It certainly includes resurrecting good ideas that have never made it into the mainstream. But the most important innovations either don’t follow that pattern or make additions to it. They involve taking a step back and looking at the problem from a broader perspective: looking at transportation and realizing that people don’t need better horses, they need affordable cars at scale. Ford may have done that. Steve Jobs did that—both when he founded Apple and when he resuscitated it. Generative AI can’t do that, at least not yet.
Footnotes
Sorensen, Charles E. & Williamson, Samuel T. (1956). My Forty Years with Ford. New York: Norton, p. 116.