Too Big to Succeed
Why the AI industry has probably already failed
On a very windy October day in 1983, a group of engineers and utility executives gathered on the coast near Hamburg, Germany to witness the maiden test of Growian. It was the biggest wind turbine built to date, coming in at a full 3 megawatts. The turbine was supposed to serve as forerunner to the coming renewable energy future, a gargantuan testament to Germany’s engineering prowess. Too bad it ended up being a piece of shit.
Growian, a contraction of grosse Windenergieanlage (literally “big wind turbine”), failed after a mere 420 hours, or the equivalent of 17 and a half days of continuous operation. No crane was tall enough to assemble the massive turbine, so parts had to be made small enough to be pulled up the interior of the units tower. And the resulting welded joints of the turbine’s housing proved to be too weak handle the mechanical stresses that electricity generation produced.
Energy scholars Craig Morris and Anne Jungjohann, suspect intentional corporate sabotage, given that at least board member at the utility stated that “we need Growian to prove that wind power won’t work.” However, wind turbine prototypes across the pond were equally pathetic. One should be skeptical of stories that involve malice and conspiracy when plain old foolishness is the more likely root cause.
In America, almost a half a billion dollars and all the expertise that engineers from both NASA and firms like Boeing and General Electric could muster ended up producing a full decade of wind turbine failures. Many prototypes required serious maintenance after as a little as 40 hours. None operated for more than a few years. Nearly all were gigantic, especially so for the era. Based on these cases, a person in the 1990s could be forgiven for believing that wind energy was a failed technology.
But in Denmark the wind turbine was transformed into something commercially viable. The path to success here didn’t involve millions in federal R&D or aerospace engineers using computer models and the latest material science. The motive force for wind energy innovation came from a seemingly unlikely set of firms: farm machinery builders.
Known today as the biggest turbine manufacturer in the world, the Danish firm Vestas started out making milk coolers and small hydraulic cranes. Although it is tempting to see airplanes as the closest analogue to a wind turbine (given the need for propellers and the importance of aerodynamics), hands on experience with agricultural implements proved decisive for Danish companies. No matter how good a turbine’s power to weight ratio or its predicted energy capacity, if it isn’t resilient, it’s worthless. German and American engineers chased technical sweetness, a turbine that might have even had a fighting chase at getting airborne, when what they needed was something that was as robust as a tractor.
The innovation process in Denmark was also far better organized to support ongoing learning. Designs had to be proven on a government test stand prior to licensing, which helped provide performance feedback and access to engineering expertise to firms still new to the turbine business. And the companies shared both successes and failures through an industry journal. And most importantly, in the absence of massive influxes of other people’s money, these firms had to make things that they could actually sell.
Moreover, government policy ensured a place for wind turbines within the energy grid. Regulations were developed to ease their integration. Feed-in tariffs and other relatively light forms of subsidy were implemented to help commercialize them. All these factors put together not only turned wind turbines from a small-scale farm implement into a serious energy technology but also resulted in Danish companies absolutely dominating the market in its early years.
One common truism is that while history doesn’t exactly repeat itself, it certainly rhymes. As a number of economic indicators now signal a coming burst of the AI bubble, I couldn’t help but hear the word Growian ringing in my ears.1 If the bubble indeed bursts, the last five years of AI development should be seen in similar terms as the American and German failure to turn a theoretically plentiful resource into a practical reality: an ongoing act of pompous hubris, of arrogant stupidity.
We can see it in the innovation strategy being employed. Propped up by billions in venture capital, AI firms have single-mindedly chased narrow technical measures of success. Like the builders of the Growian turbine or its equivalents in America, they have acted as if they believed that an AI system only needs to be big, fast, and comprehensive enough, then its deployment and integration across the planet’s sociotechnical systems is basically a done deal: Build Skynet and it takes over the world.
Just a little knowledge of the history of innovation shows us that the technological advancements are more than mere technical achievement. They’re a matter of reweaving currently dominant sociotechnical systems so as to make way for new technologies. Thomas Edison’s genius didn’t lie in superior knowledge or analytical ability, he sought to develop or acquire, and then deploy, the technological infrastructure and regulatory policies that enabled electric technologies like the light bulb to easily become a part of citizens’ everyday lives. One does not simply build a technology and think the rest will come, at least if you want to be successful.
But that is exactly what the main AI players are doing.
What should have been done instead? Or what should AI advocates do after the bubble bursts and we likely land in yet another AI winter?
Rather than make yet another spectacular gamble on artificial general intelligence, the industry would be better served by a more humble and incremental approach. Discover the problems that AI truly help solve and develop small marketable products that can address them. Commercialization shouldn’t be an afterthought but one of the initial steps. It’s such basic UX thinking (find out what the user needs and wants before building a product) that I have a hard time not wondering if there has been an unreported epidemic of serious head injuries among AI leaders, an ill-fated billionaire rager in Aspen.
In my conversations with colleagues, it’s become clear to me that AI won’t be anything more than a hinderance to universities in the near future, simply because Silicon Valley developers had never bothered to understand education as a sociotechnical system. I would immediately petition my department chair if readily available AI products were explicitly designed to enhance students’ learning, to make them better at exercising expert judgement, rather than to give them an alluring escape from the frustrating uncertainties of doing their own work.
As Colin Garvey pointed out last week, what we wanted was artificial personal assistants. Instead the AI industry has given us digital sycophants that try to do everything possible, including lie to us, to keep us forever distracted, forever engaged.
Yet, as we’ve noted in past posts, the Tech world isn’t set up to reliably solve people’s actual problems. Corporate leaders believe in moving fast and breaking things. They want to monopolize new market segments in order to make monopoly rents. Whether it actually makes citizens’ lives better is seemingly irrelevant. The trouble is that in many cases what actually breaks is the technology’s own chances for a sustainable future. A hubristic vision of progress is ultimately self-stultifying, delaying broadly shared advancement rather than accelerating it, as countless dollars get wasted on failed technological megaprojects—and the rest of us are left sifting through the wreckage when it all goes to shit.
Without a set of brakes (and a steering wheel) on the overenthusiasm and the discretionary power of venture capital, we’re seemingly destined to hear the rhymes of Growian over and over again.
You could call it GroKIan, which stands for grosse KI-Kackanlage (“big AI crap system”)
 | A guest post by
|