Could “AI” end the post-industrial economy?

I’m prepared to be wrong, but my present best guess is that the “AI” bubble has now reached its peak and is about to unwind. It’s not out of the question that this will trigger a GFC-like financial crisis given the stupendous amounts of money that have been poured into it. If so, the underlying cause will be much the same as the GFC, i.e., far too much money chasing far too few returns.

The pattern that we have seen around the rollout of LLMs is not new, and is worth exploring in more detail, since it may very well be on its last legs.

Entire books have been written about Ford and the huge paradigm shift he brought about. Among other things, he set the stage for the post-war consumer economy. He was also an advocate for early forms of “social justice”. That concept was not always the province of university communists. In fact, even in our time, much of the funding still comes from various billionaires, including the charitable foundation that Henry Ford’s son set up back in the day.

For our purposes here, we’ll focus on just a few key properties of Ford and his work, so that we can compare the system that created the automobile to the one that has created “AI”; industrial capitalism vs post-industrial capitalism.

Ford was not alone in beginning his career on the production side of the business world. Many of the great “capitalists” of the 19th and early 20th centuries were also engineers. Furthermore, what we now think of as public utilities were largely built out by such private engineers. That was true even here in Australia where the market was tiny compared to the US.

The second point to make about Ford’s car manufacturing business was the manner in which he achieved success. He had a number of failed attempts before hitting the winning formula with the Model T. Although relatively large sums of capital were required to manufacture cars, these initial attempts were fairly small bets where Ford himself assumed much of the financial risk.

More importantly, once the Model T took off, Ford expanded the business by reinvesting profits earned, not through speculative capital. This was no accident. Ford was highly critical of the banking industry and saw the concentration of power in the hands of banks as a disincentive to real innovation and growth in favour of profit-taking. That’s also why Ford quickly assumed full control of his company by buying out all other original investors and becoming sole owner. This made him independent of financial institutions and the stock market.

The third crucial point is that Ford was one of the enlightened capitalists of his era who realised that paying workers the lowest possible wage was counterproductive because it created employees who didn’t have enough money to buy the products that were being manufactured. He shocked the business world by introducing the “five dollar wage” at his company, which was almost double what the average worker was earning at that time. This enabled him to snap up the most talented workers, who he further rewarded by giving stock options to the best performers. Later, he introduced a five-day work week.

The combination of these and other practices formed the basis of the economic strategy that drove capitalism for most of the 20th century. Increased productivity led to higher wages and reduced working hours, thereby giving consumers the time and money to purchase the products being created, thereby driving sales, thereby allowing wages to increase, and so on. At the core of this model were efficiency gains driven by the engineering talent that Ford was able to attract and nurture.

Of course, the power of productivity had been known about at least as far back as Adam Smith. What made Ford successful was that he found a way to apply it to complex manufacturing. What also separated Ford was his willingness to share the gains with the workers, in contrast to the old capitalists who tried to screw every last penny out of the labour force.

With these basic elements of Ford’s model sketched out, we are now ready to understand some of the major differences between the industrial economy of the 20th century and the post-industrial economy that has produced things like “AI”. One of the main differences is the way technologies such as LLMs have been developed and financed.

Remember that Ford tried and failed several times to release a viable car to market and then, when he finally succeeded, built his business from profits earned, not from speculative capital. With LLMs, we have an almost exact inversion of that paradigm. LLMs are not a product that have any kind of real consumer demand behind them. Nobody has purchased an LLM. Rather, they have been made freely available. This has only been possible due to the enormous amounts of speculative capital that have been thrown at them. In the post-industrial model, products are not required to be profitable in order to gain investment. The assumption is that they will be monetised later once a market has been created.

The beginnings of this new kind of model can be found in the early days of the home computer industry. Nominally, that industry had a great deal in common with Henry Ford’s kind of capitalism. Bill Gates, Steve Jobs, and Steve Wozniak were all engineers and enthusiasts who mucked around with the tech in their spare time. They invested their own money into their early ventures and absorbed the cost of failure. When they finally stumbled across the winning formula, home computers saw similarly explosive growth to the Model T Ford. All of this looks like a repeat of the old paradigm.

There were at least two big differences, however. Firstly, both Apple and Microsoft were launched on the stock market very early on. This made Gates, Jobs, and Wozniak very rich, but it also meant they lost control of their companies. In fact, Jobs and Wozniak were famously elbowed aside shortly after Apple was publicly listed. That is the price to be paid for selling out to capital. It’s a price that Henry Ford explicitly avoided.

The second big difference was that sales were not really driven by the computers themselves, since only people who knew how to code could use them. Instead, it was the pre-installed ready-to-use software that became the key selling point. For example, Apple got its initial popularity by bundling its computers with one of the first spreadsheet programs. It would later win the high-end of the market by incorporating graphic design and printing software. It turned out that software was the main game, not hardware. Over time, this led to a pattern of offering software free or almost free in order to win market share.

With the arrival of the internet, there was a further abstraction away from hardware in favour of software. One result of this was that websites were offered for free and then monetised later. Capital also got involved much earlier in the process. Thus, Zuckerberg barely needed to invest any money in Facebook before the investment firms came knocking on his door to give him a truckload of money. Google had a very similar story in its early days.

Home computers had followed the same pattern as early automobiles in that they started out expensive but got cheaper as economies of scale and technology improvements came along. By contrast, the internet has followed an opposite pattern. Because the price of nearly every website started out at zero, it was impossible for it to get cheaper. What happened instead was that quality degraded over time once the market was captured and the investment funds started to monetise. This is what Cory Doctorow has usefully called the “enshittification” paradigm. It is, in fact, a form of market failure based on exactly the kind of predatory financial practices that Henry Ford warned about a hundred years ago.

If we think back to the original paradigm that Henry Ford operated under, another core component was productivity gains. There can be no doubt that the computer revolution led to massive efficiency improvements, and so this fitted the old pattern of industrial capitalism. However, the internet has been qualitatively different in this respect. Yes, it certainly saves time to have a search engine that returns the exact results you want, but this useful feature rarely translates into meaningful increased workplace productivity. Google’s revenue came not from productivity gains but from advertising resulting from market capture that included the entire globe, not just local markets.

We can see, then, that the internet has broken with Ford’s paradigm in several ways. Firstly, it has not been about productivity gains. It turns out that increasing the speed of information does not improve its quality. Most of the time, it degrades it. Secondly, the internet has not been about creating a virtuous cycle where prices are lowered over time because the starting price of any website has been zero. Therefore, there has been no incentive to increase quality. On the contrary, predatory finance practices only lead to enshittification.

It is against that background that LLMs have arrived on the scene. Perhaps ironically, one of the main marketing angles for LLMs and “AI” is that they will improve productivity. While there have been some promising results in relation to machine learning more broadly, there is very little evidence that LLMs lead to efficiency gains. The strength of LLMs turns out to be a weakness in this regard. The technology does a pretty good job of impersonating human language, but human language is not optimised for efficiency. It is flexible and context-dependent.

The result is that, for any task of reasonable complexity, it is very difficult even to craft the right instructions to give to an LLM to ensure you get the answer you want. Moreover, even if you fine-tune the instructions, it is in the nature of LLMs to return somewhat random answers, including answers with basic factual errors. There was amusing example of this recently when the consulting firm Deloitte got busted using LLMs in a report for the Australian government. The LLM did its fairly common trick of “hallucinating” facts, i.e. making stuff up.

The whole reason why the computer revolution led to such huge productivity improvements was because computers give the right answer every time as long as they have been programmed correctly. It was the speed and accuracy of computers that led to efficiency gains. LLMs might be quick, but they produce results that cannot be relied on.  

That would be bad enough, but it gets even worse when you consider that LLMs require about an order of magnitude more electricity to produce an inferior result. This is where the whole narrative around LLMs becomes almost schizophrenic because, in order to make them in any way viable, you need a massive build-out of energy infrastructure, and you need that at the exact same time that we’re making energy more expensive and less reliable via solar and wind. In short, LLMs need far more electricity to produce less accurate computation. That’s not a very good deal. In fact, it’s the opposite of the dynamic that Henry Ford created whereby productivity gains led to a reduction in price.

We are now living in a very different kind of “capitalism” from the one that Henry Ford helped to create. The system is no longer predicated on first creating a product that people want to buy and then driving down prices through economies of scale and productivity gains. Rather, the LLM craze is a finance-first phenomenon. It involves throwing huge sums of money at new technologies in the hope that they might be profitable at some undisclosed time in the future. In practice, this involves creating the illusion of a viable product by offering it for free and then enshittifying it later. Along the way, a tsunami of manufactured marketing buzz is needed to keep the investment money flowing.

In truth, the real customers for these kinds of “products” are not consumers at all but bankers and other interested parties who choose winners mostly on ideological grounds. Furthermore, because of the huge sums of money involved, only products which can potentially generate enormous revenues are ever pursued, since that’s the only way to generate returns that could justify the initial investment. That’s the reason why the marketing hype always stresses that the product is going to “change the world”. The change is always just around the corner, too. It has to be to ensure the release of the next round of investment funding, which is the only thing that keeps the whole shebang going. Ford built his empire on profits. LLMs are not being built on profits because there are none.

The LLM craze is following a pattern we have seen before. About ten years ago, the exact same fluff was being peddled about self-driving cars. I distinctly remember friends and colleagues talking as if the age of car ownership were a thing of the past. They genuinely believed that all cars on the road would be self-driving within a period of months or years. Well, it should be obvious that self-driving cars did not take over the world. The revolution didn’t happen. Once the marketing money dried up, the whole thing was quickly forgotten about and we moved on to the next craze.

The self-driving car flop is just one of a pattern of failures of post-industrial economics. Remember when the blockchain and NFTs were going to change the world? All that happened instead was that Sam Bankman-Fried and his buddies got to have amphetamine-fuelled parties in the Bahamas before going to jail. Another fraudster who ended up in jail was Elizabeth Holmes, peddling one of the seemingly infinite number of biotech boondoggles that appear and disappear just as quickly, unless a “pandemic” comes along to help the marketing effort (hello, mRNA vaccines). The list of these kinds of failure could go on and on.

A big part of the reason for these failures is that these products are not chosen by the market but by “the system”. Gone are the days of the engineer iterating towards the right solution in his spare time. The process is no longer driven by independent-minded men like Henry Ford but by whole industries of finance and marketing whose specialty is the creation of bullshit.

What differentiates the current “AI” craze, I think, is the huge sums of money that have now flown into it. Unlike the other boondoggles listed above, which were able to be quietly put to bed without anybody really noticing, I doubt the same can happen with the AI bubble.

The promise that “AI” will “change the world” might turn out to be true, just not in the way that the true believers think. If the AI bubble really does trigger another GFC, there is a genuine chance that this will be the last straw, because it’s not clear how much longer can we afford the post industrial economy.