What if Generative AI Signals Decline and Stagnation, Not a Productivity Boost?

I work in an academic field called science and technology studies—or STS—which is focused on analyzing the contingent emergence, development, and changes in scientific fields and technological innovation—what we call “technoscience.” One of the founding premises in STS is that technoscience is configured by social, political, and economic drivers—that is, technoscience isn’t driven by some internal logic of increasing performance, efficiency or productivity. 

For many people, this is quite a radical starting point. And the same people can often take offence at the very idea that technoscience (which seems to be the epitome of objectivity and neutrality) is actually a social construction. Even the notion of “social construction” is misunderstood—it does not mean that it’s all in our heads!

The point of STS is not to denigrate the science and technology we see around us but rather to understand why we have the technoscience we have at any given point in time and in any given place. There are now numerous studies of the highly contingent nature of scientific research and technological developments, from mathematics through physics to biotechnology and now artificial intelligence. And I’d encourage you all to explore STS, as it’s become a very vibrant field.

How is STS Useful for Understanding Generative AI?

STS lets us get beyond the hype machine surrounding generative AI, especially the promotion, advocacy, and boosterism of it by highly interested parties—by which I mean those who’ve invested in generative AI or work in the sector or have jumped on the bandwagon as a great way to attract funding or attention. Basically, those whose livelihoods are dependent upon it being adopted widely.

There are numerous previous examples of other amazing technoscience which can do equally incredible things, but which never take off for one reason or another. I’ve been studying technoscience long enough to have witnessed several overhyped technologies, from biotech through biofuels to blockchain and the metaverse. They were meant to transform the world, but they didn’t. The point to emphasize is that the success (or not) of technoscience is highly social and highly contingent because technologies are socio-technical systems. Whether a particular technology will change the world is not down to technology alone. That means we have to understand the context in which it emerges. 

My starting point for analyzing generative AI is to remember that it’s not actually an autonomous intelligent system; with generative AI, there is no weighing up of facts (and values) and then human-like thinking or deciding on the basis of those facts (and values) what to do next. As Bender and colleagues point out, generative AI is better understood as a “stochastic parrot”—a system that makes a series of probabilistic selections based on a training dataset. And to do this, especially at scale, requires an array of socio-technical devices, infrastructures, relations, processes, and practices behind it. For example, generative AI is entirely dependent upon massive computing capacity for processing and data storage. Current uses of things like ChatGPT are not profitable and will require significant scaling up to make money, which will require continuing levels of massive investment in computing capacity, dwarfing what Big Tech firms are currently spending. 

This is why I don’t think generative AI will lead to a productivity boost or will even become economically profitable. There are some great discussions about this produced by venture capitalists, investment banks, journalists, and many, many, many others. 

Is Generative AI Overhyped?

My take on why generative AI is such a hot topic and interest right now is pretty simple but requires a little grounding in theories of socio-technical change. Joseph Schumpeter famously characterized capitalism’s dynamism as the “gales of creative destruction” that enable capitalism to overcome recurrent crises. Schumpeter used creative destruction to conceptualize the economic and technological changes that lead to both growth and decline over 40- to 60-year cycles. Simply put, as a new techno-economic paradigm emerges, it leads to massive investment and expansion before hitting maturity and then slow decline; the next cycle then gets kickstarted. These cycles should not be thought of as natural, however, in the sense that there is an inherent or intrinsic logic underpinning the expansion and decline of a particular technology. Human decisions, actions, and beliefs drive societal transformations, especially economic logics.

Critically, if we look at the current techno-economic paradigm, which can be defined as the rise of a digital economy (e.g. information and telecommunications, internet, smartphones, etc.), we can see all the signs of maturity, stagnation, and decline setting in. Productivity has been falling for decades, besides a short upward blip in the early 2000s; societal investment is stagnating; R&D spending has not translated into rising GDP; and there is a range of societal crises we’ve been unable to resolve, from climate change to inequality to housing affordability. Despite this, a small number of companies, largely American Big Tech firms (e.g. Amazon, Alphabet, Microsoft), are investing billions and billions in expanding computing capacity. And they’re doing so at a higher and faster rate over time.

Much of this expansion in computing capacity is understood to be driven by generative AI, which requires significant computation to process massive training datasets. However, these Big Tech companies started expanding their digital infrastructure investments well before the likes of ChatGPT or others were launched—this is evidenced in their annual reports, where you can see the investments they’ve been making in their tangible asset base. My take on this is that the rise of generative AI can be seen as a symptom of the dying stages of the digital economy. If you read their earnings calls or financial reports, Big Tech firms explain that they’re extracting more from their capital investments by extending the lifespan of digital infrastructure as well as investing in new capacity and data centres. I’d argue that this is a strategy to offset the upcoming losses they expect from their decaying and depreciating asset base or the potential stranding of their asset base as a new techno-economic paradigm emerges. Much of this asset base is under threat of irrelevance as their services and products become increasingly useless, whether that’s poorly performing search engines, ad-riddled social media, or scams and counterfeits swamping e-commerce sites, alongside a slew of allegations about monopolistic and other actions. 

As the management guru Peter Drucker pointed out a long time ago, at a certain point in the innovation cycle, businesses stop investing and start reaping the returns on their investments. Right now, though, Big Tech can see the endpoint coming to capture returns on their previous investments, but they can also see that while we wait for the next techno-economic paradigm, they can continue to reap further returns from the current dying paradigm by propping up generative AI as a way to also prop up their expensive asset base. Unfortunately, the enormous sums invested in computing capacity for generative AI have to make an enormous return, or they’ll be wasted. Jim Covello, Head of Global Equity Research at Goldman Sachs, argues that although AI looks like it’ll attract US$1 trillion in investment in the coming years, there really isn’t a US$1 trillion problem for it to solve. Someone’s going to lose a lot of money.

To conclude, none of this would be an especially big problem except that more and more government, public capital, and private capital investment is going into AI, especially generative AI. This investment could be going elsewhere rather than being narrowly deployed to support one (highly speculative) technological system. And, if I’m right that this is the death throes of the current techno-economic paradigm, then all this investment is being sucked into recouping the investment of Big Tech firms rather than laying the groundwork for the start of the next cycle. Our future is being starved to feed a defunct technoscientific cycle.

Calls to Action:

• We need to strengthen competition policy. This is happening in Canada and other countries, but it needs continuous support so that specific sectors, companies, and investors cannot undermine markets.
• We need to diversify ownership of the “digital infrastructure stack” on which we all depend. This means investing in broadening the range of infrastructure providers, including investing in publicly-owned or non-profit alternatives in Canada. This is becoming more important with President Trump’s threatened tariffs.

• We need to rethink the way we fund technological innovation, especially productivity-enhancing technologies. Currently, early-stage funding tends to be focused on a few startups that have the potential to capture whole markets, locking us into a (potentially problematic) socio-technical system. Lock-in can very easily lead to stagnation and decline while limiting our capacity to develop new ideas and technologies.
• We need to broaden the focus of technology investment so that it doesn’t (mainly) end up going into one sector or technology (e.g. generative AI). This means finding ways to incentivize diverse investment strategies and disincentivize narrow ones.
• We need to value a range of social, political, and economic outcomes when it comes to technological innovation. Focusing on monetary outcomes to the near exclusion of social or political outcomes undermines our social fabric and political institutions; this means supporting and nurturing technologies which are not monetizable.