The Political Economy of AI — Now & In The Decades to Come

Eight months after the world was awed by GPT, we are beginning to see traces of policy and legislation about artificial intelligence. The European Union is set to pass its comprehensive AI Act, Southeast Asia is set to publish a governance code for AI usage, and the US Congress recently held a hearing on the topic of AI regulation which featured among others, Sam Altman, CEO of ChatGPT’s parent company, OpenAI. Make no mistake: sooner or later, policy is coming on AI and along with it, an economic and moral debate that will in some ways resemble that of previous technological breakthroughs, but in other ways, diverge from what we’ve seen before.

Behind every policy to come, there’ll be an implicit economic argument — a basis for decision-making that may either serve as the inherent logic for the policy or a post-hoc justification for it. Either way, not many of us are versed in the economic lexicon, ideas, and models that may soon be used to legislate AI, and in my view, that’s a problem. In this piece, we’re going to summarize different economic viewpoints on technology, extrapolate what they may have to say about AI based on reactions to previous technological innovations, and we’ll end with a discussion on real policy proposals emerging from each economic perspective. Ready?! Let’s go!

The multiple angles of Political Economy

Political economy is a rather elegant and academic reminder that what we think of economics — a set of models, curves, and theories about decision-making, inflation, employment, etc.— isn’t quite the full picture. An economy is a construct, something we’ve built over history; something that has taken different shapes and forms and has interacted with societies in different ways over time. While some schools of thought have become more dominant over time, the idea of political economy is one that invites us to look at economics from different lenses, each revealing a unique set of assumptions, emphasis, and conclusions. For this article, we’ll focus on four major schools of economic thought. We’ll start with classical economics, the dominant school of thought in the 18th and 19th centuries, but quickly switch focus to three major schools of thought today: neoclassical, Keynesian, and Marxian economics. I’ve taken the time to synthesize — and let’s be real, oversimplify — the main characteristics of each school below.

• Classical Economics: Taught in most schools as the beginning of economics as we know it today. Among the most influential classical economists are Adam Smith, David Ricardo, John Stuart Mill, and Thomas Malthus. Classical economics focused on the idea of a free market, one that required little to no intervention, and one that had emergent properties such as trade, comparative advantage, and self-regulation.
• Neoclassical Economics: What most schools teach today as mainstream economics. Like classical economics, it focuses on free markets and self-regulation, but unlike classical economics, it seeks to formalize and abstract concepts into models and mathematical representations. Notably, this field popularized supply and demand diagrams and introduced concepts such as utility, indifference curves, and the production possibility frontier.
• Keynesian Economics: Conceptualized by James Maynard Keynes after the Great Depression, it challenges the notion of the free market as an optimally regulating agent and instead focuses on the role of government and monetary policy to create stable economies. Keynesian economics has largely drawn from neoclassical thought in its fundamental abstractions but has offered its own modifications with concepts such as aggregate supply and demand, and the Phillips curve.
• Marxian Economics: Based on a critique of capitalism by Karl Marx during the industrial revolution, it focuses on class structures, labor relations, and market commodities, and introduces a number of concepts such as surplus value, socially necessary labor time, appropriation, and exploitation. Influenced by German philosophy, and many popular ideas at the time, Marxian economics definitely has a different vocabulary and diverges in major ways from classical and neoclassical economics in what they emphasize.

Perspectives on Technology & Growth

Classical economists, in general, saw technology positively, although they focused more on factors like land, labor, and capital, and how they drove economic growth — or as Adam Smith put it, the ‘wealth of nations’. Technology was seen as something that improved efficiency and therefore was good, but it was mostly treated as an external factor.

Neoclassical economists, on the other hand, are a little more interested in technology as an endogenous factor. Many neoclassical models that focus on modeling economic growth use technology as an input so that varying levels of technology maturity can be used to predict economic output. A popular model both used by neoclassical and Keynesian economists is the Solow Growth Model, which models the technological level as a constant that directly and positively influences the level of output.

The Solow Growth model shows that as technological level (A) increases, so does the overall output per worker, hereby expressed as y. The change in capital per worker, expressed by Δk also goes up. This measure is known as capital accumulation and can also be expressed as the difference between the investment rate share of output (sy) minus the depreciation rate (δk). If none of this made sense to you, just notice how the curve with more technology, shown above in green, is taller and corresponds to more output, which is good.

Another way in which both neoclassical and Keynesian economists incorporate technology is by graphically showing how technological innovations can decrease the price of a good by lowering the production cost and therefore increasing the supply of the good. This is referred to as a downward shift of the supply curve and there are indeed many examples of this in our daily lives — a good one is the cost of running computing power in the cloud, which is relatively cheap today as microprocessing capabilities and software sophistication have evolved.

In a typical supply and demand visualization, the equilibrium point is represented by the intersection of the demand curve, namely, the quantity consumers are willing to pay for a given price; and the supply curve, namely, the quantity producers are willing to sell at a given price. We can see that the equilibrium point in blue, showing an economy with more technology, is one where more quantity is produced at a lower price.

Using similar models of supply and demand, we can also see how technological change can affect wages in various ways. If a technology serves as a substitute for labor — that is, if a human laborer can be replaced entirely by a technology — then the wages for that profession will go down. On the other hand, if a technology is a complement to labor — that is, if the technology becomes a high-demand or helpful skill for laborers to have — then laborers in that class will see higher wages. Again, this is quite apparent in our modern lives as we compare average salaries for tech workers vs. average salaries in other industries.

In this visualization, the supply curve now refers to the supply of laborers while the demand curve refers to the demand for that particular kind of labor. It’s in a way an inversion of how we typically think of supply as companies selling things and demand as consumers buying them. (John Lynham, 2023)

Yet another discussion of technology in the economy lies in the concept of creative destruction — originally coined by Marx, but later popularized by neoclassical economist Joseph Schumpeter. Creative destruction refers to the transformation technology can bring to an economy by destroying an industry whilst simultaneously creating a new one. Examples include the carriage industry being replaced by the automotive industry, the CD industry being replaced by digital media and streaming services, and to some degree, manufacturing jobs being replaced by automated machines. Creative destruction is going to be a widely important term in the coming AI discussions as more professionals ask themselves whether AI is more of a substitute or more of a complement to their work.

US manufacturing has been on the decline for some time, and many experts attribute the decline to automation and factory technology. That is not an uncontested view, however, with some economists arguing much of the shift can be attributed to outsourcing and supply chain transformations with globalization.

Finally, it’s time to add the Marxian perspective to the equation. Marx made a clear distinction between labor and labor power — the former, the actual process of work, and the latter what the capitalists were actually purchasing and paying for in wages. This distinction is important to Marx because labor power, as a commodity, is bought and sold for a price equal to its value, but labor is capable of producing additional value. When that surplus value is not appropriated by the laborers who produced it but instead goes to the capitalists as profit — that’s what he called exploitation (commonly referred to with the variable ‘s’ in Marxian economics).

This chart is somewhat of an oversimplification — the true units that are ultimately discussed by Marx are actually the units of socially necessary labor time (SNLT), measured in hours, and not as simply abstract concepts. Exploitation rates can be quantitatively calculated from the ratio between surplus labor and necessary labor time, profits, and wages.

Labor itself has multiple components in Marx’s analyses. Living labor refers to the actual physical and mental work performed by laborers — the innovation, the craft, the execution — and is what Marx thought of as the source of the surplus value creation. Meanwhile, dead labor refers to the machines, tools, and processes that could make living labor more productive but ultimately did not actually create surplus value itself. Marxian economists like to point to a modern chart as strong evidence of this logic — the gap between productivity and pay since the middle of the 20th century when major technologies started to emerge in the world. Indeed, productivity has gone up nearly 65% since the introduction of computers and other modern technologies, but hourly pay has only gone up by roughly 17%. Profit is definitely generated in larger volumes— but it is not always the case that overall wages will see their reflection.

From the Economy Policy Institute’s website, “Productivity measures how much total economy-wide income is generated in an average hour of work. As productivity grows and each hour of work generates more and more income over time, it creates the potential for improving living standards across the board.” Indeed, productivity has grown, but Marxian economists, emphasizing questions of appropriation, would ask whether living standards have indeed increased for those who have become more productive themselves.

What’s different about AI?

None of the economic perspectives examined today make claims about specific technologies — rather, they speak of technological progress in general. In that sense, there’s nothing special about how models should treat AI versus how they may have treated the internet, computers, or automobiles. From the Marxian perspective, for example, David F. Ruccio elegantly writes in his latest book:

“Just as in the first Industrial Revolution, new production methods and new ways of organizing workplaces are redefining the process of work and the possibilities of exploiting workers during the immediate process of production. For example, with the further development of machinofacture, the conditions are eventually created for the introduction of artificial intelligence, when not only the human hand but the human brain, is replaced in the production of machines by machines. That’s why contemporary capitalism is both radically different from and, in other ways, exactly the same as it was in the nineteenth century when Marx first formulated his critique of political economy”

Yet, there are two aspects of AI that may be a little unusual — the speed of adoption, and the scale of impact. On the former point, consider how long it has taken for households in the USA — historically, the country with the earliest access to many technologies — to adopt different innovations. It took nearly 60 years for telephones to be widely adopted, but only took the internet a little more than a decade.

Chart created by Nicholas Felton (HBR, 2023) — the link provided also shows the adoption curves for smartphones, showing an even more exponential rise. It would be nice to see a similar chart for worldwide adoption but I’m certain the trends there are less defined given the unfortunate inequality and access to resources between countries in different parts of the world.

It’s too soon to model the adoption curve for AI, but ChatGPT — one of the most salient examples of AI in the last year — was the fastest tool to ever reach 1M users, doing so in just 5 days. Some argue that the speed of development in the AI space is faster than the appropriate use cases for it and that can make it a challenging one-to-one comparison with previous innovations. However, that isn’t stopping companies from proactively thinking about how to incorporate AI into their products and indeed, consulting firm Accenture predicts the ‘AI transformation’ in companies will be quicker than the ‘digital transformation’ they have been undergoing in the many years past.

The study, conducted last year, could perhaps be revised for a post-GPT world, where at least in terms of companies' appetite, generative AI is far more explicitly on the table. Nonetheless, the conclusion seems to be corroborated by other sources — it will not take as long for AI to go mainstream in many companies.

Secondly, the scale of AI’s impact is something that is still being forecasted but likely will be of unprecedented reach. According to McKinsey, it is possible that 30% of the human workforce would be automated by 2023 — a displacement of up to 375 million workers. These figures would have seemed far-fetched a few years back, but it is becoming increasingly clear white and blue-collar workers alike may see their jobs being replaced by code or code-based systems. A recent report by Goldman Sachs estimates, in fact, that as much as two-thirds of occupations could be partially or completely automated by AI. Wendy’s is now using AI chatbots to take drive-thru orders, and tools like JasperAI is adding serious disruption to content writing and SEO optimization. Even highly specialized fields such as radiology are already seeing AI software tools capable of high-accuracy scan reads — a possible threat to a position that has typically taken many years of training, and one that boasts high salaries in hospitals.

Though apocalyptic at first, this chart features a caveat by Goldman researchers: “Although the impact of AI on the labor market is likely to be significant, most jobs and industries are only partially exposed to automation and are thus more likely to be complemented rather than substituted by AI,” (Goldman Sachs, 2023)

Jobs will also be created, to be sure. In fact, occupations such as prompt engineers and large language model researchers have already emerged. Some estimates say up to 97 million jobs could be created by AI industries by 2025. Undoubtedly, a new cycle of creative destruction is coming. But how much shift will be caused to supply curves and how quickly? How much will wages be impacted? Will new profits be reflected in the well-being of average workers? These are the questions economists and policy-makers will soon begin to ask.

AI, the Economy, and Society

It’s easy to get lost talking about economic abstractions and forget they have very real implications for society at large. AI’s impact on employment, wage rates, and innovation will be reflected in many facets of daily life. Politics, for example, may be highly impacted. In his 1944 influential book ‘The Great Transformation’, economist Karl Polanyi introduced the concept of the double movement — in his view, a cycle between economic re-regulation worsening working conditions and therefore giving rise to legislative and political reactions in response.

The double movement is conventionally illustrated with the analogy of the pendulum — indeed, if we think about almost any democratic country, there are visible swings of governance between parties who promote de-regulation and those who promote more governmental control. Many Keynesian and socialist economists would likely reference or invoke elements of the double movement in explaining such political shifts.

To Polanyi, there could be many possible outcomes from the double movement — things like demands for policy reform, labor protections, and more social programs, but also, the possibility of populism and authoritarianism. The double movement doesn’t converge to a quantitative model that may be seen in neoclassical and Keynesian economics, but many neoclassical and Keynesian economists have also made parallels to the relationship between the rise of populism and factors such as inequality, uncertainty, and unemployment, which are mentioned by Polanyi. Today, Lucas Chancel and Thomas Piketty lead the World Inequality Lab which comes out with annual reports on global inequality. As of 2021, they concluded the top 10% of the global population captured 52% of the total income and 76% of total wealth in the world, while the bottom 50% captured only 8.5% of the income and 2% of the wealth. Clearly, we already have high levels of inequality and have seen populist regimes get stronger in some parts of the world. Could AI add fuel to the fire?

From Piketty’s website: “Global inequality, as measured by the ratio between the average income of the top 10% and the average income of the bottom 50%, more than doubled between 1820 and 1910, from less than 20 to about 40, and stabilized around 40 between 1910 and 2020. It is too early to say whether the decline in global inequality observed since 2008 will continue.” (Piketty, 2021)

The economic impact of AI adoption could also be societally reflected in a broader discussion of both the meaning and shape of work. More companies are experimenting with shorter work weeks, and some countries are beginning to trial universal basic income (UBI) as a possible safety net measure for the threat of large unemployment. UBI programs have been thought to be expensive and a disincentive for those without jobs to get back into the labor force. Such claims could turn out to be true. But it is also true, from nearly every economic viewpoint, that the economy functions by a kind of circular flow — an exchange of money between households who seek to purchase and businesses who seek to sell. In a world of mass unemployment, there would not be enough money to flow from households, and the economy, in its entirety, could suffer.

The circular flow diagram is shown in most microeconomics textbooks. The basic idea is that households (individuals) sell their labor, which is consumed by businesses who in turn sell their goods and services to — you guessed it — the households themselves. Marxian economists may have a few modifications here, for instance, making a distinction that what is purchased is labor power, and adding that other forms of exchange can take place such as unpaid household work, trading, gift exchanges, and more.

There is far more that will be societally reflected by AI’s economic impacts — some possibilities are outlined in a recent report by the White House. It could be that by making it easier for individuals to write and produce media with generative tools, more members of society would have side gigs and pursue part-time creative passions. Just as much, it is possible that AI’s economic impacts in healthcare could translate to new paradigms of interaction between providers and patients — and improve access to healthcare by giving more economical options for AI-driven patient care. Drug discovery, also potentially powered by AI, could become a lot faster and cheaper. AI-influenced fashion, design, architecture, and engineering may also become part of our lives — though I sure hope it doesn’t mean we all start to dress the same and live in places indistinguishable from one another.

Conclusion

Economics, as a social science, is comprised of a multitude of perspectives that have changed over time. In today’s piece, we saw how different aspects of technology’s impact on the economy may be emphasized by different economic perspectives — and how we could extrapolate these perspectives to better understand what economists and policymakers may have to say about artificial intelligence. Innovations in AI will not only affect employment, wages, and the circular flow but may also lead to political movements and specific policies such as stronger safety nets for the unemployed. They may also have undesired effects that some economic perspectives have emphasized in previous technological advancements — exploitation, inequality, and a poor allocation of rewards compared to the gained productivity allowed by the new technology. As we begin witnessing public discourse about artificial intelligence, keeping in mind the different economic perspectives and their lexicons can help us navigate headlines and better understand the backbone of each policy— most importantly, it can help us better understand how we, individually, feel about them.