The Economic Future of AI: Scarcity, Labor Share, and Redistribution

The advent of advanced AI and automation raises profound questions about the future of our economy. Will capital's share of wealth increase dramatically? What will become scarce, and how should the immense wealth generated by AI be taxed and redistributed? These are the central questions explored in a discussion with Alex Imas, Director of AGI Economics at Google DeepMind and Professor of Economics at the University of Chicago, and Phil Trammell, Head of Economics at Epoch and research scholar at Stanford.

What Will Be Scarce?

One plausible candidate for future scarcity lies in the "relational sector"—services and goods where the human element is integral to their value. As automation makes many other things abundant, human involvement in services like those provided by a barista or a performer could become a premium. This creates a dynamic where a "human economy" of services for humans coexists with a "machine-only economy" producing automated goods.

However, the notion of the human-only economy shrinking as a proportion of the total economy is not a foregone conclusion. Economists have historically struggled with forecasting, as evidenced by debates dating back to David Ricardo in the 1820s. Ricardo predicted that automation would lead to widespread unemployment and social unrest. While his observation that jobs of his era would be automated proved correct, he underestimated the economy's capacity for structural change. As automated goods became cheaper, people had more disposable income to spend on services, leading to new job creation and a surprisingly stable labor share.

This historical perspective highlights the difficulty of predicting future economic outcomes. Instead of precise forecasts, a more useful approach might be to explore potential scenarios and identify the dimensions of scarcity that would drive them. This, in turn, can guide data collection efforts.

Labor Share and Capital Share: A Historical Puzzle

The economy's total output is divided between labor (wages) and capital (profits, rent, dividends). Historically, labor has consistently received around 60% of the economy's output, with capital taking the remaining 30-40%. This stability, known as a Kaldor fact, is surprising given the extensive automation that has already occurred. Some argue that labor share has been declining in recent decades, while others contend that accounting changes obscure this trend, suggesting labor share has remained remarkably constant.

The relationship between labor and capital is crucial. If they are complements, meaning both are needed for production, then both would logically be compensated. Even in seemingly fully automated sectors, labor often plays a role further down the supply chain, contributing to the value of machines and components.

A significant qualitative shift could occur if entire supply chains for certain goods become fully automatable, with no intrinsic human role desired. The implications for the overall capital share are, however, ambiguous. If demand for these fully automated goods becomes satiated rapidly, their marginal utility could fall faster than their quantity increases, potentially shifting economic value towards sectors where human involvement remains crucial.

The Relational Sector and Consumer Preferences

The concept of the "relational sector" is not solely about scarcity but also about intrinsic human preferences for empathy, connection, and interaction with other people. Experiments have shown that consumers are willing to pay more for goods and services where a human is involved in the process, even if AI could perform the task. This preference is not necessarily analogous to a horse as a mere input; it suggests a value derived from the human interaction itself.

However, the persistence of this preference is an open question. As AI becomes more sophisticated, consumer preferences might evolve. The key is whether the value derived from human interaction in a service or good decreases significantly when a human is replaced by AI. If the value of the output diminishes when the human element is removed, the relational sector's importance will persist.

The "Messy Middle" Scenario

A potential future scenario, termed the "Messy Middle," envisions AI automating many jobs without generating sufficient wealth to compensate those displaced or create a Pareto improvement for everyone. While the resources saved from not paying human workers theoretically exist, allocative inefficiencies and political challenges could prevent their equitable redistribution. For instance, the political sustainability of providing high-wage former employees with equivalent compensation when many working individuals earn less is questionable.

While some believe the pie will grow rapidly enough to offset these issues, others point to scenarios where automation offers only marginal cost savings. In such cases, short-term effects like widespread layoffs could occur, even if long-term demand for new applications of technology eventually emerges. The political economy of unemployment, even a few percentage points, can dramatically shift public sentiment and policy.

The "drip" scenario, where individuals are not massively unemployed but move into lower-paying jobs, is also a concern. This mirrors historical events like the automation of telephone operators, where displacement occurred gradually over decades, leading to underemployment. The core of the "Messy Middle" concern is that the pie doesn't grow significantly, leading to displacement without broad-based abundance.

Taxation and Redistribution of AI Wealth

Several mechanisms for taxing and redistributing AI-generated wealth are being considered, each with its own complexities:

A more nuanced approach might involve separating revenue generation from distribution. Governments could use broad-based taxes to acquire stakes in key AI companies and then distribute these shares widely. Alternatively, consumption taxes like VAT could fund such initiatives. The key is to ensure that the mechanisms for raising and distributing wealth are robust and equitable.

The Unlikelihood of Demand Collapse

Despite fears of a white-collar apocalypse, current data suggests that mass automation and unemployment due to AI are not yet occurring. While there might be subtle signals, such as slightly slower job growth for junior developers, overall demand for software engineers, particularly senior ones, remains strong. Anecdotal evidence of difficulty finding jobs might be amplified by an "AI narrative" rather than reflecting a systemic collapse.

The "O-ring theory" of jobs, where a job is a series of tasks and automation of complementary tasks can increase the value of remaining human labor, offers a counter-argument to widespread job destruction. If automation makes a product cheaper and demand responds elastically, it could lead to increased hiring. This elasticity of demand is crucial. While Jevons paradox suggests that increased efficiency can lead to increased consumption, this is not guaranteed for all goods. For some, like essential medicines or basic food, demand is inelastic.

The claim that software is a unique good for which demand will always increase as it becomes cheaper is also debated. The scenario of negative economic growth, as posited in some viral planning scenarios, requires highly improbable conditions, such as a hard bound on demand and a failure for capital holders to reinvest their wealth.

Human Integration in the Machine Economy

Integrating human employees into future production flows organized for AI could become increasingly difficult. AI operates at speeds and in ways that might be incompatible with human workflows. Even if humans possess a comparative advantage in certain tasks, the transaction costs and reliability concerns associated with integrating them into AI-centric production lines could be prohibitive.

This could explain why professions like law or accounting are not yet fully automated. Beyond the relational aspect, regulatory layers, licensing requirements, and the need for human accountability create frictions that keep humans in the loop. However, these regulatory layers might be transitional as AI capabilities advance and political systems adapt.

The Evolution of Preferences: Humans and AIs

The future economy will be shaped not only by human preferences but also by the preferences of AIs. Just as evolution on Earth favored agents with specific drives, future AI evolution might favor entities that grow and accumulate resources. This raises questions about whether entities with a preference for human-intrinsic goods will be the ones that accumulate the most resources.

It's uncertain whether AI entities will develop preferences for interacting with humans. However, human preferences for interacting with other humans, for empathy and trust, might be evolutionarily stable. Individuals who prioritize these human connections may be more likely to reproduce and pass on these preferences.

The wealth accumulation of the richest individuals also offers a clue. Many wealthy individuals, even when they could convert their wealth into immediate consumption, choose to reinvest and compound their capital. This suggests a preference for accelerating capital growth, a "Nick Landian" drive for accumulation. This drive, even without AI, could lead to a high capital share in the future.

What Should Developing Countries Do?

Economists are still grappling with how middle-income and developing countries can navigate the AI revolution. There are optimistic scenarios where AI technology levels the playing field, but also pessimistic ones where countries without AI production capabilities are left behind, especially as automation allows developed nations to produce commodities more cheaply.

The "Messy Middle" scenario could be particularly challenging for developing nations, as they start from a lower economic base. A key factor is the interest rate and the rate at which the price of capital-produced goods falls. If savings can be translated into significant future consumption due to high interest rates or rapidly falling prices of goods, even a small amount of savings could benefit developing countries.

The ability to "index" the economy—to gain broad exposure to the returns of AI—is crucial. Historically, this was difficult, but the rise of index funds made it more accessible. However, recent trends towards concentrated returns in private companies might make indexing harder. Developing countries need strategies to gain access to AI's economic gains, whether through sovereign wealth funds or subsidies for citizens to invest.

The analogy of AI to electricity is useful. Electricity is a fundamental utility that transformed the economy, but its benefits were broadly distributed. Social media, in contrast, concentrated rents with platforms. If AI becomes a fundamental utility like electricity, its benefits might be more widely shared. Open-source AI models could play a significant role in democratizing access.

Key Takeaways