Phases of Economic Growth

From Malthus to AI

The statement that “Malthus was wrong” is often repeated with a sense of triumph, but it reflects a misunderstanding of what Thomas Robert Malthus actually observed. For most of human history, the logic he described was painfully accurate. Populations grew more quickly than the means of subsistence, pressing wages and living standards back toward a steady state of scarcity. Real incomes fluctuated with the harvest, technology, and disease, but across centuries, the average worker lived and died close to subsistence (Malthus 1798; Clark 2007). The true mystery of economic history is not Malthus’s pessimism, but the fact that his world eventually disappeared. Understanding how humanity escaped the Malthusian trap remains essential for understanding why growth happens at all and why it sometimes stops.

If we look across time, the pattern is clear: societies spend long stretches in near-equilibrium, with living standards tied to resource constraints, punctuated by rare and transformative breakthroughs that reshape production and knowledge. These are what might be called “productivity spurts,” moments when new general-purpose technologies (GPTs) such as the printing press, the spinning wheel, the steam engine, or the microprocessor drastically increase the efficiency of work and the reach of human knowledge. In these moments, the basic arithmetic of scarcity temporarily breaks. When productivity rises faster than population or resource depletion, the Malthusian logic gives way to sustained gains in welfare. Yet this escape is never permanent. Every wave of progress eventually encounters limits, and the world settles again into a slower, more constrained phase until another major technology emerges. Economic growth, in this view, is not linear. It is phasal. The natural condition of human economies is Malthusian equilibrium; the bursts of progress are the exceptions that must be continually recreated (Clark 2007; Galor 2011).

To appreciate the profundity of this shift, it helps to start where Malthus did. In the pre-industrial world, productivity gains were typically erased by population growth. Suppose an English farmer invented a more efficient plow or had a few good harvests. The extra income would allow earlier marriages and more children. Within a generation, the population would rise, the land would be subdivided, and wages would fall again. The average person lived only slightly better than their ancestors did centuries before. Epidemics, wars, and famine periodically reduced populations, allowing temporary recovery, but the long-run pattern was flat. Global GDP per capita barely changed for millennia. It is difficult for modern observers to comprehend how stable that equilibrium was. Malthus’s insight, that population pressure could permanently constrain prosperity, was not pessimistic for his time; it was descriptive (Malthus 1798; McNeill 1976).

And yet, history did eventually bend. Somewhere between the fifteenth and nineteenth centuries, the equilibrium cracked. For the first time, the relationship between population and subsistence loosened. The world shifted from a static economy to a dynamic one. To understand how that happened, we must examine how information and technology interacted with institutions to change the rules of the game.

The story begins not with steam, but with ink. When Johannes Gutenberg perfected movable type in the mid-fifteenth century, he did more than revolutionize communication. He reduced the marginal cost of reproducing information by orders of magnitude. A single press could produce hundreds of identical copies of a text in the time it once took a monk to copy one by hand. This lowered the barriers to entry for knowledge, which had previously been guarded by religious and political elites. The printing press democratized learning. Within fifty years of Gutenberg’s invention, presses had spread to over two hundred European cities (Eisenstein 1979; Dittmar 2011).

The economic effects were profound. Cities that adopted the printing press early saw faster population growth, higher literacy, and greater innovation in subsequent centuries (Dittmar 2011). Ideas that once took years to circulate could now spread in months. Technical manuals, scientific treatises, and commercial guides diffused through Europe’s growing merchant networks. Knowledge compounded. The Protestant Reformation, the Scientific Revolution, and the Enlightenment all relied on this infrastructure of cheap information (Eisenstein 1979; Mokyr 2002). Knowledge is not only a cultural force but an economic one: it is the ultimate input into productivity. When more people can access, test, and improve on prior ideas, innovation accelerates. The printing press was, in this sense, an early general-purpose technology of information, and it paved the way for industrialization.

But technology alone was not enough. Many societies invented or adopted new tools, only to stagnate. What distinguished the countries that escaped the Malthusian trap was not only what they invented but how they organized themselves. Institutions—the formal and informal rules that govern incentives—determined whether new technologies translated into broad-based growth. As Douglass North famously wrote, institutions are the “rules of the game” that shape economic performance (North 1990). When those rules protect property rights, reward innovation, and allow entry, growth compounds. When they are extractive, protecting elite rents and suppressing competition, innovation stalls.

Empirical history supports this. Acemoglu, Johnson, and Robinson (2001) showed that colonies where Europeans settled with inclusive property regimes, such as in North America, developed sustainably higher incomes than those where extractive institutions dominated, as in parts of Latin America and Africa. Similarly, the divergent paths of North and South Korea in the twentieth century reveal how political and institutional design determine technological diffusion. The two Koreas share geography, language, and culture, but the South’s inclusive, market-oriented institutions produced one of the most advanced economies on Earth, while the North’s closed, extractive system remains impoverished. The same logic applied centuries earlier. Where governments protected intellectual freedom and competition, the press, and later industrial machinery, flourished. Where censorship and monopolies prevailed, progress slowed.

The Industrial Revolution was therefore not simply a mechanical event but an institutional one. It required a combination of accessible knowledge, inventive activity, and governance structures that allowed entrepreneurial experimentation. James Watt’s improvements to the steam engine, for example, built on a long lineage of published research and practical tinkering. Britain’s patent system and relatively open financial markets allowed inventors to commercialize ideas and attract investment. Schumpeter later called this process “creative destruction”: innovation destroys old equilibria and creates new ones (Schumpeter 1942). But destruction without inclusion can just as easily create oligarchy. Inclusive institutions, which distribute opportunity and ensure competition, are what keep the system dynamic rather than brittle.

By the mid-twentieth century, economists began to formalize what had long been implicit in history. Robert Solow’s neoclassical growth model (1956) demonstrated that capital and labor accumulation alone could not sustain long-term growth because of diminishing returns. Only technological progress—an exogenous factor in his model—could explain persistent increases in output per worker. Paul Romer later internalized that factor. His theory of endogenous growth (1990) treated technology as the product of purposeful economic activity, subject to incentives, institutions, and ideas. Technological change arises from investments in human capital and research, and it can be influenced by policy.

More recently, unified growth theory (Galor 2011; Kremer 1993) has woven these threads together. It describes a long Malthusian phase where population growth absorbs economic gains, followed by a demographic transition where falling fertility allows per-capita income to rise, and finally a modern growth regime sustained by innovation and education. The shift between these phases occurs when the rate of technological progress exceeds the rate at which population growth erodes gains. In other words, growth depends not only on invention but on the social and demographic conditions that let it accumulate.

Each major technological era since Gutenberg has followed a similar pattern: new means of information production lead to wider participation in knowledge creation, which fuels innovation, which in turn supports new institutions and further growth. The digital revolution of the twentieth century was an extension of this sequence. Computers, telecommunications, and later the internet exponentially reduced the cost of transmitting, storing, and replicating data. The diffusion of knowledge became nearly frictionless. Yochai Benkler (2006) described this as the rise of “commons-based peer production,” where open collaboration and information sharing enabled large-scale innovation without traditional market hierarchies.

Yet the digital era also introduced new pathologies. The same networks that democratized access to information also amplified misinformation, polarization, and manipulation. Studies show that false or sensational information spreads faster than accurate news on social media (Vosoughi, Roy, and Aral 2018). The abundance of data has not always translated into abundance of wisdom. When signal becomes overwhelmed by noise, the productivity of knowledge itself can decline. The lesson echoes Malthus in spirit, if not in mechanism: unchecked proliferation—whether of population or information—can reintroduce scarcity in another form, scarcity of attention, trust, and meaning.

Today, the world stands again on the edge of a new phase. Artificial intelligence, particularly in its generative form, is the latest in a lineage of general-purpose technologies that promise to reshape the economy. The question is whether AI will deliver another productivity spurt or merely a transient spectacle. The answer depends on two things: whether AI genuinely enhances productivity at scale, and whether institutions guide it toward inclusive and responsible use.

Empirical evidence from early adoption suggests significant potential. In controlled studies, access to AI tools has increased productivity for customer-service agents by double-digit percentages, with the largest gains among less experienced workers (Brynjolfsson, Li, and Raymond 2025). Other experiments show that professionals using large language models complete tasks faster and with higher quality, particularly for mid-level writing and analysis (Noy and Zhang 2023). These findings echo the early stages of past technological waves: efficiency rises first at the task level, then diffuses through complementary innovations in workflow and organization.

The challenge is scaling these gains across the economy. Task-based macroeconomic models developed by Acemoglu and Restrepo (2019) demonstrate that aggregate productivity depends on how technologies reallocate tasks between humans and machines. Automation alone can displace workers without creating new productive niches. True growth occurs when technologies not only substitute for existing labor but also create new tasks that exploit human comparative advantages (Acemoglu 2024). If AI tools amplify human judgment, creativity, and coordination, they can expand total output. If they merely replicate existing tasks more cheaply, they risk deepening inequality while stagnating overall productivity.

Accurate measurement is therefore critical. Current AI benchmarks, such as the Massive Multitask Language Understanding (MMLU) and Holistic Evaluation of Language Models (HELM), test models on cognitive or linguistic tasks. These measures are useful for assessing capability, but they are poor proxies for economic value. Real progress should be measured in productivity terms: changes in output per worker hour, reductions in defect and rework rates, improvements in energy efficiency, and acceleration in research cycles. These are the statistics that determine whether an economy escapes stagnation. Productivity is, after all, the engine of prosperity.

Reliability is part of productivity. The costs of hallucination, misinformation, or error can offset the benefits of automation. Systems must therefore be designed with rigorous validation, traceable provenance, and human oversight. The engineering challenge is less about creating ever-larger models and more about embedding them safely and effectively into production. The future of AI benchmarking should align with industrial metrics, not entertainment. The true test is whether AI improves the precision, speed, and reliability of human effort.

Institutions will again decide how this phase unfolds. The diffusion of past GPTs has always depended on the surrounding social architecture. Inclusive systems that promote competition and learning translate technology into broad-based prosperity; extractive systems that concentrate control capture the gains for the few. The same dynamic applies to AI. If governments and firms deploy AI primarily to reduce labor costs, the wage share may fall even as output rises. If they deploy it as a tool to augment workers, create new professions, and expand access to expertise, productivity and inclusion can reinforce each other (North 1990; Acemoglu, Johnson, and Robinson 2001; IMF 2024). The next productivity spurt will depend as much on our institutions as on our algorithms.

Seen from this long arc, economic history resembles a series of escapes from Malthusian gravity. Each escape relied on technologies that multiplied the productivity of knowledge, combined with institutions that encouraged broad participation. The printing press democratized access to ideas; the Industrial Revolution transformed knowledge into machinery; the digital era connected billions of minds; AI now offers the possibility of automating and amplifying cognition itself. The pattern repeats, but the outcome is not predetermined. Our systems must channel these tools toward productive and ethical ends.

The central argument of this essay is simple but consequential. Economic growth is not a smooth continuum but a sequence of phases in which periods of stagnation are punctuated by technological and institutional breakthroughs. Humanity’s default condition remains Malthusian in nature, and our brief eras of prosperity are built atop the productivity spurts that general-purpose technologies enable. As the world approaches the threshold of the AI era, the question is whether we will treat intelligence as a spectacle or as an instrument of real production. The right metrics, incentives, and safeguards can make the difference between another plateau and a new epoch of growth. Malthus was right about his world. The task before us is to be right about ours.

---

References

Acemoglu, Daron. 2024. “The Simple Macroeconomics of AI.” Working paper, MIT.

Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2001. “The Colonial Origins of Comparative Development.” American Economic Review 91 (5): 1369–1401.

Acemoglu, Daron, and Pascual Restrepo. 2019. “Automation and New Tasks: How Technology Displaces and Reinstates Labor.” Journal of Economic Perspectives 33 (2): 3–30.

Benkler, Yochai. 2006. The Wealth of Networks: How Social Production Transforms Markets and Freedom. New Haven: Yale University Press.

Bloom, Nicholas, Charles I. Jones, John Van Reenen, and Michael Webb. 2020. “Are Ideas Getting Harder to Find?” American Economic Review 110 (4): 1104–44.

Brynjolfsson, Erik, Danielle Li, and Lindsey R. Raymond. 2025. “Generative AI at Work.” Quarterly Journal of Economics 140 (2): 889–930.

Clark, Gregory. 2007. A Farewell to Alms: A Brief Economic History of the World. Princeton: Princeton University Press.

Dittmar, Jeremiah. 2011. “Information Technology and Economic Change: The Impact of the Printing Press.” Quarterly Journal of Economics 126 (3): 1133–72.

Eisenstein, Elizabeth. 1979. The Printing Press as an Agent of Change. Cambridge: Cambridge University Press.

Galor, Oded. 2011. Unified Growth Theory. Princeton: Princeton University Press.

Gordon, Robert J. 2016. The Rise and Fall of American Growth. Princeton: Princeton University Press.

Hendrycks, Dan, et al. 2020. “Measuring Massive Multitask Language Understanding.” arXiv:2009.03300.

International Monetary Fund (IMF). 2024. “Generative AI and the Labor Market: Risks and Policy Responses.” Washington, DC.

Kremer, Michael. 1993. “Population Growth and Technological Change: One Million B.C. to 1990.” Quarterly Journal of Economics 108 (3): 681–716.

Liang, Percy, et al. 2022. “HELM: Holistic Evaluation of Language Models.” arXiv:2211.09110.

Malthus, Thomas R. 1798. An Essay on the Principle of Population. London: J. Johnson.

McNeill, William H. 1976. Plagues and Peoples. Garden City, NY: Anchor Press/Doubleday.