← Writing

Aditya Mehta & Ana-Maria-Elena Radu

Nexus: How AI and Creative Destruction are Changing the World

Key takeaways

  • Whether AI leads to shared prosperity is not preordained.
  • AI is both an enabler of technology and a crystallization of the theory of creative destruction.
  • AI sits in the chain of historical technological diffusion—after the Industrial Revolution, electrification, and mass production.
  • Firms often adjust employment through hiring slowdowns, attrition, and task substitution rather than mass layoffs, capturing productivity gains while minimizing cost.
  • History shows productivity gains do not automatically raise wages—as in Engels' Pause during the Industrial Revolution.
  • AI's importance for scientific research in materials science and medicine is rising, producing valuable results at remarkable speed.

Introduction

Economists cannot seem to agree on what AI is doing to jobs. In August 2025, researchers at Stanford's Digital Economy Lab analyzed high-frequency payroll data from ADP (see ADP Research's discussion)—the world's largest payroll provider—to track employment changes after the release of ChatGPT. They found a 13–20% decline in employment among workers aged 22–25 compared with other age groups in the most AI-exposed occupations, with software development among the hardest hit. Interestingly, the mechanism was not layoffs: firms were simply hiring fewer junior workers. A project page for the same line of work is here.

Around the same time, researchers at Yale's Budget Lab examined aggregate labor-market indicators and reported no discernible disruption. Unemployment rates remained low; occupation employment shares looked broadly stable. At the macro level, the labor market appeared unchanged. Press coverage of that theme appears in TechTarget's summary of the Yale findings. One study detects a missing rung on the ladder; the other counts the ladder still standing.

Employment shocks are not new, but the way they manifest today is unlike most historical examples. How can workers on the front lines feel real and immediate harm while aggregate indicators stay calm? How can technological disruption be both present and absent at once?

The seeming contradiction is simple to resolve. It arises not because the economy is behaving strangely, but because we are measuring different layers of the same process. To make sense of it, we need a framework that distinguishes where disruption appears from what constrains adjustment.

Micro and macro: measurement and constraint

Looking further back helps. History shows tension between individual experience and economic data—sometimes inverted. Local innovations and tech shifts occurred while overall living standards stayed flat because of demographic gravity: the Malthusian trap produced micro changes without macro progress.

As old arrangements break and new ones take shape, diffusion and creativity do not unfold evenly. The former arrives quickly and locally; the latter is often delayed, diffuse, and unevenly distributed. Relying too heavily on averages flattens temporal and structural complexity into misleading calm.

Two dimensions are often conflated. First, measurement: some effects appear only at the micro level—tasks, hiring, firm reorganization—while others appear only at the macro level—occupations, sectors, national statistics. Historically, disruption almost always surfaces at the micro level first, long before it registers in aggregate data.

Second, constraint: at different moments, economic activity is limited either by labor (skill, time, organization) or by infrastructure (networks, capital-intensive systems, standards, access). Confusion arises when we debate labor outcomes while the binding constraint has shifted to infrastructure, or when we look for macro effects while the damage is still micro.

The earliest disruption often appears when technology reshapes what people do before it eliminates whether they work at all. Labor remains essential, but its role is reorganized. Late eighteenth-century Britain offers a clear illustration.

Lessons from the Industrial Revolution

Beginning in the 1760s, innovations such as the spinning jenny and water frame automated the upstream task of spinning yarn (background on industrialization). Yarn became cheap and abundant almost overnight. Weaving—the downstream process that turned yarn into cloth—remained a manual craft. The result was not immediate unemployment but a boom in handloom weaving. Employment expanded and wages even rose temporarily, because weaving became the bottleneck. Technology destroyed one occupation while temporarily intensifying demand for another—acting as a complement to labor rather than a substitute.

The power loom, widely adopted by industrialists in Great Britain, accelerated profits but also contributed to a significant rise in unemployment and deep societal divisions. The Luddites, who smashed power looms in response to mechanization, were not simply resisting progress; they were early detectors of a shifting task structure that would eventually transform their world. They were the canaries, signaling danger before aggregate data caught up. Ultimately, industrialists prevailed as the British government enacted legislation prohibiting destruction of power looms and publicly denounced the Luddites.

Modern technologies follow the same logic: disruption first appears where tasks are most easily codified. Entry-level work sits closest to routine, repeatable activities. When firms adopt new tools, they rarely begin by firing experienced workers; instead, they quietly hire fewer novices. That is why micro-level evidence can show real harm even while macro indicators stay stable—the damage concentrates at the point of entry, not yet at the level of occupations. Over time, task-level reorganization can give way to occupational displacement once substitution replaces complementarity.

Engels' Pause and wages

Economic historian Robert C. Allen explains why wages stagnated despite a per capita GDP boom driven by immense technological upheaval—a theory he calls Engels' Pause. In Engels' pause: Technical change, capital accumulation, and inequality in the British industrial revolution, he argues that between 1800 and 1830 the famous inventions raised aggregate total factor productivity growth to about 0.69% per year. That technology shock pushed output-per-worker growth to about 0.63% per annum but had little impact on capital accumulation or real wages—which remained constant.

Technological innovations and booms do not translate quickly into higher wages or quality of life for workers at the bottom of the pyramid; that increase is often a slow process taking decades. Wages rose much faster toward the end of the nineteenth century as society moved away from older systems toward capitalism and a belief that the future belonged to the bold.

Creative destruction

John T. Dalton and Andrew J. Logan, in Creative Destruction, examine the Industrial Revolution from a Schumpeterian perspective. They describe creative destruction as a force that "sculpts the riverbed over which the waters of economic history flow"—which is why it captured Schumpeter's imagination generations before Silicon Valley reintroduced "disruption" to the public.

Joseph Schumpeter described how technological advancement propels society by replacing outdated systems with new innovations. Destruction is inherent to progress as disruptive technologies supplant existing industries. Dalton and Logan clarify: "Creative refers to new innovations brought to market," while "Destruction refers to the fate of those antiquated products, processes, and social modes of organisation that such innovations replace." Schumpeter contended that long-term effects are often positive—the decline of one industry gives rise to several new ones. From a Schumpeterian angle, the fear and hardship of the Industrial Revolution were transitional challenges; subsequent expansion generated many new professions.

People often reject this when facing the present—understandably, self-preservation favors the near term. The instinct to oppose technologies that could cause widespread unemployment was shared not only by workers but by rulers. In The Power of Creative Destruction, the authors recount that in 1589 William Lee presented a machine for knitting stockings to Queen Elizabeth I, hoping for support for large-scale adoption. After reviewing it, she was against it, telling the inventor to consider what his invention could do to her poor subjects—"It would assuredly bring them ruin by depriving them of employment, thus making them beggars."

Escaping the Malthusian trap

Tomas Kögel and Alexia Prskawetz summarize Malthus's two assumptions (Agricultural Productivity Growth and Escape from the Malthusian Trap): higher per capita income above an equilibrium leads to population growth, and larger population dilutes per capita resources so consumption falls back to equilibrium—generating the trap in which technological innovation fails to produce stable GDP gains because population growth eventually causes stagnation.

In The Power of Creative Destruction, Aghion, Antonin, and Bunel note that in Malthus's world, only demographic decline from abstinence or restricted childbearing could increase GDP. Joel Mokyr, in Why Was the Industrial Revolution a European Phenomenon?, argues the trap's application is historically contingent: if human knowledge and economic infrastructure are sufficiently positive functions of population, Malthusian constraints can be overcome.

How did Britain escape? Aghion, Antonin, and Bunel emphasize the shift from agriculture to manufacturing, moving the economic focus from land to capital—eliminating setbacks from population growth and unlocking compound growth. Capital is not fixed like land; it accumulates over time. They summarize: the shift relies on threshold effects—population, demand, investment. Mokyr complements this with an institutional story: industrial societies adopted a scientific approach to innovation, emphasizing practical knowledge. After 1820, innovation did not trail off—it gathered force in continuous improvements and new applications.

Electricity and uneven access

Not all disruption is governed by labor; sometimes the constraint is access to the system. Electricity transformed the nineteenth century. Before it, cities depended on whale oil for light, lubrication, and more. Large-scale electrification made whale oil obsolete, with widespread unemployment in that industry—reported in the Wall Street Journal and New York Times—yet destruction of whale oil did not automatically guarantee universal access to electricity.

The 1880s "War of the Currents" pitted Edison's direct current against Tesla's alternating current (U.S. Department of Energy overview). DC could not easily be converted across voltages; AC could, using transformers. AC prevailed as cities were designed for a future where electricity was baseline, not luxury.

By 1930, nearly 90% of urban and rural non-farm dwellings in the U.S. had electricity—yet only 10.9% of American farms were on the grid. The barrier was not technical; it was the unit economics of distribution. Transmission lines cost roughly $1,500–$2,000 per mile (tens of thousands in today's dollars). In cities, cost could be amortized across many customers per mile; in sparse rural areas, private utilities had little incentive to extend the network. Old ways declined everywhere; new capabilities concentrated where networks already existed.

The Rural Electrification Administration did not halt creative destruction—it tried to reshape outcomes. Engineers cut construction costs with better conductors and lines; the REA paired that with low-interest loans to customer-owned cooperatives where coverage—not profit—was the goal (Lewis & Severnini on rural electrification; Encyclopedia entry on the REA; Strong Towns on the Rural Electrification Act). By 1950, cooperatives had extended power to millions of farms private utilities had written off. Deliberate public action overcame market incentives favoring concentration over inclusion.

The War of the Currents also shows how standards and architectures allocate value. Edison's DC implied local, fragmented networks; Tesla's AC enabled long-distance transmission and centralized grids. Edison waged a campaign emphasizing AC's dangers—public electrocutions of animals, and supporting the electric chair using AC to associate Westinghouse with death. Despite that, AC's practical advantages were decisive. Infrastructure battles are rarely settled by technical merit alone; incumbency, narrative, and power matter. Before standards settle, firms duplicate effort and productivity lags—not because technology fails, but because coordination and trust lag behind capability.

Model T and the ladder

Henry Ford's Model T (1908) became transformative not only as a car but as a system: in 1913 the moving assembly line cut assembly time from roughly 12 hours to about 93 minutes (The Henry Ford on the assembly line). Within a decade, an automobile cost less than a high-quality horse-drawn carriage. Carriage makers were not gradually reassigned—their industry ceased to exist. New sectors followed: petroleum, civil engineering (roads, bridges, highways), and services—motels, diners, repair shops, suburban real estate.

New industries absorb labor, but not necessarily the same workers. A fifty-year-old wheelwright did not seamlessly become a petroleum engineer; skills were rendered valueless while opportunities flowed to a younger, differently skilled cohort. For some, the ladder is not just hard to climb—it is removed. Firms hoard labor, freeze hiring, or use attrition; displaced workers may move into informal roles missed by headlines. By the time macro disruption is visible, structural change is already advanced—destruction becomes legible at scale years after it began in quieter ways.

AI and the four patterns at once

Artificial intelligence is unusual because it activates multiple patterns simultaneously. Stanford-style hiring data shows task-level disruption concentrated among junior workers (Digital Economy Lab on AI and labor markets); Yale-style aggregates reflect stability with new industries buffering the shock. The global compute divide shows infrastructure inequality—as Gregory C. Allen (CSIS) has identified as a defining geopolitical bottleneck of the AI era. Inside firms, shadow AI tools and competing standards expose organizational friction.

Looking only at aggregate statistics is like watching average sea level to detect a storm: the water looks calm in the aggregate while the coastline—entry-level cognitive work—is battered. Low overall unemployment does not negate structural signals in exposed sectors; it can confirm them. Task-level disruption appears quickly; standards stay unsettled; infrastructure advantages concentrate; aggregate statistics adjust slowly, buffered by inertia and demographics. Different layers yield different truths—they are not contradictions.

That is less evidence that creative destruction has changed its nature than that its internal dynamics are moving faster while habitual measures lag the lived experience of those at the margins. The canaries signal change before aggregate data catches up.

Conclusion

Schumpeter was right that creative destruction drives long-term growth, but he warned its benefits are not automatic or evenly shared. Gains depend on institutions, policies, and efforts to rebuild after disruption. The Industrial Revolution eventually raised living standards—after decades of stagnation and conflict. Electrification transformed society—after public action widened access. In each case, progress was real, but its moral character depended on how the transition was governed. The handloom weaver's prosperity rested on a fragile bottleneck; once power looms matured and factory weaving became viable, the bottleneck—and its protection—disappeared.

Historians still debate what let the Industrial Revolution escape the Malthusian trap: capital and thresholds versus institutions and the "industrial Enlightenment." The coexistence of micro-disruption and macro-stability is not a paradox to dismiss but a warning to understand. Creative destruction destroys first and creates later—and social strain accumulates in the gap. Whether that strain hardens into inequality or is softened by adaptation is not determined by technology alone, but by choices societies make while averages still look reassuringly calm.

None of our frameworks is certain; all are imperfect. Waiting for certainty is itself a choice. The missing rung is visible now, before the ladder fully collapses. The canaries have already signaled danger. The question is whether we respond in time—for everyone downstream of the next wave of change.

Sources & links (from the PDF)

The PDF contained 30 hyperlinks (some duplicated). Below are the canonical destinations we used; additional JSTOR links from the manuscript are listed for completeness.