HomeCountiesArticlesMethodologyContactThe Dozen Terms You Need to Read the Debate — and Your Own County — Without Getting Fooled
Every field protects itself with vocabulary, and the economics of artificial intelligence is worse than most: it borrows jargon from labor economics, computer science, urban planning, and climate science all at once, then uses the same words to mean different things depending on who is talking. A consultant says your county has "high exposure" and a council member hears "we are going to lose jobs" — which is not what the word means. This glossary is built to fix that. It defines the terms a municipal leader actually encounters, in plain language, with the real research behind each one, so that you can read an analyst’s report, a newspaper story, or your own CAERI county page and know exactly what is being claimed — and, just as usefully, what is not. Keep it on the desk. Cite it in a memo. The goal is a shared language, because you cannot plan around a word your team defines three different ways.
The single most misunderstood term in the field. Exposure measures the degree to which the tasks that make up a job overlap with what AI can currently do. It is built by breaking occupations into their component tasks — a framework pioneered by Autor, Levy, and Murnane (2003) — and scoring how many of those tasks a given AI capability could perform. The leading measures come from Felten, Raj, and Seamans (2021), Webb (2020), and Eloundou and colleagues at OpenAI, whose 2024 Science paper estimated that around 80 percent of U.S. workers have at least 10 percent of their tasks exposed to large language models (Eloundou et al., 2024).
Here is the part everyone gets wrong, stated by the exposure researchers themselves: exposure is a proxy for potential impact and deliberately does not distinguish between whether AI will augment a worker or replace them (Eloundou et al., 2024). A high exposure score means AI can touch the work — not that the job disappears. A radiologist and a court reporter can both be "highly exposed" while one’s job is made more productive and the other’s is automated away. Exposure tells you the water is nearby. It does not tell you the flood is coming.
Borrowed, fittingly, from climate-resilience science, adaptive capacity is the ability of a system — here, a local economy — to adjust to disruption, moderate the damage, and take advantage of new opportunities. The concept is central to the resilience frameworks used by bodies like the Intergovernmental Panel on Climate Change, and it is what separates two equally-exposed communities into very different fates. A county with a diverse employer base, an educated and mobile workforce, strong institutions, and healthy public finances can absorb a shock that would flatten a concentrated, thinly-staffed neighbor. Exposure is how close you stand to the hazard; adaptive capacity is how well you can move when it arrives. Any honest analysis of AI and a community has to measure both — which is precisely why they sit as separate pillars in a composite index rather than being mashed into one scary number.
The distinction that determines whether AI is a raise or a pink slip. Automation means a machine performs a task instead of a person; augmentation means a machine helps a person perform a task better or faster. Economists Daron Acemoglu and Pascal Restrepo have built much of the modern framework here, showing that automation displaces labor and pushes down its share of income, while new tasks and augmenting technologies can create labor demand and raise it (Acemoglu & Restrepo, 2019). The same tool can do both: a translation model that replaces a staff translator (automation) also lets a small exporter reach customers it never could before (augmentation). Which effect dominates is not a property of the technology — it is a choice made by firms, shaped by policy, incentives, and skills. This is why "AI will automate X" is always an incomplete sentence. Automate, or augment? For whom? On what timeline?
Why the middle falls out. Employment polarization is the well-documented tendency of technology to hollow out middle-skill, routine jobs — bookkeeping, clerical work, repetitive production — while employment grows at both the high-skill end (managers, engineers) and the low-skill manual end (care work, food service, cleaning) that is hard to automate. David Autor and David Dorn documented the pattern in the U.S. labor market, and Maarten Goos and Alan Manning gave it its most memorable name: the growth of "lovely and lousy jobs" at the expense of the ones in between (Autor & Dorn, 2013; Goos & Manning, 2007). For a municipal leader the lesson is sharp: technology does not lower all boats or raise all boats — it can drain the middle of the harbor, which is exactly where a lot of your stable, family-supporting jobs sit.
The reason jobs clump. Agglomeration economies are the productivity benefits that firms and workers gain from being near one another — shared suppliers, a deep labor pool, and the fast spread of ideas. The idea dates to the economist Alfred Marshall in 1890, and Enrico Moretti’s work has shown how powerfully it shapes the modern map, with a handful of "brain hub" regions pulling ahead precisely because talent and firms concentrate there (Moretti, 2012). Agglomeration is why economic gains from a technology do not spread evenly across the map, and why some regions capture the upside of AI while others get only the disruption. It is also the force a struggling county is fighting when it tries to build a new cluster from scratch — and the reason place-based investment succeeds only when it manages to seed a self-reinforcing agglomeration rather than a subsidy that evaporates.
Why your workers’ jobs are not where they sleep. A commuting shed is the real geography of a local labor market — the area across which people actually travel to work — which almost never matches a city or county line. It matters enormously for exposure analysis, because the jobs located inside your borders (workplace exposure) can look very different from the jobs held by the people who live there (residence exposure). A bedroom community whose residents commute to exposed office jobs in the next county carries risk that a map of local employers would completely miss. Any serious look at a community’s exposure has to reckon with where its residents actually work — not just what sits inside the municipal boundary.
The old engine, named. Coined by economist Joseph Schumpeter in 1942, creative destruction is capitalism’s process of continuously replacing old industries and jobs with new ones — the automobile destroying the carriage trade, the spreadsheet transforming accounting. It is the optimist’s favorite term, and it contains a real truth: over the long run, this churn has raised living standards, and more than 60 percent of the jobs Americans work today are in occupations that barely existed in 1940. But the phrase hides a brutal detail in the word "destruction." The new jobs and the destroyed ones rarely belong to the same people, in the same places, at the same time. Creative destruction is a comfort at the national scale and a catastrophe at the local one — which is exactly the gap local government exists to manage.
Why one lost job is never just one. Displacement is the involuntary loss of a job to economic or technological change; the classic finding, from Jacobson, LaLonde, and Sullivan (1993), is that displaced workers suffer large and persistent earnings losses that can last decades. The multiplier is the amplifier: each "traded" or anchor job supports additional local jobs through supplier and consumer spending, so losing it removes more than one job from the community. Research on trade-driven displacement found that local employment fell by roughly two additional workers for every directly displaced one (Autor, Dorn & Hanson, 2013). The multiplier is why a single plant closing can take a whole town down with it, and why exposure concentrated in a few large employers is more dangerous than the same exposure spread thin.
Why the things AI can’t do get expensive. Named for economist William Baumol, cost disease describes how wages rise even in sectors with little productivity growth, because they must compete for workers with sectors where productivity is soaring. It explains a counterintuitive feature of the AI economy: as AI makes cognitive work dramatically cheaper, the human-intensive services it cannot easily do — hands-on healthcare, skilled trades, in-person care — may take up a larger and larger share of what households spend. For a community, cost disease is a hint about where durable, hard-to-automate employment and rising costs will both concentrate, and why "just move everyone into care work" is not the free lunch it appears to be.
The predecessor idea. Skill-biased technical change is the theory that technology tends to raise demand for higher-skilled workers faster than for lower-skilled ones, widening the wage gap between them — the framework Lawrence Katz and Kevin Murphy used to explain rising wage inequality in the late twentieth century. It is worth knowing partly because AI may complicate it: some evidence suggests generative AI helps less-experienced workers most on certain tasks, potentially compressing rather than widening gaps in those settings. Whether AI is skill-biased in the old direction or something genuinely new is one of the live questions in the field — and a reminder that the last technology’s rulebook does not automatically apply to this one.
Why the robot can pass the bar exam but can’t fix your toilet. Named for roboticist Hans Moravec, the paradox is the long-observed finding that tasks humans find hard — abstract reasoning, calculation, passing professional exams — are comparatively easy to automate, while tasks humans find effortless — perception, dexterity, moving through a cluttered physical world — are extraordinarily hard (Moravec, 1988). It is the single most useful idea for working out which local jobs are exposed soonest. It is why high-paid cognitive work is on the front line of the current wave while a plumber crawling through a ninety-year-old basement is, for now, among the safest workers in your county. It also carries a warning: "for now" is doing real work in that sentence, because the frontier of what machines can perceive and manipulate keeps moving.
Read in sequence, the terms tell a coherent story. A technology arrives that can perform certain tasks (exposure), and whether it lifts or replaces a given worker depends on choices (augmentation vs. automation). Its effects fall unevenly across skill levels (polarization, skill-biased change) and across places (agglomeration, commuting sheds), amplified when jobs are concentrated (the multiplier) and cushioned where communities can adjust (adaptive capacity). Over the long run the churn has paid off (creative destruction) while inflicting real, lasting, local harm along the way (displacement) — and the work least touched is often the physical work we least expected (Moravec’s paradox). That whole story is what a composite exposure-and-resilience index is built to capture in a single, comparable picture, which is why no one number in it means much on its own.
For how those measurements should and should not be read — see Position, Not Prophecy. For why the time horizon changes every one of these terms’ implications, see Mind the Clock. And for how polarization and the absorption of displaced workers play out in practice, see Painful Growth.
Vocabulary is not a side issue in this debate — it is where most of the confusion, and a good deal of the fear, actually lives. A leader who knows that "exposure" is not a forecast, that "automation" is a choice and not a destiny, and that "creative destruction" is a national comfort but a local emergency is already better equipped than most of the people who will brief them. The words are the instruments. Learn to read the dials, and the rest of the conversation — in the newspaper, in the consultant’s deck, on your own county page — gets a great deal clearer.
Every term in this glossary shows up on a CAERI county page — exposure, adaptive capacity, concentration, and more, scored for your community and explained in plain language. See your county’s numbers →
Acemoglu, D., & Restrepo, P. (2019). "Automation and New Tasks: How Technology Displaces and Reinstates Labor." Journal of Economic Perspectives, 33(2), 3–30.
Autor, D., & Dorn, D. (2013). "The Growth of Low-Skill Service Jobs and the Polarization of the US Labor Market." American Economic Review, 103(5), 1553–1597.
Autor, D., Dorn, D., & Hanson, G. (2013). "The China Syndrome: Local Labor Market Effects of Import Competition in the United States." American Economic Review, 103(6), 2121–2168.
Autor, D., Levy, F., & Murnane, R. (2003). "The Skill Content of Recent Technological Change: An Empirical Exploration." Quarterly Journal of Economics, 118(4), 1279–1333.
Eloundou, T., Manning, S., Mishkin, P., & Rock, D. (2024). "GPTs Are GPTs: Labor Market Impact Potential of LLMs." Science, 384(6702), 1306–1308.
Felten, E., Raj, M., & Seamans, R. (2021). "Occupational, Industry, and Geographic Exposure to Artificial Intelligence: A Novel Dataset and Its Potential Uses." Strategic Management Journal, 42(12), 2195–2217.
Goos, M., & Manning, A. (2007). "Lousy and Lovely Jobs: The Rising Polarization of Work in Britain." Review of Economics and Statistics, 89(1), 118–133.
Jacobson, L., LaLonde, R., & Sullivan, D. (1993). "Earnings Losses of Displaced Workers." American Economic Review, 83(4), 685–709.
Katz, L., & Murphy, K. (1992). "Changes in Relative Wages, 1963–1987: Supply and Demand Factors." Quarterly Journal of Economics, 107(1), 35–78.
Marshall, A. (1890). Principles of Economics. Macmillan (origin of agglomeration economies).
Moravec, H. (1988). Mind Children: The Future of Robot and Human Intelligence. Harvard University Press.
Moretti, E. (2012). The New Geography of Jobs. Houghton Mifflin Harcourt.
Schumpeter, J. (1942). Capitalism, Socialism and Democracy. Harper & Brothers (origin of "creative destruction").
Webb, M. (2020). "The Impact of Artificial Intelligence on the Labor Market." Working paper, SSRN.
On Baumol’s cost disease: Baumol, W. J. (1967), "Macroeconomics of Unbalanced Growth," American Economic Review, 57(3), 415–426.