Intro
L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
AI and the Future of Work · Introduction

Every labor-saving technology rearranges work. AI is rearranging it faster than any before.

What does that mean for your job, your team, the economy, and the next generation?

The steam engine eliminated the occupation of water carrier and created railroad engineer and factory foreman. The telephone eliminated telegraph operators and created phone-company executives. The personal computer eliminated typing pools and created software engineers. Every major technology has done this — and every time, the people whose work was displaced absorbed the majority of the cost while the gains concentrated elsewhere.

AI is the next round. The transition is compressed: tasks that were safe a year ago are automatable now, and the definition of a knowledge worker is shifting faster than institutions (employers, schools, governments) can adapt. The net number of jobs may be fine; the specific humans in specific jobs, facing specific transitions, are not.

This course is an honest assessment of work in the AI age. It covers what AI is actually automating, what it isn't, the historical pattern of automation booms, the economics of labor displacement, the specific occupations most affected, how organizations are (and aren't) handling the transition, and the questions parents, students, workers, and leaders should be asking about their own futures.

If you finish every module, here's who you become:

  • You'll understand why task automation and job automation are not the same thing — and why that distinction changes everything about how you assess your own exposure.
  • You'll be able to read labor market research critically, separating signal from noise in the headlines about AI eliminating or creating work.
  • You'll know which occupational categories face near-term disruption, what the realistic timeline looks like, and why proximity to AI doesn't guarantee safety.
  • You'll walk into any conversation about AI adoption — with your team, your leadership, or your board — able to name the human costs that optimistic productivity framing tends to erase.
  • You'll understand the centaur model of human-AI collaboration and recognize which of your current tasks are candidates for augmentation versus displacement.
  • You'll become someone who can evaluate a reskilling program or workforce transition plan and tell the difference between one that works and one that performs effort while changing nothing.
  • You'll think about AI and work the way a labor economist and an organizational leader would — simultaneously, without flinching at the tension between the two.
Module 1 · Lesson 1

The Productivity Paradox — Revisited

Every major technology wave has triggered the same fear. The numbers now arriving are different.
What does controlled research — not punditry — actually tell us about AI's impact on worker output?

In the autumn of 2023, researchers at Harvard Business School embedded a controlled trial inside BCG, one of the world's premier strategy firms. They gave 758 consultants access to GPT-4 for specific tasks and withheld it from a control group. The results landed hard enough to reshape the internal debate about AI adoption at nearly every firm that read them.

The BCG Study: Numbers That Changed the Conversation

The Fabrizio Dell'Acqua–led study, published in September 2023, found that consultants using GPT-4 completed 12.2% more tasks, did so 25.1% faster, and produced output that was rated 40% higher quality by blind evaluators — compared with the control group on the same assignments. These were not simple tasks; they included market sizing, creative ideation, and persuasive writing components drawn from real BCG client work.

Critically, the gains were largest at the lower end of the performance distribution. Consultants who had previously scored in the bottom third of the firm's internal benchmarks improved their output quality by over 43 percentage points on evaluated tasks. The study authors labelled this a "levelling up" effect — not a replacement of the weakest performers, but a compression of the performance gap between junior and senior staff.

+25%
Speed increase
BCG / Harvard, 2023
+40%
Quality increase
BCG / Harvard, 2023
+43pt
Bottom-third gain
BCG / Harvard, 2023
The MIT–Cybersecurity Analyst Study

A parallel 2023 study by Noy and Zhang at MIT assigned knowledge workers — this time, college-educated professionals writing reports, press releases, and analytical documents — either access to ChatGPT or no AI tool. Workers with AI completed assignments in 17 minutes on average versus 27 minutes in the control group. Quality ratings improved by 18 percentage points on a standardised rubric.

The finding echoed BCG's: the largest absolute gains accrued to workers who began with the weakest baseline writing skills. Workers already rated "excellent" showed moderate gains. Workers rated "poor" or "fair" at baseline showed dramatic convergence with the excellent group. The mechanism appeared to be task scaffolding — AI supplies the structural backbone of a document, freeing the worker to focus on judgment calls rather than formatting and phrasing.

Important Caveat

Both studies measured specific, clearly defined cognitive tasks over short time horizons. Neither study measured what happens to overall employment, wage levels, or firm headcount over years — only immediate task-level productivity. Extrapolating from task gains to macroeconomic outcomes requires significant additional evidence.

The Stanford Customer Service Study

In 2023, Erik Brynjolfsson and colleagues at Stanford studied a Fortune 500 company that deployed an AI assistant to 5,179 customer support agents. The tool provided real-time suggested responses drawn from a database of successful past conversations. After deployment, agents resolved 14% more issues per hour. Critically, new agents (fewer than two months on the job) improved by 35% — and attrition rates dropped, since the AI gave new hires the institutional knowledge that previously took months of trial and error to acquire.

The Stanford study introduced an economic wrinkle not present in the BCG or MIT work: the AI system was effectively transferring the tacit knowledge of the firm's best workers to its newest hires at near-zero marginal cost. This has implications for wages, training budgets, and the premium attached to experience — themes that recur throughout this module.

Key Terms
Task-level productivityOutput per unit of time on a defined, measurable task — distinct from firm-level or economy-wide productivity, which integrates many factors including capital investment, management quality, and market conditions.
Performance compressionThe narrowing of output quality gaps between low- and high-performing workers when both gain access to the same AI tool. Documented in BCG, MIT, and Stanford studies.
Tacit knowledge transferThe transmission of embedded, experience-based expertise — typically slow and costly — at scale through AI systems trained on records of expert behaviour.
Why These Studies Matter — and Their Limits

Before 2023, most claims about AI's productivity impact were either theoretical projections from economists or anecdotal reports from technology vendors with obvious conflicts of interest. These three studies — all peer-reviewed or working-paper quality, all using randomised or quasi-experimental designs — provided the first credible causal evidence from real workplaces.

Their limits are equally important. All three studied white-collar knowledge work. All three measured short-term task completion, not career trajectories. None measured whether firms retained the productivity gains or passed them to shareholders versus workers. And none controlled for the learning effects that might reverse over time as novelty wears off or as workers adapt their work style to accommodate AI. These are the open questions that the rest of this module pursues.

The Honest Summary

On narrow, well-defined cognitive tasks performed by knowledge workers, AI assistance in 2023–2024 produced real, statistically significant productivity gains of 14–40% depending on task type and worker baseline. These gains were not evenly distributed — lower-performing workers gained most. What these studies cannot tell us is whether those gains translate into job security, wage growth, or net employment increases. That question requires different evidence.

Lesson 1 Quiz

Five questions · select the best answer
1. In the 2023 BCG / Harvard study, consultants using GPT-4 produced output rated how much higher in quality by blind evaluators?
Correct. Blind evaluators rated GPT-4-assisted consultants' output 40% higher quality — one of the study's most striking findings.
Not quite. The BCG study found a 40% quality improvement on blind evaluation. The 12% and 25% figures relate to task completion and speed respectively.
2. Which group of workers showed the LARGEST gains in the BCG study?
Correct. The study documented a "levelling up" effect where bottom-third performers improved by over 43 percentage points — the largest absolute gain of any group.
Incorrect. The BCG study found that lower-performing consultants gained the most — a "levelling up" effect that compressed performance gaps across the firm.
3. The Stanford customer service study found that new agents (fewer than two months on the job) improved their resolution rate by approximately:
Correct. New agents improved by 35% — far above the firm-wide average of 14% — because the AI compressed the learning curve normally associated with months of experience.
Incorrect. The overall firm average was 14%, but new agents (under two months) gained approximately 35%, reflecting accelerated knowledge acquisition.
4. What concept describes the AI system effectively giving new hires the knowledge of experienced workers at near-zero marginal cost?
Correct. The Stanford study specifically highlighted tacit knowledge transfer — the AI encoded expert behaviour and delivered it to novices at scale.
Not quite. "Tacit knowledge transfer" is the key concept here — the AI encoded patterns from expert workers and delivered them to new agents without the usual years of mentoring or experience.
5. Which of the following is a stated limitation shared by all three studies discussed in Lesson 1?
Correct. All three studies measured immediate task-level productivity gains. None tracked how those gains translated into wages, employment levels, or long-term career trajectories.
Incorrect. The shared limitation is time horizon — all three measured short-term task performance, not careers, wages, or macroeconomic outcomes. BCG and MIT used blind evaluation; the Stanford study used firm-tracked resolution data.

Lab 1 — Interrogating the Evidence

AI research assistant · 3 exchanges to complete

Your task

You've read about three real studies measuring AI's impact on worker productivity (BCG, MIT, Stanford). Now interrogate that evidence more deeply. Ask the assistant about methodology, potential confounds, what the studies can and cannot prove, or how to apply these findings to a real workplace situation.

Starter prompts: "What's the biggest methodological weakness in the BCG study?" · "Could the performance gains disappear over time as novelty wears off?" · "How should a manager use these findings when deciding whether to roll out AI to their team?"
Research Assistant
Lesson 1 · Evidence Analysis
Welcome to Lab 1. I'm here to help you think critically about the productivity studies we covered — BCG, MIT, and Stanford. What aspect of the evidence would you like to probe? You might ask about research design, alternative explanations, or practical implications for real workplaces.
Module 1 · Lesson 2

Jobs at Risk — Separating Tasks from Occupations

The history of automation teaches that jobs rarely vanish overnight. Tasks do.
How do researchers measure AI's exposure to different jobs — and what does the exposure data actually predict about employment?

In March 2023, OpenAI researchers Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock published a paper titled "GPTs are GPTs" — a deliberate reference to economist Erik Brynjolfsson's concept of general purpose technologies. Their core claim: large language models expose workers to augmentation or automation of significant portions of their work, and this exposure is highest, paradoxically, among college-educated professionals rather than lower-wage manual workers.

The O*NET Exposure Framework

Eloundou et al. used the U.S. Department of Labor's O*NET database — which catalogues thousands of occupations by their component tasks — and had human raters plus GPT-4 itself classify each task according to whether an LLM could reduce the time to complete it by at least 50%. An occupation was classified as "exposed" if more than 50% of its tasks met this criterion.

Results: approximately 80% of the U.S. workforce has at least 10% of their tasks exposed to LLMs. About 19% of workers are in occupations where more than 50% of tasks are exposed. Critically, the most exposed occupations include mathematicians, tax preparers, financial quantitative analysts, writers and authors, and web designers — not truck drivers, janitors, or agricultural workers, whose tasks remain largely physical and therefore outside current LLM capabilities.

Exposure ≠ Displacement

The OpenAI paper is careful to note that "exposure" means AI could assist or automate a task — not that it will, or that the job will disappear. Decisions about whether to deploy AI, how to reorganise work, and whether gains accrue to workers or employers are made by humans and institutions, not by technical capability alone.

The Goldman Sachs Analysis (2023)

In a widely-read March 2023 report, Goldman Sachs economists Jan Hatzius and colleagues estimated that generative AI could automate tasks equivalent to 300 million full-time jobs globally — but immediately clarified that this figure described potential task automation, not net job loss. Their model distinguished between:

Substitution effect: Tasks currently done by humans that AI could perform instead, potentially reducing demand for labour in specific roles.
Complementarity effect: Productivity gains that expand output and demand for the goods and services workers produce, increasing total employment.
New job creation effect: Entirely new roles created by the technology itself and by firms managing and implementing AI systems.

Goldman's central scenario predicted that AI-driven productivity gains equivalent to a 7% boost to global GDP would occur over ten years, but that the net employment effect remained "highly uncertain" and dependent on policy, wage bargaining, and the pace of adoption.

80%
US workers with ≥10% tasks exposed
Eloundou et al. / OpenAI, 2023
19%
Workers with ≥50% tasks exposed
Eloundou et al. / OpenAI, 2023
+7%
Projected 10-yr GDP boost
Goldman Sachs, 2023
The Task vs. Occupation Distinction — Why It Matters

Economic historian David Autor has spent two decades documenting what he calls "task-biased technological change." His central insight, refined from 2003 onward with collaborators Levy and Murnane: technology tends to automate routine cognitive tasks (data entry, basic analysis, rule-following) while complementing non-routine cognitive tasks (judgment, creativity, persuasion, novel problem-solving) and remaining largely unable to replicate physical dexterity in unstructured environments.

Applied to AI: a radiologist's job involves tasks ranging from reviewing scan images (increasingly automated) to communicating diagnoses to anxious patients, managing clinical relationships, and making judgment calls in ambiguous cases (not easily automated). The occupation persists even as specific tasks within it are transformed. Autor's framework predicts job transformation as the primary near-term effect — not wholesale job destruction.

Historical Comparison: What the ATM Tells Us

When automated teller machines were introduced at scale in the United States from the late 1970s onward, economists predicted significant job losses in bank teller employment. The opposite occurred. By 2004, there were more bank tellers employed in the U.S. than before ATMs, according to economist James Bessen's research published in 2015. Why? ATMs reduced the cost of operating a bank branch, so banks opened more branches. Each branch needed fewer tellers, but the number of branches more than compensated. Meanwhile, the teller role itself shifted from cash handling toward customer service and financial advising — tasks ATMs could not perform.

This does not mean AI will follow the same pattern. ATMs automated one narrow mechanical task. LLMs can operate across dozens of cognitive task categories simultaneously. But the ATM case demonstrates that technological displacement of specific tasks routinely co-exists with job count stability or growth when the technology also reduces costs and expands markets.

Task-biased technological changeThe tendency of new technology to automate specific task types within an occupation rather than eliminating the occupation as a whole; associated with David Autor's labour economics research from 2003 onward.
O*NET exposure metricA classification of U.S. occupations by the fraction of their constituent tasks that AI could reduce completion time by 50% or more; developed by Eloundou et al. using Department of Labor data.
Complementarity effectThe labour-demand-increasing effect of productivity gains — when workers produce more output, demand for goods and services rises, potentially creating new employment elsewhere.

Lesson 2 Quiz

Five questions · select the best answer
1. According to Eloundou et al. (2023), approximately what percentage of U.S. workers are in occupations where more than 50% of their tasks are exposed to LLMs?
Correct. About 19% of workers are in highly exposed occupations (≥50% of tasks), while 80% have at least some exposure (≥10% of tasks).
Not quite. 19% of workers have ≥50% of tasks exposed. The 80% figure refers to workers with at least 10% of tasks exposed — a lower threshold.
2. Which of the following worker groups does the OpenAI exposure study identify as MOST exposed to LLM-based task automation?
Correct. The highest-exposure occupations are cognitive and knowledge-based — mathematicians, tax preparers, financial quantitative analysts, writers — not physical or manual roles.
Incorrect. Physical jobs remain largely unexposed because current LLMs cannot replicate manual dexterity. The highest-exposure occupations are cognitive — mathematicians, analysts, writers.
3. David Autor's "task-biased technological change" framework predicts that technology primarily causes:
Correct. Autor's framework emphasises job transformation — routine tasks within occupations are automated, shifting workers toward non-routine, judgment-heavy tasks — rather than wholesale occupation elimination.
Incorrect. Autor's framework predicts job transformation: routine tasks within occupations get automated while non-routine tasks grow in importance. Occupations persist even as their content changes.
4. The ATM case study shows that bank teller employment in the U.S. actually increased after ATM deployment, primarily because:
Correct. Lower operating costs enabled bank branch expansion. Fewer tellers per branch, but more branches overall, produced net employment growth. The teller role also shifted toward higher-value customer service tasks.
Incorrect. The mechanism was cost reduction enabling expansion: ATMs lowered branch costs, so banks opened more branches, each needing fewer tellers but adding up to more total teller jobs.
5. Goldman Sachs' 2023 analysis estimated that AI could automate tasks equivalent to 300 million full-time jobs globally. What did the report emphasise this figure represents?
Correct. Goldman explicitly distinguished substitution, complementarity, and new job creation effects. The 300M figure is a task-automation potential estimate, not a net employment forecast.
Incorrect. The report distinguished three effects — substitution, complementarity, and new job creation — and emphasised the 300M figure measures task automation potential, not confirmed net displacement.

Lab 2 — Mapping Exposure in Your Field

AI research assistant · 3 exchanges to complete

Your task

Apply the task-exposure framework from Lesson 2 to a real occupation or industry you care about. Name a job or sector, and work with the assistant to identify which specific tasks within that role have high LLM exposure versus low exposure — and what that implies for how the role might transform.

Starter prompts: "Apply the O*NET exposure framework to a paralegal's job." · "Which tasks in a marketing manager's role are most vs. least exposed to AI?" · "How should a nurse think about AI exposure — what tasks are safe and what might shift?"
Research Assistant
Lesson 2 · Task Exposure Analysis
Welcome to Lab 2. Tell me about an occupation or industry you'd like to analyse through the task-exposure lens. We'll break it down into specific tasks and classify each by LLM exposure level — then explore what the overall picture means for how that role might evolve.
Module 1 · Lesson 3

The Wage Question — Who Captures the Gains?

Productivity gains are real. Whether workers share in them is a different question entirely.
When AI raises a worker's output, does that worker earn more — or does the surplus flow elsewhere?

When researchers at the International Monetary Fund analysed wage data around AI adoption across multiple countries in 2023, they encountered a consistent pattern: productivity gains from AI disproportionately accrued to capital owners and firms in the short run, with workers seeing modest wage gains only in occupations where AI complemented rather than substituted for their skills. The distributional question — who benefits — proved as important as the productivity question — how much more gets done.

The IMF's January 2024 Analysis

IMF chief economist Pierre-Olivier Gourinchas released an analysis in January 2024 — timed to coincide with the Davos World Economic Forum — projecting that AI would affect 40% of jobs globally and up to 60% of jobs in advanced economies. The IMF's framing was explicitly distributional: in advanced economies, AI exposure is roughly split between augmentation (higher productivity, potentially higher wages) and displacement (tasks replaced, potentially lower wages or unemployment). In emerging markets, the share of jobs with high displacement risk is lower, but so is the capacity to manage the transition through social safety nets.

The report identified a specific risk: AI tends to raise the value of high-skill workers who can direct and verify AI output, while potentially depressing wages for mid-skill workers whose tasks are automated. This is different from the "hollowing out" of the middle that characterised the 1990s–2010s automation wave. AI's exposure is concentrated among cognitive rather than routine blue-collar tasks, affecting a different set of workers.

GitHub Copilot and Software Developer Wages

GitHub Copilot, launched in June 2022, provides one of the cleaner natural experiments in AI's wage effects. In a 2023 controlled study by Peng et al., programmers using Copilot completed coding tasks 55% faster on average. However, wage data for software developers did not deteriorate in 2022–2023 despite this productivity increase — according to U.S. Bureau of Labor Statistics data, median annual wages for software developers remained at approximately $120,000 and vacancy rates stayed elevated.

The likely explanation: software development faces persistent labour shortages, meaning employers cannot easily substitute AI for workers without the context and judgment needed to manage AI output. Productivity gains expanded output rather than reducing headcount. This is consistent with the complementarity effect — but it is a sector-specific outcome, not a universal law. Labour market conditions (tight vs. slack, unionised vs. non-unionised, high-skill vs. low-skill) profoundly shape whether productivity gains translate to wages.

The Monopsony Problem

When a few dominant employers control hiring in a sector, workers have limited ability to capture productivity gains through job switching or wage bargaining. Research by economists Azar, Marinescu, and Steinbaum (2020) documented significant employer market power (monopsony) in U.S. labour markets. In sectors with high monopsony concentration, AI productivity gains are more likely to accrue to employers rather than workers — irrespective of how much task efficiency improves.

The Acemoglu–Restrepo Framework

MIT economists Daron Acemoglu and Pascual Restrepo have produced the most rigorous econometric work on automation and wages. Their 2019 paper "Robots and Jobs" found that each additional industrial robot per thousand workers in a commuting zone was associated with a wage decline of 0.42% and employment decline of 0.2 percentage points. Their 2022 extension argued that most automation since 1980 has been "so-so" — generating efficiency gains that flow to capital without creating enough new tasks to restore worker incomes.

Acemoglu has been consistently more pessimistic than many technology economists about AI's wage effects. In a widely-circulated 2023 MIT working paper, he argued that the productivity gains from current AI are likely to remain concentrated in specific task types, and that without policy intervention (portability of benefits, wage supplements, antitrust enforcement against labour market concentration), the gap between AI-exposed workers and everyone else will widen.

What the Early Firm-Level Data Shows

A 2024 NBER working paper by Humlum examined Danish firms that adopted AI software tools between 2018 and 2022. The study found that adoption raised firm-level revenue and profits, but workers in adopting firms saw only a 1–3% wage increase — substantially below the productivity gains implied by the firm-level revenue data. The "pass-through" rate from AI productivity gains to worker wages appeared low, consistent with a world where firms capture most of the surplus and return it to shareholders.

The pass-through was higher in firms with works councils (formal worker representation bodies) and in sectors with sectoral collective agreements — suggesting that institutional bargaining power shapes how AI gains are distributed, not just technical exposure levels.

60%
Jobs affected in advanced economies
IMF, January 2024
+55%
Speed gain for Copilot coders
Peng et al., 2023
1–3%
Wage increase at AI-adopting firms
Humlum / NBER, 2024
Pass-through rateThe fraction of a firm's AI-driven productivity gain that is transmitted to workers as higher wages, rather than retained as profit or returned to shareholders.
MonopsonyA labour market condition where one or a few employers dominate hiring, limiting workers' ability to capture wage gains through competition; documented as widespread in U.S. labour markets by Azar et al.
So-so automationAcemoglu and Restrepo's term for technological change that raises firm efficiency without generating sufficient new tasks to compensate displaced workers, producing net wage stagnation or decline.
The Distributional Bottom Line

AI productivity gains are real and documented. Worker wage gains from AI adoption are also real — but substantially smaller than the productivity gains, and highly variable depending on labour market tightness, unionisation, institutional bargaining structures, and the specific nature of the tasks automated. The evidence does not support either "AI will make everyone richer" or "AI will impoverish workers." It supports a more uncomfortable finding: outcomes depend on the distribution of power between workers and employers, which is shaped by policy and institutions, not technology alone.

Lesson 3 Quiz

Five questions · select the best answer
1. The IMF's January 2024 analysis projected that AI would affect what percentage of jobs in advanced economies?
Correct. The IMF projected AI would affect up to 60% of jobs in advanced economies — higher than the global average of 40% — reflecting the higher concentration of cognitive work in wealthier nations.
Incorrect. The IMF's figure for advanced economies was 60%. The 40% figure applies globally; the higher proportion in advanced economies reflects their greater share of cognitive vs. manual work.
2. The Humlum (2024) NBER study of Danish firms found that workers at AI-adopting firms received wage increases of approximately:
Correct. Workers received 1–3% wage increases — modest relative to the firm-level revenue and profit gains from AI adoption, suggesting low pass-through from productivity to wages.
Incorrect. The Humlum study found only 1–3% wage increases for workers at AI-adopting firms, despite substantially larger productivity and profit gains — a low "pass-through rate."
3. Acemoglu and Restrepo's concept of "so-so automation" refers to technological change that:
Correct. "So-so automation" generates efficiency gains that primarily flow to capital, without creating the new complementary tasks that would raise labour demand and wages.
Incorrect. "So-so automation" specifically means automation that raises firm efficiency (good for capital) but fails to generate enough new task creation to compensate displaced workers (bad for labour).
4. The Humlum Danish firm study found that wage pass-through from AI productivity gains to workers was HIGHER in which types of firms?
Correct. Works councils and collective agreements gave workers institutional bargaining power to capture a larger share of AI productivity gains — suggesting distribution is shaped by institutions, not technology alone.
Incorrect. The Humlum study found higher pass-through in firms with works councils and sectoral collective agreements — demonstrating that institutional bargaining power shapes distribution of AI gains.
5. In the GitHub Copilot natural experiment, software developer wages did NOT decline despite a 55% productivity gain primarily because:
Correct. Tight labour market conditions in software meant firms used productivity gains to expand output (complementarity effect) rather than substituting AI for workers.
Incorrect. The mechanism was labour market tightness — persistent vacancies meant firms used Copilot to do more work, not fewer workers. This is a sector-specific condition, not a universal protection.

Lab 3 — The Distribution Debate

AI research assistant · 3 exchanges to complete

Your task

Lesson 3 established that AI productivity gains don't automatically translate to worker wages — pass-through depends on institutions, market power, and labour conditions. Use this lab to explore specific scenarios where the wage question plays out differently, or to challenge the evidence you've read.

Starter prompts: "In a sector with weak unions and high employer concentration, what happens to workers when AI boosts productivity?" · "Is Acemoglu too pessimistic? Make the case that workers will capture AI gains." · "What policies would actually improve pass-through of AI gains to workers?"
Research Assistant
Lesson 3 · Wage Distribution Analysis
Welcome to Lab 3. The wage question is where the evidence gets politically charged. I'll help you examine the distributional arguments from multiple angles — the pessimist case, the optimist case, and the policy levers that actually shape outcomes. What aspect of the wage debate would you like to explore?
Module 1 · Lesson 4

New Jobs, Transition Costs, and the Speed Problem

History shows new jobs emerge. History does not show they emerge fast enough, or in the right places.
What does the evidence say about how quickly displaced workers actually transition — and what happens to those who don't make it?

Between 1999 and 2011, the United States lost approximately 5.5 million manufacturing jobs — a substantial portion attributable to import competition from China following WTO accession. Economists David Autor, David Dorn, and Gordon Hanson documented in their landmark 2013 paper that workers in the most exposed local labour markets did not transition to new sectors as standard trade theory predicted. They remained in depressed regions, left the labour force, or found lower-wage service jobs. The transition costs — in lost income, health outcomes, and community cohesion — were severe and persistent. This "China Shock" episode established the evidentiary foundation for the modern understanding of transition risk.

The New Jobs Question: What AI Will Create

The World Economic Forum's "Future of Jobs Report 2023" surveyed approximately 800 companies covering 11.3 million workers across 27 industries. The report projected that by 2027, AI and automation would displace 83 million jobs globally while creating 69 million new jobs — a net loss of 14 million positions, or 2% of current employment. The new jobs projected include AI and machine learning specialists, data analysts, digital transformation specialists, cybersecurity experts, and roles in green energy transition.

The critical problem is not the aggregate number — it is the match quality. A displaced data entry clerk in a mid-sized city does not straightforwardly become an AI prompt engineer. The skills required, the geographic location of new opportunities, and the age and education profile of displaced workers all create frictions that aggregate job-count comparisons obscure.

The Speed Problem — Evidence from Past Transitions

Economic historians Daron Acemoglu and Pascual Restrepo documented in a 2018 paper that during the first Industrial Revolution, real wages for English workers stagnated for approximately 60–70 years after mechanisation began — despite dramatic increases in industrial output. This period, which economic historians call the "Engels Pause" (after Friedrich Engels' 1845 documentation of Manchester factory conditions), is the historical archetype of a transition that worked out eventually but imposed catastrophic costs on the living generation during the transition itself.

More recent evidence: the U.S. Bureau of Labor Statistics Trade Adjustment Assistance data shows that workers displaced from manufacturing jobs after trade shocks took an average of three to four years to reach stable re-employment — and many never returned to their prior wage level. A 2019 paper by Autor, Maestas, Mullen, and Strand found that workers displaced in mid-career (ages 45–54) experienced permanent earnings declines averaging 15–20% relative to comparable non-displaced workers, and significantly elevated disability application rates.

The Retraining Effectiveness Problem

Public retraining programs for displaced workers have a mixed evidence record. A landmark 2021 meta-analysis of U.S. federal retraining programs by Heckman, LaLonde, and Smith found that while classroom training showed modest positive effects for adult women, effects for adult men were near zero or negative. The implication: creating new jobs is not the same as enabling displaced workers to access those jobs. Training frictions and credential barriers are large and poorly addressed by most existing policy tools.

Where New AI Jobs Are Actually Appearing

LinkedIn's 2023 "Future of Work" report, drawing on job posting data from over 950 million users, identified the fastest-growing job categories associated with AI: AI prompt engineers, machine learning operations (MLOps) engineers, AI ethics officers, AI-augmented healthcare coordinators, and AI content editors. These roles are concentrated in three metropolitan areas — San Francisco/Silicon Valley, New York, and Seattle — and require programming, data science, or graduate education backgrounds for the technical roles.

The geographic and educational concentration matters enormously. A call-centre worker in a mid-sized Midwestern city whose job is automated by an LLM is not in a realistic position to relocate to San Francisco to become a prompt engineer. Geographic immobility — driven by homeownership, family ties, and community anchors — is one of the most robust findings in labour economics, and it is a central reason why aggregate job-count optimism does not translate into individual worker security.

The Unique Speed of Current AI Diffusion

ChatGPT reached 100 million users in 60 days after its November 2022 launch — the fastest product adoption in recorded history. For context, it took the telephone 75 years to reach 100 million users; the internet, 4 years; Instagram, 2.5 years. The pace of LLM diffusion into enterprises accelerated even faster: McKinsey's 2023 survey of 1,700 organisations found that 79% of respondents had had at least some personal exposure to generative AI, and 22% were using it regularly in their work. One year earlier, these figures were effectively zero.

This speed creates a specific challenge: labour markets, educational systems, and policy institutions that adapted to previous technological transitions over decades are now being asked to adapt over months. The question is not whether transition eventually occurs — it will — but whether the institutional infrastructure for managing that transition exists at the required speed.

60 days
ChatGPT to 100M users
OpenAI / UBS, 2023
–14M
Projected net job change by 2027
WEF Future of Jobs, 2023
15–20%
Permanent wage loss, mid-career displacement
Autor et al., 2019
The Engels PauseThe 60–70 year period during the British Industrial Revolution in which workers' real wages stagnated despite dramatic increases in industrial productivity — the archetype of a technological transition whose gains arrived much later than its costs.
China ShockThe documented labour market disruption in U.S. manufacturing communities following China's WTO accession in 2001; Autor, Dorn, and Hanson's research showed displaced workers did not transition as standard economic models predicted.
Geographic immobilityThe empirical tendency of workers to remain in depressed local labour markets rather than relocate to areas with better opportunities — a major friction preventing displaced workers from accessing new jobs created by technological change.
Module 1 Synthesis

What the evidence shows, taken together: AI is producing real productivity gains on specific cognitive tasks; those gains are currently concentrated in knowledge work; jobs will transform rather than uniformly disappear; wage pass-through from productivity to workers is low and institutionally contingent; and the speed of transition is unprecedented, which is the central challenge. None of this is reason for either complacency or panic — but it is a precise set of claims, grounded in documented data, that should anchor every subsequent discussion of what to do.

Lesson 4 Quiz

Five questions · select the best answer
1. The WEF "Future of Jobs Report 2023" projected that AI would create 69 million new jobs while displacing 83 million — a net change of:
Correct. The WEF projected a net loss of approximately 14 million positions — but the lesson emphasised that aggregate job counts obscure critical issues of skills matching and geography.
Incorrect. 83M displaced minus 69M created equals a net loss of 14 million jobs, or about 2% of current employment according to the WEF report.
2. The "Engels Pause" describes which historical phenomenon?
Correct. The Engels Pause names the decades-long wage stagnation during British industrialisation — a warning that even transitions that "work out eventually" can impose severe costs on the living generation.
Incorrect. The Engels Pause refers to the extended period (~60–70 years) during which British workers' real wages stagnated despite dramatic industrial output growth — the historical archetype of a painful technology transition.
3. Research on workers displaced in mid-career (ages 45–54) found they experienced permanent earnings declines averaging:
Correct. Autor, Maestas, Mullen, and Strand (2019) found 15–20% permanent earnings declines for mid-career displaced workers — not a temporary dip but a lasting downward step change.
Incorrect. The Autor et al. (2019) study found permanent earnings declines of 15–20% for workers displaced at ages 45–54 — a substantial and lasting effect on lifetime income.
4. ChatGPT reached 100 million users in 60 days. Which of the following BEST describes why this speed matters for the AI-labour transition?
Correct. The unprecedented diffusion speed creates a specific institutional mismatch: the systems built to manage technological transitions were designed for decade-scale change, not the current pace.
Incorrect. The concern is institutional mismatch: retraining programs, educational pathways, and social safety nets were built for transitions that unfold over years or decades, not months. Speed amplifies transition risk.
5. LinkedIn's 2023 data showed that new AI-related jobs are geographically concentrated in San Francisco, New York, and Seattle. Why does this concentration matter for displaced workers in other regions?
Correct. Geographic immobility — driven by homeownership, family ties, and community anchors — is one of the most robust findings in labour economics. New jobs in San Francisco don't help a displaced worker in Ohio.
Incorrect. Geographic immobility is a major labour economics finding: workers tend to stay in depressed local markets rather than relocate. Aggregate job creation in distant cities does not automatically benefit locally displaced workers.

Lab 4 — Designing for Transition

AI research assistant · 3 exchanges to complete

Your task

Lesson 4 identified the speed problem, the retraining gap, and geographic immobility as the central friction points in the AI-labour transition. In this lab, think through what responses — from individuals, firms, or governments — the evidence actually supports. Be specific; vague "more training" answers don't engage seriously with the retraining effectiveness data.

Starter prompts: "Given the weak evidence on retraining programs, what should a firm actually do for workers displaced by AI it introduces?" · "What does the Engels Pause imply about the political risks of the current AI transition?" · "Design a policy response to the geographic concentration of AI jobs that takes immobility seriously."
Research Assistant
Lesson 4 · Transition Policy Analysis
Welcome to Lab 4. This is where evidence meets prescription — always the hardest move in policy thinking. I'll push back on vague recommendations and try to anchor your proposals in the specific friction points the research has identified: retraining effectiveness, geographic immobility, speed mismatches, and institutional bargaining gaps. What transition challenge would you like to tackle?

Module 1 Test

15 questions · 80% required to pass
1. The BCG / Harvard 2023 study found that GPT-4-assisted consultants completed tasks how much faster than the control group?
Correct. Consultants with GPT-4 worked 25.1% faster and completed 12.2% more tasks.
The BCG study found a 25.1% speed increase. The 40% figure refers to quality gains, and 55% is from the Copilot study.
2. In the MIT Noy and Zhang study, workers with ChatGPT completed writing tasks in an average of how many minutes?
Correct. The MIT study found 17 minutes with AI versus 27 minutes in the control group — a substantial time reduction.
The MIT study found 17 minutes with AI versus 27 without, representing a roughly 37% time reduction.
3. The Stanford customer service AI study covered approximately how many agents at the company studied?
Correct. The Stanford Brynjolfsson study covered 5,179 customer support agents at a Fortune 500 company.
The Stanford study covered 5,179 agents. The 758 figure is from the BCG study.
4. According to the OpenAI / UPenn exposure study, which occupation group has the HIGHEST LLM task exposure?
Correct. High-exposure occupations are concentrated in cognitive, white-collar work — not physical or manual roles.
Physical occupations have low LLM exposure. The highest-exposure roles are knowledge-intensive: mathematicians, financial analysts, writers.
5. What does "performance compression" mean in the context of AI workplace studies?
Correct. Performance compression describes AI's levelling-up effect — lower performers gain proportionally more, narrowing the gap with top performers.
Performance compression refers to the narrowing of quality gaps between strong and weak performers when both gain access to AI — documented in BCG, MIT, and Stanford studies.
6. Goldman Sachs' 2023 analysis identified three effects of AI on employment. Which of the following is NOT one of them?
Correct. Goldman Sachs identified substitution, complementarity, and new job creation effects. "Skill premium amplification" is not part of their framework.
Goldman Sachs' three effects were substitution, complementarity, and new job creation. Skill premium amplification was not in their framework.
7. David Autor's research from 2003 onward on "task-biased technological change" found that technology most consistently automates:
Correct. Autor's framework identifies routine cognitive and manual tasks as most vulnerable, with non-routine judgment and creativity tasks being complemented rather than replaced.
Autor found that routine cognitive tasks are most automated, while non-routine tasks involving judgment and creativity are complemented — the pattern that shapes his predictions about AI.
8. The IMF January 2024 report identified a specific distributional risk not present in earlier automation waves. What was it?
Correct. Unlike 1990s automation that hollowed out routine blue-collar work, AI's highest exposure is among cognitive and professional workers — a different and politically novel risk distribution.
The IMF's key distributional finding: AI exposure is concentrated among knowledge workers (a different group than prior automation waves), raising novel political and social dynamics.
9. Acemoglu and Restrepo's "Robots and Jobs" (2019) found that each additional industrial robot per thousand workers was associated with a wage change of approximately:
Correct. Each additional robot per thousand workers was associated with a –0.42% wage decline and –0.2 percentage point employment decline in the affected commuting zone.
Acemoglu and Restrepo found a –0.42% wage decline per additional robot per thousand workers — a negative effect on worker wages from automation.
10. The Humlum Danish firm study found that higher wage pass-through from AI gains was associated with:
Correct. Institutional worker representation — works councils and collective agreements — was the key variable associated with higher wage pass-through from AI productivity gains.
The Humlum study found that works councils and collective agreements predicted higher wage pass-through — institutional bargaining power shapes distribution of AI gains.
11. The "China Shock" research by Autor, Dorn, and Hanson demonstrated that workers displaced from manufacturing:
Correct. The China Shock research showed persistent local labour market depression, labour force exit, and downward wage mobility — not the smooth reallocation that economic models predicted.
Autor, Dorn, and Hanson found the opposite of smooth transition: persistent regional depression, workforce exit, and lower-wage reemployment — a warning for AI-era displacement.
12. The WEF's "Future of Jobs Report 2023" covered approximately how many workers across 27 industries?
Correct. The WEF 2023 report covered ~800 companies representing 11.3 million workers across 27 industries.
The WEF 2023 report surveyed approximately 800 companies covering 11.3 million workers across 27 industry sectors.
13. Which product reached 100 million users faster than any previous technology in recorded history?
Correct. ChatGPT reached 100 million users in 60 days after its November 2022 launch — the fastest product adoption on record.
ChatGPT reached 100 million users in 60 days after its November 2022 launch — beating Instagram (2.5 years) and all other previous technology products.
14. The meta-analysis by Heckman, LaLonde, and Smith on U.S. federal retraining programs found that for displaced adult men, effects on re-employment were:
Correct. For adult men, the meta-analysis found near-zero or negative effects from classroom-based retraining programs — a sobering finding for policy proposals centred on "just retrain workers."
The Heckman et al. meta-analysis found near-zero or negative effects for adult men from federal retraining programs — a major challenge for policies assuming retraining solves AI displacement.
15. Which of the following BEST summarises Module 1's core synthesis of the evidence on AI and work?
Correct. This is the four-part synthesis: productivity gains are real and documented; transformation rather than wholesale job destruction is the near-term pattern; wage pass-through is low and institutionally contingent; speed of change creates unprecedented institutional mismatch.
The module's synthesis avoids both panic and complacency. The honest picture: real productivity gains, job transformation over destruction, low wage pass-through, and the speed of transition as the central challenge requiring institutional response.