Case Study 2 — AI and the Future of Work: Three Scenarios

In early 2023, OpenAI released ChatGPT-4 — an AI system that could write essays, analyze data, generate code, create marketing copy, summarize legal documents, and converse on virtually any topic. Within months, similar systems from Google (Gemini), Anthropic (Claude), Meta (Llama), and others had entered the market. By 2025, AI tools were embedded in word processors, spreadsheets, design software, coding environments, customer service platforms, and dozens of other workplace tools.

The labor-market implications are enormous and uncertain. This case study presents three scenarios for how AI might affect work over the next two decades — each consistent with some evidence and each with different policy implications.

Scenario 1 — The Augmentation Scenario (optimistic)

The thesis: AI makes human workers more productive but doesn't replace them. Just as spreadsheets made accountants more productive (but didn't eliminate accounting jobs) and CAD software made architects more productive (but didn't eliminate architecture jobs), AI tools make knowledge workers more productive without replacing them.

What this looks like: - Lawyers use AI to do legal research in minutes instead of hours — but clients still need lawyers to exercise judgment, argue in court, and advise on strategy - Doctors use AI to read radiology scans faster — but patients still need doctors to explain diagnoses, make treatment decisions, and provide care - Programmers use AI to write boilerplate code — but companies still need programmers to design systems, debug complex problems, and make architectural decisions - Writers use AI to draft and edit — but publishers still need writers who bring voice, insight, and creativity

The labor-market prediction: demand for human workers stays roughly constant or rises (because AI-augmented workers are more productive and can serve more customers, expanding the total market). Wages rise for workers who learn to use AI effectively. Wages stagnate or fall for workers whose tasks can be fully automated but who don't acquire AI-complementary skills.

Evidence for: most studies of AI adoption in 2023–2025 show productivity gains but not mass layoffs. A 2024 study by MIT economists found that ChatGPT increased writing productivity by 40% for business professionals — but employers did not lay off 40% of their writers. They used the productivity gain to produce more content, serve more clients, and expand into new areas.

Policy implication: invest in training workers to use AI tools. The biggest risk is a skills gap — workers who can use AI become very productive and well-paid; workers who can't fall behind.

Scenario 2 — The Displacement Scenario (pessimistic)

The thesis: AI replaces large categories of human work, creating prolonged unemployment for millions of workers who can't easily transition to AI-proof jobs.

What this looks like: - Customer service: AI chatbots handle 80% of customer interactions; human agents are needed only for complex escalations. Call-center employment drops by 50–70%. - Data entry and processing: AI automates most data-entry, form-processing, and routine analytical tasks. Clerical employment drops substantially. - Translation: AI translation quality approaches human quality for most languages. The translation industry shrinks. - Basic legal and financial services: AI handles contract drafting, tax preparation, basic legal advice, and routine financial planning. Entry-level positions in law and finance shrink. - Content creation: AI generates basic news articles, marketing copy, social media posts, and product descriptions. Demand for entry-level writers, copywriters, and content creators falls.

The labor-market prediction: total demand for human workers falls, at least in the medium term (5–15 years). Some workers transition to AI-complementary roles; many do not. Unemployment rises, particularly among middle-skill workers. Inequality widens: AI-owners and AI-complementary workers earn more; displaced workers earn less.

Evidence for: some early indicators are concerning. In 2024, several major employers (including tech firms, media companies, and financial institutions) announced layoffs specifically attributed to AI adoption. The Freelancer's Union reported declining rates for freelance writing and graphic design — sectors where AI tools directly compete with human labor.

Policy implication: strengthen the safety net (unemployment insurance, retraining programs, possibly universal basic income). The displaced workers can't all become AI engineers; the transition needs to be managed.

Scenario 3 — The Polarization Scenario (the mixed middle)

The thesis: AI accelerates the labor-market polarization that has been underway since the 1980s (Chapter 13). High-skill workers who complement AI become much more productive and earn much more. Low-skill workers in in-person services (care, cleaning, food, physical delivery) are hard to automate and see moderate demand. Middle-skill workers in routine cognitive tasks (data processing, clerical work, basic analysis) are most at risk.

What this looks like: - High-skill jobs (medicine, law, engineering, management, creative work): AI augments, doesn't replace. Workers who master AI tools earn substantially more. Demand rises. - Low-skill in-person jobs (eldercare, childcare, housekeeping, food preparation, physical delivery, construction): hard to automate (require physical presence and dexterity). Demand stays constant or rises (aging population increases care demand). Wages stay low because supply is abundant. - Middle-skill routine jobs (data entry, bookkeeping, call centers, basic writing, basic analysis, administrative support): most vulnerable to AI replacement. Employment falls. Workers displaced from these jobs face downward mobility.

The labor-market prediction: the "barbell" distribution. Good jobs at the top, decent jobs at the bottom (but poorly paid), and a hollowed-out middle. This is consistent with the pattern that has been developing since the 1980s — AI accelerates it.

Evidence for: the strongest empirical support. The Autor (2019) framework of "work of the past, work of the future" describes exactly this pattern. The data on occupational employment growth since 2000 shows polarization: growth in high-skill professional occupations, growth in low-skill service occupations, and decline in middle-skill routine occupations. AI is expected to intensify this pattern.

Policy implication: the hardest to design. The displaced middle-skill workers can't all go "up" (not everyone can become a data scientist) and shouldn't have to go "down" (from $50K/year clerical work to $30K/year care work). Policy needs to: (a) invest in education and retraining for the workers who can move up, (b) improve wages and conditions for the service jobs that can't be automated (minimum wage, EITC, benefits), and (c) maintain a strong safety net for the transition period.

What economics can and cannot tell us

Economics can tell us: - The direction of change: AI will change the composition of jobs (some tasks automated, some augmented, some unaffected) - The groups most at risk: middle-skill workers in routine cognitive tasks - The policy toolkit: education, retraining, safety nets, minimum wage, EITC, possibly UBI - The historical precedent: previous technological revolutions displaced workers in the short run but created new jobs in the long run

Economics cannot tell us: - The magnitude of displacement: nobody knows whether AI will affect 10% of jobs or 50% - The speed of transition: will it take 5 years or 25? - Whether this time is different: AI's breadth (cognitive + creative + analytical + physical) is unprecedented - What the new jobs will be: by definition, we can't predict the jobs that don't exist yet

The honest position: prepare for significant disruption, invest in the tools that help workers adapt, strengthen the safety net, and watch the data carefully. The worst outcome is not the automation itself — it's the failure to prepare for it.

Discussion questions

1. Which of the three scenarios do you find most plausible? Why?
2. The "augmentation" scenario is optimistic but requires workers to learn new skills. What happens to workers who can't or won't adapt?
3. The "displacement" scenario could justify a universal basic income (UBI). Is UBI a good policy response to AI displacement? What are the tradeoffs?
4. The "polarization" scenario predicts a hollowed-out middle. If you're a college student choosing a career, how does this affect your decision?
5. The historical precedent (previous automation created as many jobs as it destroyed) is often cited by optimists. Is this precedent reliable for AI? What's different this time?