Evaluate the Impact of AI on Employment and Society

Artificial intelligence (AI) affects employment and society through a mix of capability gains, task reallocation, productivity changes, and institutional responses (labor markets, education, regulation, and social protection). Evaluating impact therefore requires distinguishing job displacement from task substitution, short-run disruption from long-run adjustment, and private gains from societal outcomes.2

To ground the evaluation, this section uses a commonly cited framing: AI can change what workers do by automating some tasks, augmenting other tasks, and creating new roles—while the net effect depends on adoption speed, market demand, skill distribution, and policy choices.2

Key terms used throughout:

• Task substitution
• Job polarization
• Labor market frictions
• Productivity effect
• Distributional impact

1) How AI changes employment: tasks, not just jobs

A rigorous evaluation starts at the task level. Many studies emphasize that AI (and automation broadly) affects which tasks are performed by humans versus machines, rather than eliminating entire occupations uniformly.2 When tasks are modular, AI can substitute for routine or structured work, while human roles shift toward tasks needing judgment, social interaction, or oversight.

Mechanisms to evaluate

1. Automation (substitution): AI reduces the need for certain task inputs (e.g., transcription, basic screening, routine analysis).
2. Augmentation (complementarity): AI increases productivity for workers who supervise, interpret, or apply tools—potentially raising earnings for some skills.2
3. Creation via demand: New AI-enabled products/services can expand overall economic activity and create new roles (including AI-adjacent occupations).
4. Reorganization inside firms: AI can change internal workflows, management practices, and timing of work—affecting hiring patterns and contract structures.2

Practical evaluation lens (what to measure)

• Task coverage: How many task components in an industry are AI-addressable?
• Adoption intensity: What fraction of firms adopt, and how quickly?
• Worker reallocation speed: How fast can workers move to changing tasks?
• Wage adjustment: Do earnings rise for those with complementary skills, and fall for those displaced?
• Equity: Who bears the burden—by skill, region, gender, or race?

2) Net employment effects: why predictions vary

Public discussion often oscillates between “mass job loss” and “net job creation.” The reason is that AI effects depend on interacting forces:

• Substitution vs complementarity: Whether AI substitutes for labor or increases productivity of labor. Evidence consistently suggests task substitution is real, while labor augmentation may offset some losses depending on skill and workflow design.
• Firm and sector heterogeneity: Some sectors adopt faster or benefit more, so impacts differ across industries and occupations.
• Time horizon: Short-run displacement can occur before new demand and job creation materialize—especially where retraining is slow.
• Productivity-to-jobs transmission: If AI boosts profits without translating into expanded employment, outcomes may be weaker job growth and higher inequality.2

Evidence anchors to use in your evaluation

The World Economic Forum’s Future of Jobs provides estimates of job shifts across categories (roles declining vs emerging) and emphasizes that transformation is expected to be gradual and varies by region and industry. Meanwhile, ILO materials highlight that automation/AI changes must be managed with skills development and social protection to avoid worsening inequality. OECD work similarly stresses careful distinction between exposure and realized displacement.

3) Society-wide impacts: inequality, democracy, and cohesion

Beyond employment counts, AI reshapes society through how benefits and risks are distributed, how institutions respond, and how information ecosystems change.

A) Inequality and social mobility

If AI rewards high-skill labor and capital more than mid/low-skill labor, it can worsen income inequality and slow mobility—especially where education systems and labor markets cannot adjust quickly.2 Evaluations should include:

• wage distribution changes,
• employment stability (volatility, contracts, gigification),
• regional disparities,
• barriers to retraining.

B) Work quality and bargaining power

Even if aggregate employment is stable, AI can alter job quality:

• surveillance and performance scoring,
• algorithmic scheduling,
• changes to bargaining dynamics,
• “deskilling” if AI standardizes tasks and reduces worker autonomy.

C) Trust, legitimacy, and governance

Societal acceptance depends on transparency, accountability, and protection of fundamental rights. The EU AI Act establishes risk-based obligations for certain AI systems, including transparency and governance requirements—an example of how regulation can shape real-world deployment and labor impact.

4) Skills and education: the workforce adjustment channel

A central employment-impact pathway is whether workers can transition from affected tasks to complementary tasks. ILO emphasizes that skills development is key to managing technology transitions, and policy should support lifelong learning rather than relying only on market forces.

In evaluating impact, treat training not as a “nice-to-have,” but as an input to labor-market adjustment:

• Relevance: training must map to new task requirements (human oversight, domain expertise with AI tools).
• Access: ensure affordability and support for displaced workers.
• Timing: programs must start before large-scale displacement peaks.
• Evaluation: measure completion rates and job transitions.

5) Policy and institutional levers that change outcomes

Because AI effects are mediated by institutions, evaluation should include policy scenarios.

Common levers

1. Active labor market policies: job matching, wage subsidies, retraining pathways.
2. Social protection: unemployment benefits and income support that bridge transitions.
3. Education reform: curriculum updates and apprenticeship-like models for AI tool workflows.
4. Regulation and accountability: risk-based rules that constrain harmful deployment and require transparency (e.g., EU AI Act structure).
5. Sectoral bargaining and worker voice: frameworks that support negotiation over deployment, monitoring, and job redesign.

How to incorporate policies into evaluation

Use a scenario approach:

• Baseline: current policy environment.
• Reform 1: stronger retraining + job placement.
• Reform 2: stronger social protection.
• Reform 3: governance/transparency requirements for high-risk workplace AI. Compare predicted changes in unemployment duration, wage inequality, and labor force participation.