AI Is Making High-Paid Workers Richer and Low-Paid Workers Poorer — Here Is the Evidence
A 2015–2022 US study using instrumental variables finds automation AI cuts jobs and wages for low-skilled workers, while augmentation AI creates new roles and raises pay for the high-skilled. AI may be widening the wage gap.
What if the same technology is simultaneously creating jobs for some workers and destroying them for others — and the split falls exactly along the lines you would expect?
A rigorous study of US labor market data from 2015 to 2022 has found precisely that [Fact]. Automation-focused AI is reducing employment and wages in low-skilled occupations, while augmentation-focused AI is generating new work and boosting pay in high-skilled roles. The result is not a rising tide lifting all boats. It is a widening chasm.
If you work in a routine, process-heavy role, this research has uncomfortable implications. If you work in a creative, analytical, or strategic role, the news is better — but the bigger picture should concern everyone.
Two Types of AI, Two Very Different Outcomes
Economist David Marguerit developed a novel framework for measuring AI exposure that distinguishes between two fundamentally different flavors of artificial intelligence [Fact].
Automation AI replaces human labor. It handles tasks that workers used to perform — think document processing, basic data analysis, routine customer inquiries. When automation AI enters an industry, it substitutes for workers, particularly those performing repetitive, codifiable tasks.
Augmentation AI enhances human capabilities. It provides tools that make workers more productive — think AI-assisted medical diagnosis, AI-powered financial modeling, or AI-enhanced creative workflows. When augmentation AI enters an industry, it makes existing workers more valuable.
The study used instrumental variable estimators to isolate the causal effect of each type, drawing on AI development in countries with weak US economic ties to address the thorny problem of separating cause from correlation [Fact]. This is not just observing that AI adoption and wage changes happen at the same time — it is an attempt to establish that one actually causes the other.
The Numbers Tell a Clear Story
For low-skilled occupations exposed to automation AI, the effects are stark [Fact]:
The emergence of new work — entirely new job titles and tasks — declines. Existing employment levels fall. Wages drop. In other words, automation AI does not just shift what low-skilled workers do. It reduces how many of them are needed and how much they earn.
For high-skilled occupations exposed to augmentation AI, the picture flips [Fact]:
New types of work emerge. Employment in these roles grows. Wages rise. The workers whose jobs involve complex judgment, creativity, or strategic thinking find that AI tools amplify their output and their market value.
The contrast is not subtle. It is a clear divergence, and it maps onto existing inequalities in a way that should alarm policymakers.
Why This Matters for the Wage Gap
The study period — 2015 to 2022 — captures the era of AI before ChatGPT went mainstream [Fact]. The effects Marguerit documents were already in motion with earlier waves of machine learning and automation tools. The arrival of large language models in late 2022 likely accelerated these dynamics.
The implication is sobering: AI is not neutral [Claim]. It does not affect all workers equally. Instead, it reinforces and potentially accelerates the wage gap between high-skilled and low-skilled workers. Those who already earn more gain AI tools that make them even more productive and valuable. Those who already earn less face AI systems designed to do their jobs more cheaply.
This is not an argument against AI adoption. It is an argument for being honest about who benefits and who bears the cost.
What Workers Should Consider
If you are in a low-skilled or routine-heavy role, the data suggests your bargaining power may be declining as automation AI advances. The most effective response is not to resist the technology but to move toward roles where AI augments rather than replaces. Focus on building skills in judgment, relationship management, and complex problem-solving — areas where augmentation AI creates value rather than substituting for labor.
If you are in a high-skilled or analytical role, the short-term outlook is favorable, but do not become complacent. The boundary between "augmented" and "automated" shifts with every new AI capability. Today's augmented role can become tomorrow's automated one.
For everyone, the key insight from this research is that the type of AI your employer adopts matters enormously. A company investing in augmentation tools is signaling something very different from one investing primarily in automation. Pay attention to which direction your industry is heading.
Explore how AI exposure differs across specific occupations in our detailed occupation analyses, covering over 1,000 roles with automation risk scores and task-level breakdowns.
Sources
- Marguerit, D. (2025). "Augmenting or Automating Labor? The Dual Impact of AI on Jobs and Wages." arXiv:2503.19159. https://arxiv.org/abs/2503.19159
Update History
- 2026-03-31: Initial publication based on arXiv:2503.19159.
This analysis was produced with AI assistance. All statistics are sourced from the referenced research paper. We encourage readers to consult the original study for full methodology and context.