Study warns AI could permanently remove millions from workforce by 2050
A survey of nearly 200 economists, AI researchers and policy experts predicts that artificial intelligence will shrink North America's labour force significantly over the next 25 years - with the most severe outcomes potentially removing 10 million workers from the U.S. job market alone. The research, published this month by teams from the Federal Reserve Bank of Chicago, Yale University and the University of Toronto, is the most rigorous attempt yet to quantify AI's impact on employment.
For Canadian HR leaders, the timing matters. The findings arrive as Canada faces demographic aging, immigration volatility and trade uncertainty - conditions that will amplify whatever disruption AI causes.
What the numbers show
The study focuses on labour force participation rates - the share of working-age adults employed or actively seeking work. The median economist surveyed expects the U.S. rate to fall from 62.6 per cent today to 58.3 per cent by 2050, even without dramatic AI acceleration.
Under a "rapid progress" scenario, where AI systems surpass human performance across most tasks by 2030, participation drops to 55 per cent by 2050. Researchers estimate roughly half of that decline - about 10 million people - would result from AI rather than aging demographics.
Canada tracks similar workforce aging and AI adoption patterns. Statistics Canada found no clear sign yet that AI-exposed occupations are declining faster than others. But Canadian AI adoption remains concentrated in larger firms and knowledge-intensive sectors, suggesting the country is following rather than diverging from U.S. trends.
"AI adoption isn't really rocketing ahead in Canada," the Canadian Labour Congress said earlier this year. "But that doesn't mean we're immune."
Which jobs face pressure - and which don't
Economists ranked dozens of occupational categories by expected employment growth through 2030. The pattern is stark.
Jobs expected to grow cluster in roles requiring physical presence, human judgment and interpersonal trust: health professionals, personal care workers, protective services workers. These are tasks AI cannot yet replicate reliably.
Jobs expected to contract are defined by information processing and rules-based decisions: general clerks, customer service representatives, data entry workers, assemblers. In Canada, this maps directly onto administrative roles in insurance, banking, government and professional services - sectors employing hundreds of thousands of workers.
Statistics Canada research published this year identified clerical and administrative occupations as showing the highest automation risk from generative AI. "We're already seeing fewer positions in clerical and administrative work, because a lot of those tasks are being complemented by AI," said Tricia Williams, research director at the Future Skills Centre.
The entry-level threat
The study documents concentrated employment pressure on early-career workers in AI-exposed occupations. This threatens the traditional apprenticeship model where junior employees learn by handling routine analytical and drafting tasks under senior supervision.
As generative AI takes on these entry-level tasks, demand is shifting toward workers who can supervise and improve AI outputs rather than produce first drafts. That changes what employers are looking for - and threatens the pipeline feeding management ranks over time.
TD Economics found Canadian youth unemployment among recent graduates at four-year highs this year, though it attributed most to student population expansion rather than direct AI displacement. The structural risk is more concerning: junior positions are not disappearing overnight, but the volume is shrinking.
Wealth inequality will likely worsen
Economists surveyed expect wealth inequality to worsen significantly as AI advances. Under the baseline forecast, the wealthiest 10 per cent of households will hold 75 per cent of national wealth by 2050. Under rapid AI progress, that reaches 80 per cent - levels unseen since before the Second World War.
The mechanism is straightforward. AI raises returns to capital - ownership of data, compute infrastructure, intellectual property and tech equity - while compressing wages in automation-vulnerable occupations. Labour's share of economic output is forecast to fall from 55.5 per cent today to 45 per cent by 2050 if AI advances rapidly.
For compensation committees, this creates tension. Companies relying on administrative and clerical roles face structural cost pressure. But layoffs rather than retraining risk fuelling broader inequality and political backlash.
Public and expert views diverge sharply
The study documents a dangerous gap between what economists recommend and what the public wants.
Economists broadly support targeted interventions: 71.8 per cent backed retraining programmes providing displaced workers up to $25,000 annually in training credits, funded by modest payroll taxes. Only 13.7 per cent of economists supported a government job guarantee paying at least $15 an hour to any adult seeking work.
The public disagreed. 57.1 per cent supported the job guarantee - a 43-percentage-point gap. If AI-driven displacement becomes visibly widespread, public demand for intervention will almost certainly exceed economists' preferred policy menu.
For Canadian employers, this signals that the regulatory and legislative environment governing workforce transitions may shift dramatically. Only 6 to 8 per cent of Canadian workers are currently covered by federal labour protections related to technological change.
The Liberal government introduced legislation this year focused primarily on stimulating Canada's AI industry. The Canadian Labour Congress and others have argued the policy response has paid insufficient attention to managing disruptions for workers.
What HR leaders should do now
The study's most important caveat is also its most important insight: the median outcome - modest disruption, gradual adjustment, manageable transition - remains most probable. Economists' overall forecasts stay relatively close to historical baselines.
But the distribution of possible outcomes is wide. The consequences of preparing only for the median are asymmetric. If disruption is slower than feared, organizations investing in reskilling and workforce adaptability will have lost little. If disruption is faster, organizations that have not will scramble in a labour market offering fewer traditional entry-level roles.
The Institute for Research on Public Policy has called for regional workforce strategies tailored to provincial economies and sector-specific reskilling investments. Statistics Canada's research shows AI risk is unevenly distributed across industries and geographies.
The researchers put it directly: employers cannot plan only for the median outcome. They must contend with tail risks, including potential deep and rapid contractions in labour force participation. In Canada, where institutions to manage that contraction remain modest and policy debate is nascent, the time to begin is now.
HR leaders can start by mapping which roles in their organization map to high-risk occupational categories, assessing entry-level hiring patterns, and developing reskilling pathways for administrative and clerical staff. AI for CHROs offers frameworks for managing these workforce transitions strategically.
Your membership also unlocks:
