Is AI going to lead to mass unemployment and increasing inequality?

From: Rosalie Wyonch

To: Artificial Intelligence watchers

Date: June 2, 2026

Re: Is AI going to lead to mass unemployment and increasing inequality?

The fear that artificial intelligence is about to upend Canadian labour markets is viewing the picture through a distorted lens. Each major technological wave has been accompanied by predictions of mass unemployment that did not materialize. Markets adjust. New jobs and products appear. Productivity gains lead to real wage growth, feeding demand and the creation of new work.

First, adoption has been growing at a steady pace with no evidence of rapid acceleration. About 19 percent of businesses in Canada currently use AI (compared to 12 percent this time last year, and 6 percent the year before that).  Four in 10 businesses say AI is not relevant to their products or services and 10 percent are uncertain of potential benefits. Adoption is concentrated in finance and insurance, professional, scientific and technical services, and information and culture industries. Compared to 2025, there was no increase in adoption in the manufacturing industry and a lower proportion of businesses are using AI in wholesale trade.

Where firms have adopted, the labour-market story is, so far, undramatic. Among businesses that integrated AI in 2025, 89.4 percent reported no related change in employment. About 70 percent of firms planning to adopt expect the same.

When general-purpose technologies arrive, the early years are spent experimenting and investing in the complementary assets – skills, data, organizational redesign – that eventually unlock real gains. Productivity does not arrive overnight. It rarely has.

For many jobs, AI is a complement to human labour, not a complete replacement for it. A decade ago, research suggested that 35 to 42 percent of the labour force was at high-risk of automation and about 15 percent of work activities could be displaced by automating technologies (of the time) by 2030.

A C.D. Howe Institute study contextualized these results with actual Canadian labour market dynamics: The potential for automation rarely translates one-for-one into actual automation, and labour markets evolve faster than the forecasted doom.

That does not mean transitions are painless. Research focused on the US labour market estimates that early-career workers in AI-exposed occupations are already seeing a 16-percent relative employment decline, even as employment among more experienced workers holds steady (Brynjolfsson, Bharat and Chen. 2025). Research on the UK labour market finds that the wage premium on AI-exposed tasks fell by 12 percent between 2017 and 2024.

These shifts are real, but they are appearing as foregone employment and wage growth, not a mass employment displacement. They are also similar to the historical pattern of technological change: gradual, sectoral, and concentrated.

Labour market disruption could affect certain groups more than others and the underlying equity dynamics are also important to consider. Black and Indigenous Canadians, younger workers, and those with less education are overrepresented in automation-susceptible occupations. That is true – but the cause isn’t biased technology. Long-standing employment and economic gaps concentrated these workers in those jobs before AI was a serious commercial technology. AI inherits the inequality; it does not create it. The cure lies where it always has; in education, training, and broader labour-market policy.

What about AI tools being used to make hiring decisions? The same logic applies. AI is a neutral instrument. It can amplify, neutralize, or reduce existing biases, and which one happens depends on how we deploy it.

A working paper (Zhou 2025) tested major large language models on 12,000 historical US job advertisements. Every model produced selections at least as fair as the human benchmarks, with most actively favouring minority candidates. The right posture is due diligence and ongoing monitoring, not pre-emptive prohibition that presumes algorithms guilty until proven innocent.

So, how should Canada address the need to adopt AI to fuel economic growth and increase living standards, while moderating the downside risks for affected workers?

First, treat equity as a labour market and socioeconomic challenge, not an AI problem. Since AI is currently most disruptive to early-career workers, and has potential equity implications given the existing structure of the labour market, bridging the widening gaps will require increasing opportunities for building AI literacy, and industry- or job- specific training beyond post-secondary co-op programs.

Second, finish building the evidence base. Statistics Canada’s TechStat initiative needs to be deployed and resourced to inform continued monitoring of AI’s impact on the labour market and economy.

Third, accelerate AI adoption by addressing the barriers that firms cite: cybersecurity and privacy concerns, cost, regulatory concerns and a lack of a skilled workforce. Updating Canada’s data privacy rules and clarifying the regulatory approach to AI would significantly reduce these uncertainties.

AI will reach Canadian workers. It will be slower than some might think. It will not be evenly distributed. And on the strength of every prior technological wave, it will leave living standards higher than it found them – but only if Canadian firms actually use the technology. The risk isn’t that AI moves too fast for us to adapt. It’s that we adapt too slowly to benefit.

Rosalie Wyonch is Associate Director of Research at the C.D. Howe Institute. 

To send a comment or leave feedback, email us at blog@cdhowe.org. 

The views expressed here are those of the author. The C.D. Howe Institute does not take corporate positions on policy matters.