Table of Contents
A Strategy That Only Counts the Good News
Canada’s federal AI strategy has a metrics problem. The government is keeping careful score of how many jobs artificial intelligence is helping to create — but when it comes to the positions being eliminated by the same technology, the official count goes completely quiet. It is a selective form of bookkeeping, and economists and labour advocates are starting to notice.
The strategy, which positions Canada as a global AI leader partly on the strength of its homegrown talent and research infrastructure, points to job creation figures as evidence that the country is navigating the AI transition well. What it does not do is measure, track, or even formally acknowledge the workers on the other side of that ledger — the ones whose roles are being automated out of existence.
That omission matters more than it might seem. A policy framework that only measures wins is, almost by definition, one that cannot respond to losses. It can celebrate an expanding AI sector while missing whether the gains are reaching workers who need them, whether displaced people are finding comparable employment, or whether particular occupations are being hollowed out faster than training systems can adapt.
Job creation is a legitimate measure. New AI companies need researchers, engineers, sales staff, legal advisers, product managers and people who can apply the technology inside existing businesses. But gross job creation is not the same thing as a full labour-market assessment. If automation removes work in one part of the economy while creating highly specialised roles somewhere else, the headline total may sound encouraging without saying much about who has gained, who has lost, or how difficult it is to move between the two.
That distinction is especially important because AI does not arrive in workplaces as a single, easily counted event. A company may eliminate a role outright. It may stop replacing departing staff. It may hand a larger workload to fewer employees with AI tools, or contract out work that was once done in-house. None of those outcomes necessarily produces a clean public record labelled “AI displacement.” Yet each can reshape a worker’s prospects.
The result is a familiar policy trap. Governments can point to investment and hiring announcements, both of which are visible and politically useful. The quieter effects of automation are harder to capture: reduced entry-level opportunities, changed job descriptions, less predictable contract work and workers pushed into lower-paid roles after their existing skills lose value. Difficulty is not an excuse for ignoring the category. It is the reason to take the measurement problem seriously.
Canada’s federal AI strategy tracks jobs created by AI but completely ignores workers displaced by automation. The gap is not merely technical. It shapes the kind of success the strategy is able to recognise.
What the Federal AI Strategy Actually Says
The federal AI strategy leans heavily on economic opportunity as its central narrative. AI investment, the argument goes, will generate new industries, new roles, and new export revenue. That is not wrong — it is just incomplete. Canada has genuine strengths here: the Vector Institute in Toronto, Mila in Montreal, and the Alberta Machine Intelligence Institute form a research triangle that has attracted significant international attention and investment.
Those institutions matter because research capacity can support company formation, attract talent and help Canadian firms build expertise rather than simply buy it from elsewhere. A country that wants a meaningful role in AI needs more than enthusiasm; it needs people and institutions capable of understanding the technology, developing it and putting it to use responsibly.
But research strength does not automatically answer the labour question. The workers most likely to benefit from a growing AI industry are not always the same workers whose tasks are easiest to automate or reorganise. New roles may be concentrated in particular regions, companies or fields. The losses may be spread across offices, customer-service functions, creative work, administrative jobs and other roles where employers can use AI to produce drafts, process information or handle routine communication with fewer people.
That is why “more jobs” is an inadequate endpoint for public policy. The relevant questions are more demanding. What kinds of jobs are being created? Are they permanent or precarious? Who can qualify for them? What happens to workers whose experience is in occupations that are shrinking? And when a business reports productivity gains from AI, how much of that gain comes from creating new capacity versus reducing its need for labour?
But the strategy’s success metrics are built around inputs and outputs that make the picture look tidy. Jobs created. Dollars invested. Companies scaled. There is no equivalent tracking mechanism for jobs eliminated, sectors destabilised, or workers left without a clear retraining path.
That imbalance can distort public debate. Investment is easy to quantify, and a new company or research initiative offers a concrete story of progress. Displacement is messier. A role may disappear gradually, and an employer may not identify AI as the sole cause. Economic change also has many causes at once: broader software adoption, outsourcing, changing consumer demand and corporate cost-cutting can overlap with AI deployment. Still, uncertainty should lead to better monitoring, not an assumption that the costs are too diffuse to count.
A serious accounting system would not need to pretend it can attribute every lost job to a single technology. It could begin by examining affected occupations, changes in hiring patterns, shifts in task requirements and the outcomes of workers moving out of vulnerable roles. It could also distinguish between a worker who finds a new job quickly and one who faces a long period of insecurity or a permanent reduction in earnings. Those are very different experiences, even if both disappear inside an optimistic aggregate figure.
Why Honest Workforce Data Changes the Policy Response
This is not a uniquely Canadian problem. Governments worldwide have struggled to build honest AI workforce accounting. The OECD’s ongoing work on AI and the labour market has repeatedly highlighted the gap between job creation narratives and the more complicated displacement data that national strategies tend to underreport. But Canada’s approach is particularly striking given how much the strategy emphasises economic opportunity and the country’s ability to lead in AI.
Leadership should mean being willing to measure the trade-offs, not just market the upside. If policymakers do not know where displacement is happening, they cannot target retraining effectively. They cannot tell whether training is reaching people before their jobs disappear rather than after. They cannot identify whether support is needed for workers, communities or sectors under unusual pressure.
There is also a basic issue of trust. Workers are unlikely to accept repeated assurances that AI is creating opportunity if their own workplace is reducing staff, cutting junior roles or asking the remaining employees to do more with automated tools. Public confidence cannot be sustained by insisting that aggregate gains settle every question. People judge the transition through their wages, their security and their chance of finding meaningful work.
Without honest workforce data, Canadian policymakers risk designing AI policy that fails the workers it should protect. That is the central weakness in a strategy that celebrates job creation while leaving job loss largely outside the frame.
Canada does not need to choose between supporting AI investment and confronting automation’s costs. Treating those goals as opposites is a false choice. The stronger approach is to judge the transition by its net effects on people as well as its visible gains for companies and research institutions. Counting the jobs AI helps create is necessary. Counting the jobs and opportunities it may erase is necessary too. Anything less leaves policymakers managing only the version of the AI economy they want to see.
