62% of Canadian workers fear AI is eroding skills: what the IBM study really says

Nearly two-thirds of Canadian employees surveyed by IBM say they worry artificial intelligence is eroding their skills. The figure — 62% — comes from a new IBM Institute for Business Value study released in Canada on 2 October, and it points to a workplace problem that is more complicated than the usual argument over whether AI will “take jobs”.

The more immediate risk may be that workers use AI every day without being given clear responsibility, training or time to check what the system produces. That can create a strange combination: more automation, but also more invisible human work.

What did the Canadian survey find?

IBM says 62% of Canadian employees are concerned about skills erosion. Among employees who already worry about that issue, 69% say AI has begun to erode at least some of their skills.

Critical thinking is at the centre of the gap. Fifty-four percent of Canadian employees and 55% of Canadian chief human resources officers say critical thinking and problem framing are among the most important skills in the AI era.

But another result shows a disconnect: 68% of Canadian CHROs say the ability to supervise, validate and override AI outputs is the workforce’s most essential capability, while only 29% of Canadian employees rank judgement as important.

Why can AI reduce a skill even when productivity rises?

A skill weakens when a person stops practising the difficult part of a task. If software always drafts the first version, summarizes every document, writes every formula or suggests every answer, the worker may spend less time building the underlying mental model.

That does not mean using AI makes people less capable by definition. A calculator did not eliminate mathematics. The risk appears when the tool becomes a substitute for understanding rather than an accelerator for someone who can still verify the result.

What is “invisible work” around AI?

IBM reports that 87% of Canadian CHROs believe AI adoption creates invisible work. Examples include validating recommendations, correcting mistakes, adding missing context and handling exceptions that automated systems cannot resolve.

That work can be easy to underestimate because it does not always appear as a separate task in a job description. An employee may save 20 minutes generating a draft and then spend 25 minutes checking whether the draft contains fabricated facts, outdated assumptions or a tone that creates legal or reputational risk.

Does AI actually increase workloads?

For many respondents, yes. Fifty-one percent of Canadian employees say AI increases their workload or that the extra work it creates goes unrecognized.

This is an important counterpoint to simple productivity claims. A company can measure how quickly a document is generated while failing to measure the time spent reviewing, correcting, documenting and taking responsibility for the output.

Who gets blamed when AI is wrong?

IBM says 39% of Canadian employees report that when something goes wrong with AI, the blame falls on them. At the same time, 32% of Canadian CHROs say unclear accountability complicates AI deployment.

That is a governance problem. If a company tells workers to use AI but does not define when they can rely on it, when they must verify it and who owns the final decision, employees carry responsibility without having a clear operating framework.

What skills become more valuable as AI improves?

The study suggests that human judgement becomes more important, not less. Workers need to frame the problem correctly, recognize when the model is missing context, challenge confident but incorrect output and know when a decision should remain human-led.

Domain knowledge also matters. A lawyer, engineer, nurse, analyst or accountant cannot reliably evaluate an AI answer without understanding the professional standards behind the task. The better the system becomes at producing fluent answers, the more important it is to distinguish fluency from correctness.

Should workers stop using generative AI?

No. The useful lesson is not avoidance but deliberate use. AI can remove repetitive drafting, accelerate research triage, generate alternatives and help people explore a problem faster. The mistake is handing over the part of the work that builds or demonstrates professional judgement.

A practical approach is to decide which steps are human-led, which are AI-assisted and which can be safely automated. IBM says organisations that make those distinctions clearly report stronger outcomes globally.

What can employees do to protect their skills?

Write your own first-pass reasoning on high-stakes tasks before asking AI for a second view.
Verify important claims against primary sources rather than trusting generated citations.
Keep practising core skills without AI on a regular basis.
Ask the model to critique your reasoning instead of always generating the answer from scratch.
Record recurring AI mistakes so you learn where the tool is unreliable in your field.
Know which decisions your employer requires a human to approve.

What should employers change?

Training should focus less on “prompt tricks” and more on judgement. Workers need examples of when to reject an AI answer, how to document a review and how to escalate uncertain cases.

Employers should also recognize verification as work. If AI produces more output, someone still needs time to evaluate it. Productivity targets that count generation speed but ignore review time can encourage unsafe shortcuts.

Why is HR itself behind?

IBM says 63% of Canadian organisations make limited or no use of AI inside the HR function, even though HR is increasingly expected to lead workforce transformation. Canadian CHROs also rate their own functions relatively low in AI literacy, performance measurement and change-management capability.

That creates a structural challenge: the department asked to redesign jobs may itself be learning how to use the technology. Companies that treat AI deployment as only an IT project can therefore miss the work-design and accountability questions.

How large was the study?

The IBM Institute for Business Value, working with Oxford Economics, ran two surveys between April and June 2026. One covered 1,500 CHROs and equivalent executives globally, including 75 in Canada. The second included 8,800 full-time employees across 28 countries, including 452 Canadian employees.

Those numbers are substantial enough to provide useful signals, but they are still survey results rather than a census of every Canadian workplace. The findings should be read as evidence of broad concern, not proof that every sector experiences AI in the same way.

What is the practical takeaway?

The most valuable employee in an AI-heavy workplace may not be the person who produces the most generated content. It may be the person who knows when the AI is wrong, can explain why, and can improve the decision that follows.

That shifts the competitive advantage toward critical thinking, judgement, problem definition and subject expertise — exactly the skills that can weaken if people stop practising them.