AI ‘fatigue’ exposes weaknesses of training programs

Recent labour market analysis shows that generative AI (genAI) is technically capable of automating key tasks in about half of Canadian jobs over the next five years  – but that doesn’t mean workers are ready to embrace it.  

According to Matthias Oschinski, a senior fellow at Georgetown’s Center for Security and Emerging Technology (CSET), Canada has a management training problem – and AI-based disruption is going to bring that issue to the forefront. 

“In international comparisons, Canada has always done a really bad job in training their own workforce,” he says. 

“And this is I think a big problem now with AI, because jobs will change at a much faster pace than the last 20 years or so.”  

That gap will necessarily raise practical questions about who will fund and organize the upskilling needed for AI-augmented work in organizations, Oschinski warns – especially considering Canadian enterprises are largely small- to medium- sized, generally with fewer resources for training and management. 

AI training and the limits of ‘capability only’ management 

Oschinski’s study, “Harnessing Generative AI: Navigating the Transformative Impact on Canada’s Labour Market”, conducted with University of Toronto researcher Ruhani Walia, found that effects of genAI on employment vary widely according to sector.  

It also stressed that realizing productivity gains will depend heavily on workforce preparation and upskilling, particularly in complementary skills like social, managerial and leadership capabilities that AI is unlikely to replace. 

Rather than capability-based training and management models, Oschinski says that the skills and knowledge required to handle AI-augmented work, and the overall faster pace of transformation, require a longer-term, more deliberate and proactive approach to training – which Canada lacks. 

“It actually requires more lifelong learning … and there needs to be more of an effort made by employers to start this up,” he says, acknowledging the challenges that small or medium-sized businesses face in achieving this. 

“But you have to come up with solutions, where maybe they band together, and also work with universities and colleges, to design better training and more frequent training.” 

Highlighting this emphasis on more frequent and focused training is EY Canada, where leaders say they are now confronting “AI fatigue” among staff who feel swamped, regardless of high levels of adoption.  

“People are overwhelmed by the pace of AI, not the idea of it,” says Biren Agnihotri, the firm’s chief technology officer. 

A recent EY survey found that 43 percent of Canadian employees worry about becoming over reliant on AI, Agnihotri says, adding that “only about five percent of Canadian employees are using AI to truly transform their work today, even though more than half says workloads have increased.” 

Segmenting workers by AI capability and confidence 

Data from the generative AI study shows that automation risk varies sharply by region and sector, with transportation and warehousing recording an estimated 56.4 percent share of at-risk occupations, compared with just 3.1 percent in educational services.  

That diversity of exposure mirrors what large employers are seeing inside their own workforces: not all jobs, and not all employees, interact with AI in the same way. 

To address this emerging fact, Agnihotri says EY Canada has moved away from a single, firm-wide model of AI training.  

“Expecting everyone to move at the same pace or rely on the same tool set did not reflect how diverse roles actually operate,” he says.  

“Employees across any organization … they engage with AI differently, not because of the labels, I would say, but because people have different experience, comfort levels, motivations.” 

Skill and interest-based cohorts 

The key takeaway for Agnihotri is that “AI user” is not a single category; beyond job titles, employees differ in their subject-matter expertise, comfort with technology and readiness to change how they work. 

EY began segmenting staff along two dimensions: capability and confidence. Agnihotri says, “This helps us to tailor learning and change management in a way that is supportive and effective, rather than assuming that everyone starts from the same place,” he says. 

In practice, that has meant different interventions and training methods for different cohorts: advanced experimentation environments and leadership opportunities for high-skill, high-confidence users; role-based learning and guided use-cases for those with interest but less technical depth; and purpose- and governance-focused approaches for more cautious groups. 

“For example, with high-skill and high-confidence [employees], our focus is on acceleration and leadership,” he says.  

“These individuals often act as an early adopters and champions. We provide advanced experimentation environments to them, deeper technical and ethical guidelines around responsible AI, and opportunities to shape use cases, mentor others and influence how AI is embedded into client and internal workflows.” 

Moving from passive to proactive on upskilling 

Oschinski’s generative AI study concluded that realizing productivity gains will require a coordinated effort on infrastructure and workforce preparation, with particular emphasis on upskilling and retraining.  

For him, the responsibility for training and upskilling employees to be AI-ready and relevant is clearly in the court of employers – and those who don’t will sink. He observes how some organizations “just rain down AI tools” on their employees, then wonder at their overwhelmed, underproducing workforce. 

“If you don't see that as part of your job, to then also train your employees up, then you have a problem,” he says. “There's some something fundamentally wrong in the understanding of how to be a manager, and how to manage people.” 

Looking ahead, Oschinski says Canada’s traditionally passive approach to manager training is “Not going to work anymore … because the skills will become obsolete at a much faster pace, and that means that the onus now is more on the employer to really become more proactive on the training.” 

Task-level training and employee consultation 

At the task level, Oschinski warns against assuming that algorithms necessarily understand workflows better than front-line staff. Workers need support in understanding how their expertise interacts with AI, he says, including how to design prompts and workflows that reflect actual job realities. 

It should also include how to push back when algorithmic recommendations do not make sense on the ground, requiring meaningful employee consultation – another area where Canada falls short, he says, and points to international evidence on the benefits of early, structured consultation.  

Referring to OECD (Organisation for Economic Co-operation and Development) and European research, Oschinski says this work “kind of shows that when you consult workers early on in the process of technology adoption, you get more acceptance of the technology, and it also leads to more training for employees.”  

In his view, part of what is now labelled as AI fatigue in Canada reflects not just the volume of tools but the way they are introduced into workflows, without sufficient input from the people using them.