By Ryan Khurana
May 22, 2025
Introduction
Canada faces a paradox in the realm of artificial intelligence (AI): despite being a global powerhouse of AI research and talent, our businesses are lagging in embracing and integrating these technologies. As a result, Canada is missing a tremendous opportunity to boost its sluggish productivity growth; the country has grown only 0.9 per cent annually since 2021, placing us second-to-last in the G7 (Gu and Willox 2023).
Canada must close the gap between AI innovation and adoption by pursuing policies that encourage productivity-boosting AI – applications that augment workers and make them more efficient, rather than simply replace them. The answer is a multi-level policy framework that accelerates the uptake of AI in ways that enhance output, job quality, and workforce participation.
Key recommendations include:
- Rebalancing incentives (e.g., tax policies) to favour human-centric automation.
- Investing in AI integration and process redesign instead of piecemeal gadgetry.
- Updating education curricula to focus on uniquely human skills like judgment and creativity.
- Fostering collaborative innovation by involving workers in AI deployment decisions.
Monitoring AI’s impacts on employment, productivity, and worker well-being will be crucial. We must ensure that the AI revolution creates a more prosperous economy with broadly shared benefits – not mass joblessness or “so-so” efficiency gains.
Swift and coordinated action on AI adoption will help secure Canada’s long-term competitiveness and high standard of living, ensuring the research leadership established domestically can translate into success stories that help own the technology going forward.
Prime Minister Mark Carney recently appointed MP Evan Solomon as Canada’s first Minister of Artificial Intelligence and Digital Innovation. This signals a consolidation of federal focus on a field that has historically been spread across numerous portfolios, offering Canada a clear opportunity to chart a unified vision for AI policy and strategy (Geist 2025). Solomon’s challenge will be to distinguish between productivity enhancing AI and “so-so” automation – harnessing the benefits of AI, while ensuring adequate regulation to mitigate associated risks.
Canada: leading the way in AI research
Canada has earned a reputation as a global AI leader in terms of research breakthroughs and talent development. Canadian labs and universities have pioneered advances in machine learning, and the federal government was the first in the world to launch a national AI strategy in 2017 (CIFAR 2023). Yet when it comes to deploying AI on the ground in businesses and public services, Canada is falling behind.
Michelle Alexopoulos, an economics professor at the University of Toronto, noted that Canada was an early pioneer in AI research but struggles with “the adoption and commercialization of things that are actually developed here at home” – a shortfall that is “feeding into the [productivity] problem” we face (Lowey 2024). In other words, our world-class innovations are not translating into widespread productivity gains in the economy.
The data confirm this adoption gap. By the end of 2021, only about 3.7 per cent of Canadian firms (with 5 or more employees) had incorporated AI in their operations in any capacity (Lockhart 2023). This uptake rate is barely half that of global leaders – for example, Denmark’s business AI adoption stands at roughly double Canada’s rate. A Statistics Canada study similarly found that as of 2024 only ~6 per cent of Canadian businesses were using AI to produce goods or services (Bryan et al. 2024). International rankings consistently show Canada lagging many peer countries in commercial AI leadership (Fattorini et al. 2024).
Numerous studies have noted the increase in productivity from certain forms of AI productivity adoption. Brynjolfsson et al. (2025) highlight that customer service AI assistants boost productivity by 15 per cent on average, especially for lower-skilled workers, and increase customer satisfaction.
The result of this slow adoption is that Canada has yet to reap significant productivity benefits from the ongoing AI revolution. Indeed, our labour productivity growth has remained anemic even as AI technology has advanced, and Canada risks falling further behind if it cannot capitalize on its homegrown innovations. Peng et al. (2023) found that coding assistants increased output for software engineers on average by 56 per cent. The OECD noted firm-level productivity increasing between 3 to 11 per cent across countries that adopted AI. The growth concentrated in a U-shape, where younger, entrepreneurial firms, and older, larger firms undergoing comprehensive digital transformation gained the most (Filippucci et al. 2024).
The disconnect between innovation and adoption comes at a pivotal moment. Canada’s productivity growth has been stagnating, and business investment in new technologies remains weak. Meanwhile, demographic pressures (an aging workforce and labour shortages in some sectors) mean we urgently need efficiency gains to sustain economic growth.
AI offers a critical opportunity to boost output per worker, stimulate innovation, and even alleviate labour shortages by enabling workers to focus on higher-value tasks. McKinsey Global Institute’s report (2023) on the economic potential of generative AI estimates that automation could boost global productivity by 0.5 per cent to 3.4 per cent annually, with generative AI contributing 0.1 to 0.6 per cent of that growth.
In short, embracing AI more fully – and strategically – could be a key part of the solution to Canada’s persistent productivity challenge. However, how we adopt AI matters tremendously: Will we use it to simply cut costs by replacing workers, or to genuinely improve the efficiency and quality of work? Canada’s AI adoption must be steered toward productivity-enhancing and participatory applications. The goal must be to boost economic output and workforce participation, while also ensuring that Canada’s AI future is one of shared prosperity rather than polarized outcomes.
The case for productivity-enhancing AI
Not all AI is created equal in its effects on productivity and jobs. It is crucial to distinguish between AI applications that complement human labour versus those that merely substitute for it. AI and automation technologies that augment workers – for example, decision‑support systems that help professionals make better decisions, or AI assistants that streamline complex workflows – tend to raise labour productivity and can even improve job quality. By taking over routine, time‐consuming tasks or by providing smarter and more contextualized tools, these technologies allow workers to focus on higher‑value activities, leading not only to higher output but also to enhanced employee satisfaction and skill development (Acemoglu and Johnson 2023).
In contrast, a narrow, cost‑cutting use of AI can fall into what some economists refer to as “so‑so automation” (Acemoglu 2019). This term describes technologies that displace workers without delivering significant improvements in task efficiency. A classic example is the self‑checkout kiosk: while it enables stores to reduce cashier headcounts, the technology often fails to meaningfully speed up the checkout process or enhance the customer experience. In such cases, the technology displaces workers but the limited productivity gains fail to create new value – an outcome that can depress the overall labour share of income and contribute to rising inequality. Other cautionary examples include fully automated assembly lines that, despite their technological ambition, have led to operational bottlenecks and additional costs, as famously evidenced in some early missteps at companies like Tesla (Khurana 2019).
So‑so automation is problematic because it primarily delivers short‑term profit gains for firms – through marginal cost reductions – while leaving the broader economic picture largely unimproved. When innovations simply replace human labour without enabling new tasks or creating a more dynamic production process, the net result is a lose‑lose scenario: displaced workers, static productivity figures, and an erosion of the labour component of national income.
The hidden costs of poorly integrated AI
AI deployments that fail to properly integrate with human workers create a cascade of productivity-draining issues that often remain invisible on balance sheets until it’s too late. While organizations rush to adopt AI technologies, many fall into implementation traps that actively reduce productivity rather than enhance it.
When AI systems are deployed without meaningful collaboration with employees, they lack the essential feedback loops required for improvement. AI models need domain-specific refinement to capture the unique relationships and values important to an organization. Without this feedback from frontline workers, AI systems perpetuate a “garbage in, garbage out” cycle where they never adequately adapt to organizational needs.
This creates what might be called “technical debt by proxy” – each interaction with a suboptimal AI system requires additional human work to fix, creating hidden labour costs throughout the organization (Waddell 2019). Many supposedly automated systems actually rely on concealed human work to appear functional, with technologists hiding the human labour that makes their AI products seem more advanced than they truly are (Khurana 2019).
Poorly integrated AI can also erode organizational knowledge. Tesla’s experience with “over-automation” actively undermined production goals, eventually leading to the admission that “humans are underrated” after automation efforts compromised manufacturing targets (Khurana 2019).
The knowledge that frontline workers possess – their tacit understanding of processes, exceptions, and adaptations – is often invisible to both AI systems and the executives implementing them. When AI replaces rather than augments these workers, this invisible knowledge disappears. Organizations then find themselves with sophisticated systems that handle the 80 per cent of straightforward cases acceptably but fail catastrophically on the 20 per cent of edge cases that human workers previously navigated with ease.
It is not sufficient to invest in AI as a point‑solution – a stand-alone tool designed solely to automate a specific task. Such point‑solutions may appear attractive because of their low‑hanging fruit nature: they are often easy to deploy and require minimal changes to existing business processes. However, when AI is implemented without a broader strategy for organizational change, the rest of the business process may become a bottleneck that restricts efficiency gains (Agrawal, Gans, and Goldfarb 2022).
For example, many companies have deployed AI‑powered chatbots to handle routine customer inquiries and streamline basic transactions, appearing as a quick win. Yet, in areas where these companies continue to operate on legacy systems, the chatbot may provide inaccurate or delayed information, leading just to more customer frustration. It also frustrates employees, who must deal with the chatbot’s errors. It demotivates everyone and may lead to a chilling effect on future AI investments. In these cases, the AI system is simply tacked onto a legacy workflow that was never redesigned to harness its full potential.
Perhaps most critically, AI deployments that sideline employees create a trust deficit that undermines effective implementation. When workers see AI as a replacement threat rather than a complementary tool, they have little incentive to provide the feedback and knowledge transfer necessary for system improvement.
Successful AI integration requires making a business’s values explicit and ensuring that employees understand the broader organizational goals and values. Without this alignment between human and machine priorities, AI systems become adversaries rather than assistants, creating organizational resistance that manifests as reduced productivity.
Workers who fear displacement also tend to withhold the very insights that would make AI systems more effective. This creates a vicious cycle: suboptimal AI performance reinforces skepticism among workers, who then provide even less input, further degrading system quality. Organizations then face a stark choice between abandoning their AI investments or doubling down on flawed implementations, neither of which delivers the promised productivity gains.
A holistic approach to AI integration
Instead, transformational gains occur when AI is integrated as part of a holistic, system‑level innovation. In this approach, AI is not simply replacing manual processes but is embedded within a re‑engineered decision‑making framework that decouples computational work from human judgment (Agrawal, Gans, and Goldfarb, 2018). This “re‑architecting” of systems enables organizations to unlock new sources of value – for example, by automating data‑intensive predictive tasks while simultaneously empowering human experts to apply nuanced judgment on what is most valuable to the business in a given context. In practice, these system‑level changes may involve redesigning workflows, adjusting job roles, and investing in worker retraining to ensure that human skills are amplified rather than diminished.
A classic example of system‑level innovation comes from the early adoption of ATMs. Instead of simply replacing tellers, banks incorporated ATMs into a broader redesign of their facilities, leading to more numerous, smaller locations. This integration reduced operating costs and enabled banks to expand branch networks, which in turn created more job opportunities. Over time, the role of bank tellers transformed from handling routine cash transactions to focusing on personalized financial advice and customer relationship management, leading to higher job satisfaction and enhanced customer service. In the US the rate of bank tellers increased by 2 per cent annually in the 15 years following the widespread adoption of ATMs (Bessen 2015). As banks introduced more ATMs, the penetration of banking into daily lives increased due to ease of access, necessitating higher employment and output.
This holistic approach starkly contrasts with isolated point‑solutions like self‑checkout kiosks. While self‑checkouts can lead to frustration and bottlenecks, the strategic deployment of ATMs illustrates how re‑architecting workflows can both boost productivity and lead to the emergence of entirely new tasks and higher‑value work that offset the roles displaced by automation.
To create lasting value, organizations must involve workers in AI design from the outset. This collaborative approach accomplishes several critical objectives. First, it ensures that AI systems address real operational needs rather than perceived inefficiencies. Second, it captures the tacit knowledge that might otherwise be lost. Third, it builds the trust and buy-in necessary for effective implementation.
Companies that adopt this more thoughtful approach to AI integration often discover opportunities for innovation that extend far beyond simple task automation. By freeing workers from routine activities, these organizations enable them to develop new skills and create entirely new value propositions. The result is not just preserved employment but enhanced productivity and worker satisfaction.
When AI is successfully implemented, it becomes part of a virtuous cycle: workers provide the domain expertise and feedback that improve AI performance; better-performing AI delivers more value to workers; and this increased value reinforces worker engagement with the technology. The result is a complementary relationship between human and artificial intelligence that drives continuous improvement.
However, a wave of “so-so” AI adoption will likely create the worst of both worlds – displaced workers, lower labour force participation, and mediocre productivity growth. The stakes are high. The following sections lay out a policy agenda to tilt the balance toward AI uses that boost both productivity and workforce participation, aligning with Canada’s long-term economic and social goals.
Key policy tools
Encouraging productivity-boosting AI adoption will require several key policy tools.
Reducing taxes on labour relative to capital
A growing body of evidence suggests that tax systems biased in favour of capital investment can promote “excessive automation” (Acemoglu and Restrepo 2018).
In Canada, the tax code exhibits this bias in favour of capital investment through mechanisms such as accelerated Capital Cost Allowance (CCA) rates for many classes of assets. Under the Canadian Income Tax Act, businesses can deduct a significant proportion of the cost of capital equipment – often including sophisticated automation and AI systems – faster than wage-related expenses can be written off. This creates an indirect incentive for firms to invest in technologies that replace or reduce human labour, even when these technologies deliver only marginal productivity benefits. In contrast, payroll taxes – including contributions to the Canada Pension Plan and Employment Insurance – remain relatively high, which increases the relative cost of employing workers compared to investing in capital.
Canada should evaluate its own tax code for such distortions and move toward a more neutral structure. This is not to suggest that vehicles like the CCA should be diminished, but rather that the same level of investment should be complemented with labour incentives. In modern information economics, skilled labour plays an equally important, if not more important, role than capital in driving innovation and growth. To level the playing field, Canada could consider a suite of policy adjustments aimed at promoting labour-friendly innovation. For instance, programs like the Scientific Research and Experimental Development (SR&ED) tax credit already offer incentives for R&D and are a boon to domestic innovation and investment by providing refundable tax credits for labour hours spent on solving a “technological uncertainty” or in doing basic or applied research. The current system contains provisions to claim hours that went into the quality assurance and testing of these investments. Expanding the ability to claim time spent retraining employees or adapting business workflows to more innovative processes would undoubtedly lead to greater pro-labour innovation.
Additionally, reforms such as reducing payroll taxes or expanded training credits would help discourage “so‑so” automation by ensuring that AI investments are made based on genuine efficiency gains and the complementary enhancement of human skills, rather than simply for tax arbitrage.
By taxing labour and capital at more similar rates (or even slightly favouring human employment), we remove an artificial incentive to automate for cost reasons alone. Acemoglu, Manera, and Restrepo (2020) discuss the employment effects of tax policies favouring capital over labour, indicating that neutral tax systems can significantly boost employment by over 4 per cent.
Overall, smarter tax policy can discourage “so-so” automation and reward AI adoption that grows the economic pie.
Investing in AI initiatives focused on system redesign
Policy-makers should prioritize funding and support for AI projects that aim for holistic process improvement rather than isolated automation of tasks. One reason AI adoption has not yet yielded big productivity jumps is that many firms try to drop in an AI tool without adjusting surrounding workflows or organizational structures.
However, according to MIT scholar Kristina McElheran, “innovate one part of a system, and the rest of the system needs to innovate at the same pace, otherwise things start to come unglued” (Eastwood 2024). In other words, maximum benefits come when AI is integrated as part of a broader transformation of how work is done.
Governments can encourage this by designing programs that fund end-to-end AI integration projects. The move to reinvigorate the Industrial Research Assistance Program (IRAP), which covered up to 80 per cent of labour costs associated with transformation, to the new Canadian Innovation Corporation (CIC), provides a major opportunity for this approach to be adopted. IRAP’s successes were curtailed mainly by a lack of proactive communication to businesses, leading to a concentration of utilization among tech companies. The more we invest in downstream adoption of AI, the greater the economic impact will be.
The federal Scale AI program already embodies this approach in supply chains, co-funding projects that overhaul logistics using AI from forecasting to delivery. Expanding such models to other sectors (health care workflows, manufacturing production lines, even government service delivery processes) will likely yield far greater productivity gains.
Point solutions have their place, but policy should nudge organizations to be ambitious – to redesign for AI. This might include multi-stakeholder “AI transformation accelerators” where firms, tech providers, and consultants work together (with public support) to implement AI across a value chain.
The result is AI adoption that is deeper and more impactful, avoiding the pitfall of scattered tech pilots that never scale or connect. By focusing on systems, we also ensure that human roles are redefined alongside AI, rather than eliminated.
Improving curricula to emphasize AI-complementary skills
Education policy needs to respond to the shifting demand for skills in an AI-powered economy. As AI handles more routine analysis and information processing, the need for human judgment, creativity, emotional intelligence, and complex problem-solving become even more critical. These are precisely the skills that are hardest to automate and will remain in high demand.
Therefore, Canadian curricula at all levels should place greater weight on developing students’ critical thinking, adaptability, and domain-specific judgment. This could mean updating school curricula to include more project-based learning, ethics, and interdisciplinary problem solving, rather than rote memorization.
At the post-secondary level, programs in business, engineering, law, etc., should integrate real-world AI case studies and focus on how professionals can interpret and add value to AI-generated insights (for instance, training medical students not just on AI diagnostic tools, but on how to combine AI outputs with clinical judgment).
The premise is that “computers still cannot think” or contextualize in the nuanced ways humans can – so our workforce must excel at the kinds of thinking and decision-making that complement computational prowess. The federal government can catalyze this shift by working with provincial education authorities and accreditation bodies to modernize learning outcomes through strategic funds that direct educational spending by the provinces and through advocacy with the Council of Education Ministers, Canada (CMEC), which historically resulted in the adoption of provincial open educational resources and spurred digitization.
Additionally, mid-career training initiatives (possibly funded through employment insurance or tax credits) should help workers cultivate the uniquely human skills that make them resilient in the face of AI. By baking human-centric skills into the curriculum, Canada can ensure its talent pool is prepared to leverage AI as a tool, rather than be displaced by it. Over time, a workforce rich in judgment and creativity will attract more employers to invest in AI here, knowing that Canadian workers can unlock its full potential.
Creating mechanisms for collaborative innovation that integrate labour voices
A critical but sometimes overlooked ingredient for successful AI adoption is worker engagement. The people on the front lines often know best where AI could improve efficiency and where it might create problems.
Policy should encourage companies to bring workers (and their unions, where applicable) into the AI design and implementation process. This could be achieved through formal structures – for example, work councils or joint management-labour committees focused on technology – or through requirements for consultation when government funding is involved.
The principle is that giving workers a voice can help guide AI toward augmenting their work rather than replacing it. It also boosts acceptance: employees are far more likely to embrace an AI tool if they have input into its development and trust its purpose.
Some countries are exploring this collaborative approach. A recent Brookings report emphasizes “fostering worker engagement in AI design and implementation” and “enhancing worker voice through unions or other means,” so that AI is deployed in ways that workers ultimately benefit from (Kinder et al, 2024).
Concretely, Canadian policy-makers could encourage this by attaching conditions to AI adoption grants – for instance, requiring an applicant firm to produce an “AI implementation plan” that includes how workers will be consulted, trained, and re-skilled. Another idea is government-sponsored “innovation labs” where companies and worker representatives jointly test how a new AI system would impact workflow, sharing the lessons across industry.
By creating formal ways for people to discuss and collaborate on technology issues, Canada can pursue a high-productivity, high-trust implementation of AI. This not only mitigates fears and resistance but also improves the quality of the AI solutions (since they are informed by on-the-ground insights).
In sum, collaborative innovation ensures AI adoption supports workers’ development and well-being, alongside efficiency.
Ensuring AI supports workforce participation and productivity
Implementing the right policies is only half the battle – we also need to track outcomes and adjust course to make sure AI delivers on the dual goals of higher productivity and robust workforce participation. Governments should establish metrics and evaluation frameworks to answer questions like: Is AI adoption raising output per worker? Is it creating new jobs or at least leading to job transitions rather than permanent losses? Are workers reporting improved job satisfaction and quality of work with AI?
These are complex metrics, but important ones. We recommend a few strategies to ensure AI’s impacts align with Canada’s long-term goals:
Measure adoption and impact rigorously
Building on existing surveys, Statistics Canada and other agencies should regularly collect data on AI adoption rates by industry, firm size, and region, and correlate this with indicators of productivity and employment.
For instance, StatsCan could report on productivity growth in firms that use AI vs. those that do not or track how the adoption of specific AI tools (say, in customer service) affects the number of employees and their roles in that firm over time. This data will help identify whether Canadian companies and organizations are using AI in a complementary way.
If, for example, firms adopting AI consistently see output rise and headcount stay the same or even grow, that’s a good sign the AI is augmenting workers. Statistics Canada’s 2024 report on robotics technologies adoption indicates that businesses implementing automation often experience workforce growth, as these technologies enhance operational capacity. We need to continuously monitor if this holds true with AI across the economy.
Set and monitor targets for productive adoption
Having national or provincial targets (e.g., percentage of businesses using AI by year X) can focus efforts. But beyond sheer adoption, we should also set targets for outcomes.
For example, Canada could aim to climb into the top 10 of OECD countries for AI-driven productivity growth by 2030 or target a certain improvement in our ranking in global innovation indices. The government might also consider a target related to the nature of AI use – such as a goal that, by 2025, a majority of AI-adopting firms report that they are using AI to augment their workforce rather than reduce it.
In surveys, nearly half of businesses using AI (in particular, generative AI) say their goal is to automate work without cutting jobs, focusing instead on accelerating output. This is a positive sign, and policy can reinforce it by publicly recognizing companies that meet “augmented AI” criteria.
Annual public reports could highlight case studies where AI adoption led to both productivity gains and net new roles or upskilled workers, thereby creating exemplars for others to follow.
Assess worker well-being and job quality
Productivity statistics alone don’t tell the whole story. It’s equally important to gauge how AI is affecting the quality of jobs and worker satisfaction. Government agencies, possibly in partnership with academic researchers, should conduct periodic surveys or studies on employee experiences with AI in the workplace.
These could ask workers whether AI tools have made their jobs more interesting or less, whether their workload has increased or become more manageable, and whether they feel more empowered or more monitored due to AI. Worker happiness and engagement are legitimate policy concerns, because an AI-driven productivity boost that comes at the expense of an engaged workforce may not be sustainable or desirable.
Companies might also track this internally and share data (perhaps anonymously) as part of corporate social responsibility reporting. Policy-makers could encourage the use of frameworks like “algorithmic impact assessments” for workplace AI systems, requiring employers to evaluate effects on workers before implementation.
By capturing these qualitative outcomes, we can promote AI designs that improve worker satisfaction (for example, by offloading drudgery and allowing employees to focus on creative or interpersonal aspects of the job) and flag those that have negative impacts (like causing stress or excessive surveillance).
Align AI policy with employment policy
To ensure that AI supports workforce participation, we will need support workers as they transition into new roles as old ones evolve. This includes ongoing investments in retraining and lifelong learning initiatives. Governments should strengthen safety nets and active labour market programs that help displaced workers find new opportunities, ideally leveraging AI itself for personalized job matching and skills training.
In addition, we may consider policies like job-sharing or reduced work weeks if AI-driven productivity gains are very high – essentially ways to distribute work rather than simply eliminating positions. The underpinning idea is that the benefits of AI (higher productivity) should enable more people to participate meaningfully in the economy, not fewer. That could mean shorter hours for the same pay, or growth in industries that were previously resource-constrained.
Policy-makers will need to be creative in this area. Metrics such as the employment rate, labour force participation rate, and income distribution will tell us whether organizations and companies are harnessing AI in inclusive ways. If we see productivity rising but employment falling significantly, that’s a warning sign that adjustments are needed (perhaps through the tax or incentive policies discussed).
On the other hand, if AI contributes to rising incomes and maintained low unemployment – as has largely been the case so far, with unemployment near record lows in Canada despite more automation – then we’re on the right track.
Canada should treat AI policy as a learning process: gather data, involve stakeholders (business, labour, academia) in reviewing what the data show, and be ready to refine programs accordingly.
By actively measuring what we care about – productivity, jobs, and the quality of working life – we can ensure that AI adoption stays aligned with the public interest. This will build public trust in AI as well, addressing the hesitancy among Canadians who are less convinced of AI’s benefits.
Public sector as AI catalyst
Canada’s most significant AI opportunity lies in the outsized role the public sector plays in the economies. As a major employer, service provider, and procurer of technology, the public sector can serve as both a pioneer and proving ground for AI transformation in Canada.
Government as First Adopter
Canadian governments represent an ideal testing ground for AI-driven transformation. Service bottlenecks often provide a drag on overall economic growth and quality of life for Canadians. Given that government services, by their nature, are not productive, but rather infrastructure for productive work, efficiency is easier to measure and more magnified in impact.
Existing service bottlenecks persist because public sector processes serve as elaborate scaffolding around longstanding inefficiencies. The systems approach to AI elaborated above serves as a means by which governments can operate more efficiently and effectively.
Building in public: A three-fold impact
Government leadership in AI adoption would catalyze Canada’s broader AI ecosystem through three critical mechanisms:
- Creating implementation models: By building AI systems in public, government would establish proven implementation pathways that private enterprises – particularly Small and Medium-sized Enterprises (SMEs) – could adopt without reinventing the wheel.
- Stimulating demand: As a major buyer, government procurement would create robust demand for Canadian AI firms, helping promising start-ups scale into global competitors. This would provide research institutions like MILA, Vector Institute, and AMII with clear pathways to commercialize Canadian innovations.
- Demonstrating value: Successful government AI implementations would showcase tangible benefits, building public trust and addressing skepticism about AI’s value.
This approach has historical precedent. In the 1970s, the federal government’s strategic support for Northern Electric’s development of digital switching was pivotal in transforming Canada’s telecommunications infrastructure, propelling Canadian firms to global leadership positions. A similar opportunity exists with AI.
A multi-level government approach
For the effort to be succeed, all levels of government will need to work efficiently together:
Municipal initiatives
Cities can serve as living laboratories for productive AI adoption. Municipal governments can lead by implementing AI in public services to improve efficiency and quality of life. The City of Toronto’s AI-driven traffic management pilots (Hudes 2024) and Quebec City’s use of AI analysis (Canadian Press 2024) for urban monitoring demonstrate how municipal-level AI adoption can yield immediate benefits.
These smart-city initiatives not only improve governance quality but also serve as proofs of concept that inspire private-sector uptake. By opening city datasets and procurement opportunities, municipalities can provide AI developers with real-world problems to solve, creating replicable models for broader adoption.
Provincial initiatives
Provinces wield significant powers over education, economic policy, and digital transformation that can accelerate productive AI adoption. To address the knowledge gap identified by the Dais study (Lockhart 2023) – where many businesses “have not been able to make a business case for AI, and many don’t know what AI tools are available” – provinces could establish:
- “AI innovation vouchers” or cost-sharing grants allowing SMEs to consult AI experts;
- Connections between provincial AI research institutes and industry needs beyond the tech sector; and
- Provincial auditors general can focus on sector-specific audits (e.g., health care, education) to evaluate the effectiveness of AI tools in improving service delivery and operational efficiency with appropriate risk-management focus.
Provinces can also modernize regulations that inadvertently hamper useful AI deployments, following models from Singapore (Monetary Authority of Singapore 2023) and New Zealand (Department of Internal Affairs 2019) where digital-first administrative processes have reduced bureaucratic friction and enabled productive technology adoption.
Federal initiatives
The federal government sets the overarching economic conditions and can use its national reach to propel Canada’s AI adoption. Beyond the recently announced AI strategy for the federal public service, the government should pursue a more transformative vision through:
- Setting ambitious targets, such as achieving a 50 per cent AI adoption rate among Canadian businesses by 2030.
- Expanding programs like the Canada Digital Adoption Program (Innovation, Science and Economic Development Canada 2022) and Scale AI Supercluster (Scale AI 2024).
- Prioritizing funding for AI projects that involve system-wide transformation rather than just automating isolated tasks.
- The Minister of AI can work with the auditors general and treasury board to perform a comprehensive process mapping across federal departments to assess systemic bottlenecks and recommend areas for AI integration.
Conclusion
Canada stands at a crossroads in the AI era. We have the intellectual muscle – from top-notch AI researchers to innovative start-ups – and we’ve made early moves on national AI strategy and responsible AI governance. Now, we must match that with an equal commitment to deployment and diffusion of AI throughout our economy.
The payoff for getting this right is substantial: higher productivity growth to fuel prosperity, globally competitive industries, better public services, and jobs that are more engaging and rewarding. The risk of inaction or missteps, however, is that we squander our AI advantage and see our economic dynamism continue to slip. If Canada remains a place where AI breakthroughs happen in labs but are commercialized and applied elsewhere, we will feel the consequences in lower living standards and lost opportunities.
We need to align our tax and incentive structures with the outcome we want – productive, inclusive AI – by removing biases that encourage automation for its own sake. We need to invest in people and processes, not just shiny gadgets, recognizing that true innovation is often organizational as much as technical. We must update our education and training ecosystems to produce an AI-ready workforce that excels in areas where AI cannot. And we should embrace a model of innovation that is done with workers, not to workers, ensuring buy-in and shared benefits.
These ideas are not mere idealism; they are grounded in evidence and echoed by leading experts around the world who warn against simplistic technology will solve everything and instead advocate a balanced approach to AI adoption (Khurana 2019).
For Canada, encouraging productivity-boosting AI adoption is more than an economic strategy – it is a nation-building project for the 21st century. It means using our policy tools wisely to turn our comparative advantage in AI research into a comparative advantage in AI-enabled productivity. It means Canadians working smarter and more creatively, with AI as a partner. It also means maintaining our values, by channeling AI in ways that uphold dignity, fairness, and broad-based prosperity.
Countries like Finland, The Netherlands, and Denmark have shown that small, advanced economies can successfully integrate new technologies and enjoy strong productivity and social cohesion (Nosratabadi, Atobishi, and Hegedűs 2023). Canada can do the same. By implementing the recommendations in this report – and acting with urgency – we can ensure that the next generation of AI breakthroughs will not only be invented in Canada but also utilized here to the benefit of all Canadians.
The time to align our AI policy with our long-term economic and workforce goals is now, before the gap between possibility and reality widens further. With thoughtful policy leadership, Canada can turn the AI paradox into an AI promise: a future where innovation and adoption go hand in hand to secure our prosperity.
About the author
Ryan Khurana is a senior fellow at the Foundation for American Innovation and a contributing author to the Macdonald-Laurier Institute.
References
Acemoglu, D., A. Manera, and P. Restrepo. 2020. Does the U.S. Tax Code Favor Automation? National Bureau of Economic Research. Available at https://www.nber.org/papers/w27052.
Acemoglu, D., and S. Johnson. 2023. “Rebalancing AI.” Finance & Development Magazine, International Monetary Fund, December 2023. Available at https://www.imf.org/en/Publications/fandd/issues/2023/12/Rebalancing-AI-Acemoglu-Johnson.
Acemoglu, D., and P. Restrepo. 2018. Artificial Intelligence, Automation, and Work. National Bureau of Economic Research. Available at https://www.nber.org/papers/w24196.
Agrawal, A., J. Gans, and A. Goldfarb. 2018. Prediction Machines: The Simple Economics of Artificial Intelligence, January 2018. Harvard Business Review Press.
Agrawal, A., J. Gans, and A. Goldfarb. 2022. Power and Prediction: The Disruptive Economics of Artificial Intelligence. Harvard Business Review Press.
Bessen, J. E. 2015. Learning by Doing: The Real Connection between Innovation, Wages, and Wealth. Yale University Press.
Bryan, V,, S. Sood, and C. Johnston. 2024. Analysis on expected use of artificial intelligence by businesses in Canada, third quarter of 2024. Statistics Canada. Available at https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2024013-eng.htm.
Brynjolfsson, E., D. Li, and L. Raymond. 2025. Generative AI at Work. Quarterly Journal of Economics, 140(2), 889–938, February 4, 2025 Available at https://academic.oup.com/qje/article/140/2/889/7990658.
Canadian Press. 2024. Quebec municipalities using artificial intelligence to track tree cover, cars, pools. Lethbridge News Now, July 28, 2024. Available at https://lethbridgenewsnow.com/2024/07/28/quebec-municipalities-using-artificial-intelligence-to-track-tree-cover-cars-pools/.
Chatti, R., A. Di Marco, and Wanling Li. 2024. Robotics Technologies Adoption: Insights from the Survey of Advanced Technology. Statistics Canada, August 28, 2024. Available at https://www150.statcan.gc.ca/n1/pub/22-20-0001/222000012024004-eng.htm.
Department of Internal Affairs. 2019. Strategy for a Digital Public Service. Available at https://www.digital.govt.nz/digital-government/strategy/strategy-summary/strategy-for-a-digital-public-service.
Eastwood, Brian. 2024. “The who, what, and where of AI Adoption in America.” MIT Sloan Management Review, February 7, 2024. Available at https://mitsloan.mit.edu/ideas-made-to-matter/who-what-and-where-ai-adoption-america.
Fattorini, L., et al. 2024. AI Index Report 2024. Stanford Institute for Human-Centered Artificial Intelligence. Available at https://hai.stanford.edu/ai-index/global-vibrancy-tool.
Filippucci, F., et al. 2024. The Impact of Artificial Intelligence on Productivity, Distribution and Growth. OECD, April 16, 2024. Available at https://www.oecd.org/en/publications/the-impact-of-artificial-intelligence-on-productivity-distribution-and-growth_8d900037-en.html.
Geist, M. 2025. “Solomon’s Choice: Charting the Future of AI Policy in Canada”. Michael Geist, May 16, 2025. Available at https://www.michaelgeist.ca/2025/05/solomons-choice-the-future-of-ai-in-canada/
Gu, W., and M. Willox. 2023. The post-2001 productivity growth divergence between Canada and the United States: The role of the information and cultural services industry. Statistics Canada, December 21, 2023. Available at https://www150.statcan.gc.ca/n1/pub/36-28-0001/2023012/article/00006-eng.htm.
Hudes, S. 2024. Rogers to deploy 5G cameras, sensors to manage traffic at five Toronto intersections. CityNews, March 21, 2024. Available at https://toronto.citynews.ca/2024/03/21/rogers-to-deploy-5g-cameras-sensors-to-manage-traffic-at-five-toronto-intersections/.
Innovation, Science and Economic Development Canada. 2022. Canada Digital Adoption Program (CDAP). Available at https://ised-isde.canada.ca/site/canada-digital-adoption-program/en.
Khurana, Ryan. 2019. “The Threat of Automation is a Self-Fulfilling Prophecy.” Palladium Magazine, July 5, 2019. Available at https://www.palladiummag.com/2019/07/05/the-threat-of-automation-is-a-self-fulfilling-prophecy/.
Lockhart, Angus. 2023. Automation Nation? AI Adoption in Canadian Businesses. The Dais, September 2023. Available at https://dais.ca/reports/automation-nation-ai-adoption-in-canadian-businesses/.
Lowey, Mark. 2024. Canada’s Productivity Crisis is Obvious, but Fixing the Problem is Complicated. Research Money, October 30, 2024. Available at https://researchmoneyinc.com/article/canada-s-productivity-crisis-is-obvious-but-fixing-the-problem-is-complicated-canadas-productivity-summit-hears.
Monetary Authority of Singapore. 2023. MAS Strengthens Regulatory Measures for Digital Payment Token Services. November 23, 2023. Available at https://www.mas.gov.sg/news/media-releases/2023/mas-strengthens-regulatory-measures-for-digital-payment-token-services.
Nosratabadi, S., T. Atobishi, and S. Hegedűs. (2023). Social Sustainability of Digital Transformation: Empirical Evidence from EU-27 Countries. Administrative Sciences, 13(5), 126. Available at https://doi.org/10.3390/admsci13050126
Peng, S., et al. 2023. “The Impact of AI on Developer Productivity: Evidence from GitHub Copilot.” arXiv preprint, February 2023. Available at https://arxiv.org/pdf/2302.06590.
Scale AI. 2024. Our Impact. Available at https://www.scaleai.ca/about-us/impact/.
Waddell, K. 2019. “The hidden costs of AI.” Axios, October 26, 2019. Available at https://www.axios.com/2019/10/26/ai-accuracy-bias-hidden-costs.
