October, 2025 – Canada has world-class strength in AI research but continues to fall short in widespread adoption, according to a new report from the C.D. Howe Institute. On the heels of the federal government’s announcement of a new AI Strategy Task Force, the report highlights the urgent need to bridge the gap between research excellence and real-world adoption.
In “AI Is Not Rocket Science: Ideas for Achieving Liftoff in Canadian AI Adoption,” Kevin Leyton-Brown, Cinda Heeren, Joanna McGrenere, Raymond Ng, Margo Seltzer, Leonid Sigal, and Michiel van de Panne note that while Canada ranks second globally in top-tier AI researchers and first in the G7 for per capita publications, it is only 20th in AI adoption among OECD countries. “This matters for the economy as a whole, because such knowledge translation is a key vehicle for productivity growth,” the authors say. “It is terrible news, then, that Canada experienced almost no productivity growth in the last decade, compared with a rate 15 times higher in the United States.”
The authors argue that new approaches to knowledge translation are needed because AI is not “rocket science”: instead of focusing on a single industry sector, the discipline develops general-purpose technology that can be applied to almost anything. This makes it harder for Canadian firms to find the right expertise and for academics to sustain ties with industry. Existing approaches – funding academic research, directly subsidizing industry efforts through measures such as SR&ED and superclusters, and promoting partnerships through programs like Mitacs and NSERC Alliance – have not solved the problem.
Four ideas to help firms leverage Canadian academic strength to fuel their AI adoption include: a concierge service to match companies with experts, consulting tied to graduate student scholarships, “research trios” that link AI specialists with domain experts and industry, and a major expansion of AI training from basic literacy to dedicated degrees and continuing education. Drawing on their experiences at the University of British Columbia, the authors show how local initiatives are already bridging gaps between academia and industry – and argue these models should be scaled nationally.
“Canada’s unusual strength in AI research is an enormous asset, but it’s not going to translate into real-world productivity gains unless we find better ways to connect AI researchers and industrial players,” says Kevin Leyton-Brown, professor of computer science at the University of British Columbia and report co-author. “The challenge is not that AI is too complicated – it’s that it touches everything. That means new models of partnership, new incentives, and new approaches to education.”
AI Is Not Rocket Science- 4 Ideas in Detail
Idea 1: A Concierge Service for Matchmaking
We have seen that it is hard for industry partners to know who to contact when they want to learn more about AI. Conversely, it is at least as hard for AI experts to develop a broad enough understanding of the industry landscape to identify applications that would most benefit from their expertise. Given the potential gains to be had from increasing AI adoption across Canadian industry, nobody should be satisfied with the status quo.
We argue that this issue is best addressed by a “concierge service” that industry could contact when seeking AI expertise. While matchmaking would still be challenging for the service itself, it could meet this challenge by employing staff who are trained in eliciting the AI needs of industry partners, who understand enough about AI research to navigate the jargon, and who proactively keep track of the specific expertise of AI researchers across a given jurisdiction. This is specialized work that not everyone could perform! However, many qualified candidates do exist (e.g., PhDs in the mathematical sciences or engineering). Such staff could be funded in a variety of different ways: for example, by an AI institute; a virtual national institute focused on a given application area; a university-level centre like UBC’s Centre for Artificial Intelligence Decision-making and Action (CAIDA); a nonprofit like Mitacs; a provincial ministry for jobs and economic growth; or the new federal ministry of Artificial Intelligence and Digital Innovation.
Having set up an organization that facilitates matchmaking, it could make sense for the same office to provide additional services that speed AI adoption, but that are not core strengths of academics. Some examples include project management, programming, AI-specific skills training and recruitment, and so on. Overall, such an organization could be funded by some combination of direct government support, direct cost recovery, and an overhead model that reinvests revenue from successful projects into new initiatives.
Idea 2: Consultancy in Exchange for Student Scholarships
Many businesses that would benefit from adopting AI do not need custom research projects and do not want to wait a year or more to solve their problems. The lowest-hanging fruit for Canadian AI adoption is ensuring that industry is well informed about potentially useful, off-the-shelf AI technologies. We thus propose a mechanism under which AI experts would provide limited, free consulting to local industry. AI experts would opt in to being on a list of available consultants. A few hours of advice would be free to each company, which would then have the option of co-paying for a limited amount of additional consulting, after which it would pay full freight if both parties wanted to continue. The company would own any intellectual property arising from these conversations, which would thus focus on ideas in the public domain. If the company wanted to access university-owned IP, it could shift to a different arrangement, such as a research contract. This system would work best given a concierge service like the one we just described. The value offered per consulting hour clearly depends on the quality of the academic–industry match, and some kind of vetting system would be needed to ensure the eligibility of industry participants.
Why would an AI expert sign up to give advice to industry? All but the best-funded Canadian faculty working in AI report that obtaining enough funding to support their graduate students is a major stressor. Attempting to establish connections with industry is hard work, and such efforts pay off only if the industry partner signs on the dotted line and matching funds are approved. There is thus space to appeal to faculty with a model in which they “earn” student scholarships for a fixed amount of consulting work. For example, faculty could be offered a one- semester scholarship for every eight hours set aside for meetings with industry, meaning that one weekly “industry office hour” would indefinitely fund two graduate students. Consulting opportunities could also be offered directly to postdoctoral fellows or senior (e.g., post-candidacy) PhD students in exchange for fellowships. In such cases, trainees should be required to pass an interview, certifying that they have both the technical and soft skills necessary to succeed in the consulting role. The concierge service could help decide which industry partners could be routed to PhD students and which need the scarcer consulting slots staffed by faculty members.
The system would offer many benefits. From the industry perspective, it would make it straightforward to get just an hour or two of advice. This might often be enough to allow the company to start taking action towards AI adoption: there is a rich ecosystem of high-performance, reliable, and open-source AI tools; often, the hard part is knowing what tool to use in what way. Beyond the value of the advice itself, consulting meetings offer a strong basis for building relationships between academics and industry representatives, in which the academic plays the role of a useful problem solver rather than of a cold-calling salesperson. These relationships could thus help to incubate Mitacs/Alliance-style projects when research problems of mutual interest emerge (though also see our idea below about how restructuring such projects could help further).
For academics, the system would constitute a new avenue for student funding that would reward each hour spent with a predictable amount of student support. Furthermore, it would offer scaffolded opportunities to deepen connections with industry. The system would come with no reporting requirements beyond logging the time spent on consulting. The faculty member would be free to use earned scholarships to support any student (regardless, for example, of the overlap between the student’s research and the topics of interest to companies), increasing flexibility over the Mitacs/Alliance system, in which specific students work with industry partners. Students who self-funded via consulting would learn valuable skills and would expand their professional networks, improving prospects for post-graduation employment.
Finally, the system would also offer multiple benefits from the government’s perspective. It would generate unusually high levels of industrial impact per dollar spent (consider the number of contact hours between academia and industry achieved per dollar under the funding models mentioned in Section 3). All money would furthermore go towards student training. The system would automatically allocate money where it is most useful, directing student funding to faculty who are both eager to take on students and relevant to industry, all without the overhead of a peer-review process. And it would generate detailed impact reports as a side effect of its operations, since each hour of industry–academia contact would need to be logged to count towards student funding.
Idea 3: Grants for Research Trios
Our third proposal is an approach for expanding the Mitacs/Alliance model to make it work better for AI. Industry–academia partnerships leverage two key kinds of expertise from the academic side: methodological know-how for solving problems and knowledge about the application domain used for formulating such problems in the first place. In fields for which the set of industry partners is relatively small and relatively stable, it makes sense to ask the same academics to develop both kinds of expertise. In very general-purpose domains like AI, it holds back progress to ask AI experts to become domain experts, too. Instead, it makes sense to seek domain knowledge from other academics who already have it. We thus propose a mechanism that would fund “research trios” rather than bilateral research pairings. Each trio would contain an AI expert, an academic domain expert, and an industry partner. This approach capitalizes on the fact that there is a huge pool of academic talent outside core AI with deep disciplinary knowledge and a passion for applying AI. While such researchers are typically not in a position to deeply understand cutting-edge AI methodologies, they are ideally suited to serve as a bridge between researchers focused on AI methodologies and Canadian industrial players seeking to achieve real-world productivity gains. In our experience at UBC, the pool of non-AI domain experts with an interest in applying AI is considerably larger than the pool of AI experts. One advantage of this model is that projects can be initiated by the larger population of domain experts, who are also more likely to have appropriate connections to industry. Beyond this, involving domain experts increases the likelihood that a project will succeed and gives industry partners more reason to trust the process while a solution is being developed. The model meets a growing need for funding researchers outside computer science for projects that involve AI, rather than concentrating AI funding within a group of specialists. At the same time, it avoids the pitfall of encouraging bandwagon-jumping “applied AI” projects that lack adequate grounding in modern AI practices. Finally, it not only transfers AI knowledge to industry, but also does the same to both the domain expert and their students.
Idea 4: Greatly Expanded AI Training
As AI permeates the economy, Canada will face an increasing need for AI expertise. Today, that training comes mostly in the form of computer science degrees. Just as computer science split off from mathematics in the 1960s, AI is emerging today as a discipline distinct from computer science. In part, this shift is taking the form of recognizing that not every AI graduate needs to learn topics that computer science rightly considers part of its core, such as software engineering, operating systems, computer architecture, user interface design, computer graphics, and so on. Conversely, the shift sees new topics as core to the discipline. Most fundamental is machine learning. Dedicated training in AI will require a deeper focus on the mathematical foundations of probability and statistics, building to advanced topics such as deep learning, reinforcement learning, machine learning theory, and so on. Various AI modalities also deserve separate study, such as computer vision, natural language processing, multiagent systems, robotics, and reasoning. Training in ethics, optional in most computer science programs, will become essential.
Beyond dedicated training in the core discipline, we anticipate huge demand for broad-audience AI literacy training; for AI minors to complement other disciplinary specializations; for continuing education and “micro-credential” programs; and for executive education in AI. There is also a growing need for “AI Adoption Facilitators”: bridge-builders who can help established workers in medium-to-large organizations understand how data-driven tools could offer value in solving the problems they face. Training for this role would emphasize business principles and domain expertise, but would also require firmer foundations in machine learning and data science than are currently typical in those disciplines.
Read the full report via our friends at C.D. Howe Institute here.