Building Canada’s AI Frontier

Despite its academic prowess in artificial intelligence (AI), Canada is late in building a strong AI innovation ecosystem, but it is not too late. While Canada will not outcompete the United States or China in large language models (LLMs) and other AI models constructed from massive datasets and computing power, it is positioned to be a leader in developing niche – yet important – AI tools for drug discovery and other areas of science and in building less resource-intensive AI products. To achieve this, Canada must work smarter, leveraging its fewer resources by working collaboratively and openly with like-minded nations.

Current AI models are transformative, but they are also less efficient than the hype suggests. While generative AI is great at routine, well-trodden tasks, today’s AI models struggle with deep reasoning. Anybody who has asked an AI to solve moderately complex problems has experienced these limitations. These models are getting larger and better at prediction; however, they are still less capable when it comes to genuine problem-solving. 

This creates an opportunity for Canada. Rather than trying to emulate OpenAI, Anthropic, or Google, Canada needs to create a flexible, fast-learning AI innovation ecosystem that does not depend on single countries or suppliers. To achieve this, Canada needs to better embrace open-source and open science AI development.

Open science isn’t just an ethical preference, it’s a competitive strategy. The datasets essential for training and validating AI models can only be built through open collaboration. Closed, IP-intensive approaches risk limiting Canada’s own success, while open science ensures freedom to operate for Canadian firms and keeps AI development under Canadian control. 

To reinforce this point, Canada must avoid locking up its digital assets in the hope that it can outcompete players with far more resources and significant leads. Instead, Canada must leverage its digital assets – datasets, research capacity, skilled users – in concert with countries facing similar constraints and then focus on innovation in niche areas, such as  health, electric power distribution, and AI-powered robotics,  where global leadership can realistically be attained. 

The federal government’s AI Strategy recognizes as much in its Pillar 6, where it acknowledges that open-source AI “offers transparent tools that are cheaper to deploy and easier to fine-tune and adapt to specific use-cases, lowering barriers to discovery and adoption, particularly for not-for-profit organizations and SMEs.” The Strategy understands that market consolidation around proprietary systems creates real risks for sovereignty and resilience. As the Strategy puts it, countries that depend on closed platforms face “structural dependency, vendor lock-in, diminished transparency, and a limited ability to adapt critical technologies to local needs, cultures, and values.” 

This perspective is shared beyond government. Mozilla’s CEO, Mark Surman, recently argued that Canada’s leadership in open-source AI, matched by Europe’s commitment to place “open source at the centre of the EU’s technological sovereignty” represents the path forward for middle powers seeking both independence and value creation. He noted that “transparency is what can make AI safe and accountable by design,” a key government priority.

Conscience is building platforms to support this future. Its leadership in BEACON – the Benchmarking, Evaluation, and Assessment Consortium for Science – illustrates its commitment to advancing Canadian AI sovereignty. BEACON breaks through AI hype by offering a platform across the sciences through which to evaluate the performance of AI models through competitions where the results are openly shared. This builds on the foundation of Conscience’s CACHE challenges – Critical Assessment of Computational Hit-finding Experiments – which have helped Canadian AI firms to improve their ability to predict new molecules to treat diseases and to refine their processes.

To build a truly open science AI capacity in Canada, the government needs to invest in high-quality, standardized, and open-access data generation and sharing. AI models are only as good as the data used to train them, making high-quality datasets the cornerstone of state-of-the-art AI capacity. The government also needs to spearhead benchmarking and critical assessment efforts to level the playing field for Canadian academics and SMEs with true AI capabilities. Last but not least, the government needs to build and grow an AI community that embraces open science principles.

In a competitive world, open science offers the only realistic path for Canada to build AI products, transparently ensure their safety, and customize those products to enable adoption by Canadian firms. The government’s AI Strategy is a good first step, but to translate intention into action, the government must support the creation of open, high-quality datasets and promote a shared infrastructure upon which Canadian firms can compete both in Canada and internationally.