Good morning all, 

You might remember the $2.4B announcement made by the federal government last April for the implementation of a series of measures meant to Strengthen Canada’s AI Advantage. (If you don’t, or if you’d like a refresher, you can read about it in our Budget 2024 Commentary). Within those $2.4B, $50M were earmarked for the creation of a Safety Institute. While this announcement was generally well received by our Canadian AI godfathers, little information was available at that time to properly assess what this institute would look like – and if and how post-secondary education would be impacted.

Finally, earlier this week, the Government of Canada formally announced the creation of its Canadian AI Safety Institute (CAISI). Based on the government website, the CAISI “will leverage Canada’s strong AI research community as well as in-house expertise to conduct leading-edge research in collaboration with industry and the International Network of AI Safety Institutes. [It will] study how advanced AI systems work, what are the risks, and provide tools to ensure that the Government of Canada is well-positioned to address AI safety risks as they arise.” This will include applied and investigator-led research (funded through CIFAR, who just received a $27M contribution from the federal government) and government-directed projects.

CIFAR, in a press release earlier this week, stated that the applied and investigator-led research stream will “take a multidisciplinary approach, engaging experts from diverse fields to mitigate and address both the short and long-term risk of AI systems. By partnering with research institutions across the country, it will further the research Canada’s experts have already started in areas of AI safety, such as detecting AI-generated content, contributing to the evaluation of advanced AI models, ensuring the safe adoption of AI in high-risk applications and ensuring privacy in AI systems”.

So, what do we make of all this? Well, as you can see, there is still no clear indication regarding what the role of post-secondary institutions themselves will be precisely (if they’ll have one), although researchers that work on ethical uses of AI, AI privacy, AI safety and AI detection will probably all be eager to see how they can contribute to the research efforts (and be funded to do so).

I know that I, for one, strongly believe that post-secondary institutions should stay abreast of any research developments that comes out of CAISI. Indeed, when working with post-secondary institutions across the country in developing their policy response to AI (and to GenAI, most specifically), the main areas that institutions seemed to be grappling with were related to AI safety and privacy. “What AI solutions are trust-worthy enough to be endorsed by our institution?”, “How can I be sure that private information inputted in AI tools will be protected?”, etc. All questions to which there is no one single and easy answer, especially considering the blistering pace at which new models are being developed. 

Now, while Canada might have been the first country to publish a national AI strategy (the Pan-Canadian AI Strategy), it is neither the first nor the only country to have developed an AI safety institute (AISI). The US, the UK, the European Union, Japan and Singapore have also put up bodies “tasked with evaluating AI systems’ capabilities and risks, conducting fundamental safety research, disseminating their results and facilitating information exchange among stakeholders”, based on the OECD. And it seems like we might be lagging behind. The UK’s AISI is already reflecting on its first year of operations, during which it created evaluation suites to test the capabilities and safety of AI models, and conducted evaluations on 16 models. The US’s AISIC, on its end, already has in place a series of working groups – one of which is tasked at developing a companion resource for the US’ AI Risk Management Framework.AISIC’s already encompasses more than 300 member organizations, with around 50 of those being post-secondary institutions.   

Something else to consider – the AISIs across the globe don’t seem to all share similar scopes in mandates. 

Based on the information we got through this week’s federal government announcement, it seems like our very own CAISI will mostly focus on research and knowledge dissemination. While this is of course very valuable, it isn’t clear if and how any results from those research efforts will translate into policy or national standards. On its end, the EU’s AI Office has regulatory powers: it can “request information from model providers, and can apply sanctions to enforce the AI Act”.

What is encouraging, though, is to see that countries are already eager to collaborate on this front. An International Network for AI Safety was launched last May during the AI Seoul Summit, and will reconvene this coming week (Nov 20-21) in San Francisco.