<\/p>\n
Michael Geist criticises Canada’s proposed AI training regulations, arguing they could reduce the visibility of Canadian works in global AI systems and affect cultural diversity.<\/p>\n<\/div>\n
Michael Geist has warned that a House of Commons heritage committee recommendation on artificial intelligence could leave Canadian culture less visible, not more, in the systems increasingly used to search, summarise and generate information. Writing after the committee\u2019s report on AI and the creative industries was published this month, he argued that the panel recognised the risk of Canadian works being under-represented in training data, but then embraced a rule that would make access to those works harder.<\/p>\n
At the heart of the dispute is the committee\u2019s call for an explicit opt-in requirement before copyrighted works can be used to train AI systems. Geist says that would put Canada well outside the approach taken in several other jurisdictions. He points to the European Union\u2019s text-and-data-mining regime, which generally allows use unless rights holders reserve their rights, as well as broader permissions in Japan, Singapore and parts of the United States, where AI training is treated more permissively under existing copyright doctrines. Canada\u2019s own privacy regulator has separately stressed that AI systems should be built and used with a clear legal basis, meaningful consent where personal information is involved, and greater transparency and accountability.<\/p>\n
The policy concern is not theoretical. Geist argues that Canada has already seen what happens when law and platform economics collide in the online information space: after the Online News Act created a regime requiring payment for news links, Meta blocked news content for Canadian users. He says a similar effect could follow if AI developers face extra costs and procedural hurdles before they can include Canadian material in their models. The same concern has been raised elsewhere. The Office of the Privacy Commissioner of Canada reported last year that LinkedIn paused the use of Canadian members\u2019 personal information for AI training after questions about notice and consent, underlining the sensitivity of data use in this area.<\/p>\n
There is also a deeper copyright issue. Canada\u2019s fair dealing provisions already allow copyrighted material to be used for purposes such as research and private study, and Geist contends that much text and data mining for AI training could already fit within that framework. An opt-in rule would therefore do more than add certainty: it would effectively reshape the balance by requiring prior authorisation where current law does not clearly demand it. That mirrors a wider international debate, with organisations and policy analysts split between those who argue that explicit consent best protects creators and those who warn that opt-out systems are harder to administer and may still fail to keep protected works out of training sets.<\/p>\n
Geist also questions how the committee reached such a sweeping conclusion. He says the hearings were dominated by collective rights organisations, cultural industry bodies and advocacy groups, while witnesses with more sceptical views on regulation were much less influential in the final recommendations. Industry submissions to Canada\u2019s AI consultation have also warned that rigid opt-out or opt-in systems may reduce the diversity of training data and worsen bias, a point echoed in a policy document from Innovation, Science and Economic Development Canada. The result, in Geist\u2019s view, is a report that presents itself as pro-innovation while recommending a framework that would make Canadian content harder to find inside the AI tools that are already shaping global access to culture.<\/p>\n
Inspired by headline at:<\/strong> [1]<\/a><\/sup><\/p>\n Sources by paragraph:<\/strong><\/p>\n\n