Thank you, Chair.
Good afternoon, everyone.
My name's Michael Geist. I'm a law professor at the University of Ottawa, where I hold the Canada research chair in Internet and e-commerce law. I appear in a personal capacity representing only my own views.
Thanks for the invitation to appear on this important study on AI in creative industries.
As some of you may know, I've appeared many times before this committee on questions involving technology and culture, including studies on copyright, freedom of expression and Internet regulation. In each instance, much of the discussion amounted to risk analysis, the perceived risk arising out of new technologies, whether digital copyright, online platforms, streamers or digital advertising, and concerns about risks associated with some of the proposed legislative responses, such as anti-circumvention rules, regulating user content or blocking news links. I think too often the debate frames new technology as a threat, emphasizes cross-industry subsidies and misses the opportunities that new technology presents. We therefore need risk analysis that rejects entrenching the status quo and instead assesses the risks of both the technology and the policy responses.
The debate over AI faces a similar challenge. I think that helps explain why the government has shifted from AIDA—the former Bill C-27—to now warning against overindexing on AI regulation and why groups that typically call for copyright reform find themselves arguing against it before this committee at the moment. These highlight the challenges of identifying AI risk and the fear that some regulatory responses could themselves create new risks that outweigh the problems they're trying to solve.
What are the risks I think this committee needs to think about with respect to AI and the creative sector? Three issues that often arise in this area are freedom to create, appropriate protections and Canadian content presence or discoverability. I think each of these presents its own challenges.
First, with respect to freedom to create, AI is already an integral part of the creative process, used to assist with everything from writing to film to music. Given its importance, AI has real benefits and restrictions on AI use are not only unrealistic but may be harmful. The risk comes from misinformation or public confusion that can come from “AI slop” in a video context or poorly crafted AI-generated news. This content should be properly identified, which would enhance the value of original human creativity. There is a need to work with the relevant sectors—news, video, music and AI services—to develop appropriate transparency measures to more easily distinguish between human-generated and AI-generated content.
Second, copyright invariably arises when discussing appropriate protections. Yet in the context of AI, the application of copyright isn't always clear cut. The outputs of AI systems rarely rise to the level of actual infringement given the expression may be similar or inspired by a source, but is not a direct copy of the original. The inputs—such as inclusion in large language models—are currently the subject of numerous lawsuits, but few have to date resulted in liability since those cases suggest large language models, LLM, inclusion and the resulting data analysis often qualifies as fair use or fair dealing.
What are the risks here? To paraphrase Minister Solomon, overindexing on AI regulation in a copyright context risks creating barriers that would render us uncompetitive as a market, undermining both innovation and creators. If Canada makes it more difficult or costly to develop large language models, AI development will shift outside of the country. It's therefore essential to ensure that our copyright frameworks are globally competitive. That's why we need copyright laws that continue to strike the balance through effective fair dealing rules and, given the use of text and data mining exceptions elsewhere, including the EU, the appropriate exceptions that position Canada as receptive to AI opportunities.
Third, we want to ensure AI services feature relevant Canadian results, but conventional Canadian content presence or discoverability policies such as minimum content requirements or promotional presence efforts simply don't map onto AI. Indeed, these kinds of policies could backfire, leading to the exclusion of Canadian content in large language models, which would in turn result in reduced presence in AI outputs. Essentially, it would be a replay of what we've seen with news on some social media platforms, where there are fewer conventional news sources and more presence of substitutable alternatives. In other words, the answer to Canadian AI cultural relevance is more Canada in the training data. That doesn't come from more regulation, legal barriers or higher costs, rather, it requires transparency on datasets, reducing costly barriers to access and the development of public AI systems that encourage the use and availability of Canadian content.
I look forward to your questions.