In spring 2026, Professor Dan Brown appeared before the Senate of Canada’s Standing Committee on Transport and Communications to discuss intellectual property and safety concerns raised by the rise of generative AI. He highlighted concerns about the use of copyrighted works to train generative AI systems and the potential impact of these technologies on creative professionals, particularly writers and creators from minoritized communities.
Professor Brown has been a faculty member at the Cheriton School of Computer Science since 2000. His research spans computational creativity, music information retrieval and bioinformatics. He holds a PhD in Computer Science from Cornell University and an S.B. in Mathematics and Computer Science from MIT.
Since 2015, Professor Brown’s research has focused primarily on computational creativity, a field that explores how AI systems can generate works that, if created by humans, would be considered creative. Although the field of computational creativity has existed for decades, the rapid emergence of large language models, vision language models and other generative AI technologies has brought ethical, legal and societal concerns to the forefront.
In this Q&A, Professor Brown discusses the opportunities and limitations of generative AI and explores the ethical, legal and societal questions these technologies raise about creativity, copyright, and the future of creative work.
Are generative AI systems genuinely creative?
Why don’t you start with the easy ones first?
To be more serious, this is a question that scholars have been arguing about for a long time. Some maintain that software can’t be creative: the creativity comes from the software’s programmers. In a more contemporary system like a generative AI system, they’d tend to assign the creativity to some combination of the human programmers and users and the humans who created the creative objects on which the model was trained.
Others look at the outputs of these systems that appear creative (that is, outputs that are surprising, good-quality examples of a category of object), and they say that it’s hard to draw the connection between the initial model design, the training data for a generative AI system, and how the user interacted with the system, and say that the model itself is responsible for its outputs and hence is considered creative.
I’ve spent most of the last decade trying not to take a position in this argument, but at this point I think the camp that says the models are themselves creative is probably the easier position to take. It also legitimately lets users off the hook when the model makes terrible outputs that they didn’t ask for, which I’ll talk about later.
One thing that AI systems crucially lack is a creative drive. People create for lots of reasons, but one of them is a genuine desire to see their ideas made more tangible, more real. AI systems obviously don’t have that, and for some scholars the debate ends there.
What copyright and intellectual property challenges do generative AI systems pose?
There are a few important ones.
First, does uncompensated use of copyrighted materials to train AI violate copyright law, and if so, what do we do about that violation? Who should be compensated, and how?
In Canada, this question has to be answered by looking at the “fair dealing” exemptions in the Copyright Act. None of current exemptions clearly applies to model training, and since generative AI models often create objects that compete with the objects in their training data (such as visual art “in the style of” a famous painter, or AI-generated news stories that result in people not reading human-authored newspaper articles,) it’s likely that, in Canada, this training would be considered infringement.
That said, lawsuits around the world have been filed on this very question. Here in Canada, the federal government is running a series of consultations and processes, including a review about AI and copyright, the outcome of which might make the answer clearer. My testimony before the Senate’s Standing Committee on Transport and Communications is part of a different process, and there is also the federal government’s AI Strategy Taskforce. Moving from these into a coherent legislative and policy framework will likely be challenging.
Second, can the products of these systems themselves receive any intellectual property protection, particularly when the systems themselves are the result of intellectual property infringement? This second question is especially important. A creator might be using a variety of generative AI tools in the creation, say, of an advertising design, and if they didn’t have copyright over the product of their creative process that would pose a real problem.
Can content generated by AI be copyrighted? If so, who owns the copyright?
We don’t know. The most likely answer is that this is going to depend on the outcome of lawsuits. One of the Senators asked me that question, and didn’t seem to like my answer, but that’s the best we have at this point.
One thing that’s interesting is that we do have some experience with parallel questions, where for example artists use human assistants to help them create their sculptures, or celebrities use ghostwriters to help them write their memoirs. This case law about humans assisting humans in the creative process, and how that affects who deserves copyright, is going to be very important as we develop these systems.
I’d argue that we’re at a point where the users of these systems have to be able to copyright their outputs, because otherwise the alternative is that everything created with generative AI systems exists in the public domain. There are people who would see that as a terrific opportunity, but I think that intellectual property protection is a spur for creativity. I also think it’d be very strange if certain kinds of creative works still had IP protection in the future (because generative AI isn’t used much in their creation) and others didn’t, because users started to rely on generative AI.
One specific case of this overall concern is software: if programmers are going to use code engines like Claude Code as part of their software design process, they need to have the same intellectual property protections that they’ve always had, and that includes copyright. And if Canada were to decide that software written with the use of AI tools shouldn’t have copyright protection, it would be a catastrophe for our tech industry.
How are generative AI systems affecting creative professionals such as writers, photographers, designers, illustrators and musicians?
There’s no one answer, obviously. Lots of creative professionals have used these tools and their predecessors for years in their creative process. The animation and special effects industries had been completely revolutionized by previous tools in computer graphics as well. Designers are using these tools all the time, and they’re starting to be built into the software they use in their workflows, making their use pretty much unavoidable.
I think fiction writers may be one of the angriest groups, and one of the ones most resistant to “collaborating” with AI tools, since they see their products as being quintessentially human.
So it varies. There’s also starting to be evidence that managers like these tools more than do their employees, which isn’t surprising, given how many layoffs have started to happen in a variety of industries.
There’s a hollowing out of entry-level roles in creative industries, which might make a long-term sustainability crisis for the workforce in those industries.
You recently published a research paper on the experiences of LGBTQ and disabled fiction authors with generative AI. What are the study’s main findings?
Our study was open to all fiction writers, but because of the way we sampled it turned out that a large fraction of our participants were queer and disabled; we decided this was a feature and focused our analysis on that group. They’re more pessimistic about their career prospects than were straight people or non-disabled people. They’re also more angry about the impact of AI slop on their fields than those in other creative fields.
Our participants had various levels of AI aversion. Some merely refused to use these systems, while others wouldn’t work with publishers who had in other cases used them, and other participants even cut off contact with colleagues who used them. Many of our participants said that the existence of these systems was impairing their mental health, as well.
Another outcome of our study is some recommendations about how policymakers can respond to the negative consequences generative AI creates for some creators. In particular, we suggest establishing something like the Public Lending Right (which in Canada pays authors an annual royalty based on the prevalence of their works in Canadian lending libraries, though parallel systems with different details exist in dozens of countries), paid for via a tax on the usage of generative AI systems. That said, given current politics, such a program is unlikely to be created.
Do generative AI systems pose unique challenges for creators from minoritized and underrepresented groups?
I think so. It’s worth saying that this didn’t come up in our study, to our surprise: although participants hated generative AI systems, they didn’t hate them because of their discriminatory outcomes.
But we know that previous important information systems have yielded terrible examples for minoritized groups. There’s an old case of Google search routinely showing porn ads when people searched for “black girls,” for example. Queer kids today might well use chatbots and other generative AI systems to look for examples of their community, and it’d be terrible if they didn’t receive good, supportive outputs. On the other hand, the authors in our study were adamant that they didn’t want their data used in training these models. If their work weren’t available, the generative AI system might be prone to even worse bias, or might even filter out queer content for reasons of “safety,” in a form of algorithmic censorship. So it’s quite the conundrum.
Who should be held accountable when generative AI systems produce harmful or dangerous outputs?
I think that depends on how the harm happened. If a user deliberately manipulates an AI system into generating deeply problematic sexual images, for example, the fault is mainly with the user. By contrast, if the user winds up in a bad place by happenstance, that’s no different than when a Google search user winds up on a deeply problematic link after an innocuous search, an experience I think most of us have had.
Unfortunately, and I mentioned this in my Senate testimony, it’s extremely hard to hold the creators and users of the most problematic generative AI systems to account. A user with an open-source image model and a decent GPU can train their model to create custom imagery that is deeply upsetting, and they can release it widely, with relative ease. They’re clearly at fault, but it’s very hard to hold them to account.