Canadian Heritage Committee on Oct. 27th, 2025

[Étienne Grenier]

It's important to distinguish between the various types of data used to train models.

In Quebec, some bodies are in favour of providing descriptions of cultural content instead of content in its entirety. For example, that may involve providing basic information that makes it possible to find a song, rather than the entire content of a song. Working with descriptors and using their work to ensure that content is present in models can be a strategy to help get culture discovered without necessarily contributing to theft.

Otherwise, inspiration can come from approaches such as those in France, particularly that of the Bibliothèque nationale de France. The French government's strategy is to consider that cultural data as a treasure and a heritage, which government institutions promote to boost their discoverability. The training, models and challenges of discoverability are managed by government enterprises.

Fundamentally, technology isn't a problem for us. Technically speaking, neither is artificial intelligence. The issue is really who is handling it and for what purposes.

[Martin Champoux, Bloc, Drummond, QC]

Since I have about 30 seconds left, I will try to be brief.

Mr. Roberge, you said that your two research chairs had never received a request to conduct a study funded by Canadian Heritage. In addition, there are many complaints that Canada is falling behind in regulating all new technologies.

Do you feel that the Canadian government is depriving itself of the expertise of its researchers, which would undoubtedly enable it to be a little more at the forefront or, at the very least, a little less behind in regulating these rapidly changing subjects?

[Jonathan Roberge]

There is a great deal of expertise in Canada. This includes technical expertise, more cultural expertise, and expertise in scientific disciplines. Indeed, if we held them in higher esteem, we would be able to develop more robust and longer-term strategies.

That’s possible, but if we take the example of discoverability, Minister Joly organized the Discoverability Summit a few years ago. As far as I can remember, it never led to a very robust strategy.

The same is true for the first Artificial Intelligence and Data Act.

You can talk the talk, but can you walk the walk?

That is, to some extent, the problem. In culture, it’s actually a multifactorial issue. So we need fairly precise studies to know what we’re talking about. Today, we don’t know the potential economic impact on cultural workers, among other things.

A few moments ago, your colleague Mr. Myles asked whether this would cause harm. In fact, if we think globally and logically, artists will suffer this punishment in the form of lost income or, in some cases, job loss.

So, “planning ahead” means having public policies in place to reintegrate these people, whether through training or a social safety net, but we also need to plan for the longer term instead of just rushing through 30 days.

[Martin Champoux, Bloc, Drummond, QC]

I am extremely grateful for your response.

Thank you very much.

[The Chair, Liberal, Lisa Hepfner]

Mrs. Thomas, you have five minutes.

[Rachael Thomas, Conservative, Lethbridge, AB]

Perfect.

The first thing I'm going to do is actually give the mic back to Dr. Kearney to finish her thoughts with regard to the last comment.

[Alexandra Kearney]

Thank you.

I wanted to bring up a result I saw recently that was quite interesting. One of the things that is unique is that there's a synthesis of information from these systems. A lot of people will talk about how they're just copying; they're not actually doing anything semantically interesting under the hood—they're just taking data and copying it—but you can use some of these text models to generate ASCII art.

That's kind of interesting because they're autoregressive. They're just learning to take text input and generate text output in some cases, but because there's naturally space between letters, when you're training these autoregressive models, they can learn about the space, even though that wasn't something they were actually being originally trained for. They can generate ASCII art because they spatially understand letters.

I'm bringing this as an example because it's going beyond the inputs and finding something novel that we can understand. We know that when people paint pictures with letters and make these funny things on the Internet, these models are naturally and intuitively picking up on that by studying the text, even though that wasn't the initial problem formulation.

[Rachael Thomas, Conservative, Lethbridge, AB]

Dr. Kearney, I think that begs an obvious question for many in the room, and that is, if the model is essentially thinking on its own, where does that lead us in the future?

[Alexandra Kearney]

That's a really big question. I think it's one of the profound things that enables us to do.... I think you can see these systems as computational. When people think about machine learning, they're typically thinking about inputs mapped to outputs. You're learning a model. You're learning to classify whether or not there's a cat in a given picture. Back when I started, it was a pretty novel result to be able to tell that there was a cat in a picture. Now, because they're taking inputs and generating outputs, which then become inputs again for the system, you can see them as sort of being able to do this computational thing where they're looking at what they're doing and are able to use their internal thought process to make better results.

You see this in math problems. If you ask a model to answer a math question, it does poorly, but if you let it think a little bit about how it's supposed to answer the question and do internal steps much like we would, it gets better answers; it's more accurate at these math problems.

I think we're going to enter this interesting space. A lot of people are talking about generating text or generating songs, and those are very interesting domains, but I think the most interesting—and I'm biased here—are domains where the systems are able to dynamically adapt in real time and make decisions. That's an interesting frontier, and I think it's a frontier for creativity. That's why games are most interesting to me because, fundamentally, games are a medium of agency and decision-making.

[Rachael Thomas, Conservative, Lethbridge, AB]

Thank you.

Mr. Capobianco, my next question is for you. I guess I'm hoping that you can shed some further light on how AI is being used within the creative space to reach a further audience effectively.

[Adrian Capobianco]

One way is the customization of content. I talked about the examples of either a French-speaking or English-speaking creator to be able to deliver their content in multiple languages, which they would not have been able to do previously. Also, there are different international audiences as well.

In our personal experience, those are two very simple ways we can use to open new markets and customize content to help those creators get their content to land with the right audience.

[Rachael Thomas, Conservative, Lethbridge, AB]

One of the things that you talked about in a written piece was how Google is now filtering us off pages and how we're not clicking on content in the way we used to. As a result, we're not accessing ads that would normally have our attention.

You talked about the importance of creative spaces or of influencers being able to draw attention to their pages directly. Can you expand on that?

[Adrian Capobianco]

The context of that was in relation to search and AI search. I don't know the exact number, but it's something like 60% of searches that no longer result in a click. Maybe that's not a bad thing and maybe it's a good thing, but what it's doing is that when you search and you get an AI summary, that summary is much more succinct. It doesn't lead you to further content, whatever that may be. In the context of brands and advertisers, usually that's not a good thing, because you're not getting the traffic and the audience to your content that you normally would.

Part of the solution to that—and it's a little bit self-serving for what we do—is to move away from just that search content to social content. We call it “always in”: Your content is always in feed, in context, in zeitgeist.

Part of the solution to that limit on traffic and audience is to create always-in content in social media that exists across multiple platforms and multiple points in time to help offset that AI result and drive those audiences back.

[The Chair, Liberal, Lisa Hepfner]

Thank you.

Mr. Myles, you have five minutes.

[David Myles, Liberal, Fredericton—Oromocto, NB]

I've been watching the other witnesses wanting to speak and respond to what's been happening. I want to give Wyatt a chance, because I know he was really keen on speaking. Maybe he can respond to what's been said.

[Wyatt Tessari L'Allié]

Thank you.

I wholeheartedly agree with what Dr. Grenier was saying and I want to help translate it to what it means to this committee.

First of all, yes, using data through consent should be compensated, but I guess you're not going to be able to use that data to help fund artists in the future, basically.

AI is relying less and less on human content and is doing something novel enough that it won't be a business model to allow unique creators to be able to just say, “Well, if my art isn't a model, then I'll get paid for it”, and they'll survive on that. They may get some compensation, but it's not going to be enough to sustain them.

I think that if you want to look at positive solutions for artists in the future, think of it more as whether we want to pilot it at basic income or whether we want to have a rule so that they don't need to rely on that to be able to pay their rents and they can still create, because they'll do what they want to do.

I think there's high risk in wanting to tinker with the business models of the artists right now, when pretty much everything is going to change.

I just wanted to add that.

[David Myles, Liberal, Fredericton—Oromocto, NB]

I think one really interesting thing that you said is the idea that this is actually the precursor industry that many industries are going to go through. I've seen this with the devaluing of digital content. It happened first in some of our cultural industries, and then it happened in other industries.

You were suggesting that it's a bit of a canary in the coal mine, that we're ahead of the game, and that in fact this devaluation, this process, is actually going to be taking place in all industries.

Mr. Roberge, you may respond to these questions.

If what Mr. Tessari L’Allié said is true, what future do artists have, especially if they want a job that works and provides them with an income? How do you see the future?

[Jonathan Roberge]

Earlier, representatives from the Société des auteurs et autrices de radio, télévision et cinéma spoke about the manifesto entitled L'Art est humain!, which was read in Montreal at the beginning of the summer, if I remember correctly. It is worth remembering that the world of artistic creation and the cultural world are, above all, human worlds.

This afternoon, we also discussed the issue of the discoverability of cultural content. However, we are not far from the point where we will have to talk about the “rediscoverability” of artists in Canada. Soon, it will be possible to rediscover who is making art, who is expressing themselves authentically, and who is creating cultural products that are worth listening to, reading and consuming.

From an economic standpoint, with regard to jobs in the cultural sector, monitoring the situation step by step and planning for alternative mechanisms would be a good precautionary measure. Various countries, such as France and Belgium, are already considering substitution mechanisms. If Canada lags behind, it will be problematic because it will be too late. There will be fewer creators to produce Canadian content.

For example, earlier we discussed the issue of licensing and the music streaming model. I would remind you of the staggering number of Quebec and Canadian artists who, in recent years, have publicly stated that their income from streaming was $14 or $17. Such income does not provide a living for artists. Clearly, the streaming model is not enough to generate enthusiasm. We will have to think outside the box to find solutions.

[David Myles, Liberal, Fredericton—Oromocto, NB]

Thank you.

I'm going to split this with Zoe. Zoe is going to take the next question.

[Zoe Royer, Liberal, Port Moody—Coquitlam, BC]

Ms. Broadfoot, you spoke earlier about transparency, about rights holders, about data validity. You're listening to the conversation. Is there anything you would like to add? I really appreciated your insights earlier.

Thank you.

[Lisa Broadfoot]

I think there's an important distinction to be made with regard to AI tools that are already embedded in the processes for audiovisual production. There's really exciting work being done in reaching audiences. AI tools can help producers find audiences. They can help them schedule or do pre-visualization. There are lots of ways that AI tools are already working in our sector that are quite exciting.

Then there are AI inputs into training data, which we've talked a little bit about, and it's clear that copyrighted works have been used in this training.

The third piece is AI outputs, which right now are still nascent. They're not great. These are the text-to-video platforms that are out there. They're not fantastic, but even in what you do see, it is clear that they have been trained on copyrighted works, so there is an important improvement to be done, because the content that is being put out is competing directly with the materials that have been scraped and stolen from copyright holders.

The transparency piece is all about consent. It's where the rights holder—whether that's a producer or a writer or a director or an actor—has the opportunity to know and understand what works of theirs are being used, how they can choose or not choose to be licensed and play in that playground or not, and then be able to maintain that value for themselves. It's clear that these companies are using this work because it has value. It's called text and data mining, because it's extracting.

What we don't have now is an open marketplace where two parties come to the table to negotiate freely together. We have a closed marketplace for extraction, and transparency is the key that can unlock it.

[The Chair, Liberal, Lisa Hepfner]

Thank you for that.

If members are okay, we have time for a quick three minutes to the Conservatives, three minutes to the Liberals and two minutes to the Bloc Québécois. Is that okay?

Ms. Thomas, do you want to take your three minutes, or is somebody else on your team going to take it?

[Rachael Thomas, Conservative, Lethbridge, AB]

As a friendly agreement, I would actually give my other colleagues precedence. I will be using my time to check in with the clerk on a couple of things.

[The Chair, Liberal, Lisa Hepfner]

Okay.

Mr. Al Soud, you have three minutes.

[Fares Al Soud, Liberal, Mississauga Centre, ON]

Thank you, Madam Chair.

“AI can't take away from us the joy of doing that. What it can do is take away the market share.” I thought that was very well said.

The CMPA represents independent producers in film, TV and digital media. Its focus today centres on ensuring that the creative sector remains economically viable and globally competitive as AI tools transform industry.

Mr. Strati and Ms. Broadfoot, you both have extensive experience in production finance and business affairs, so I'll take this opportunity to ask if you can speak to how you see artificial intelligence reshaping production budgets and contractual norms in Canada's screen sector.