THE STANDING SENATE COMMITTEE ON TRANSPORT AND COMMUNICATIONS
EVIDENCE
OTTAWA, Tuesday, May 5, 2026
The Standing Senate Committee on Transport and Communications met this day at 9 a.m. [ET] to examine and report on the opportunities and challenges of artificial intelligence (AI) in the information and communication technology sector; and, in camera, to consider a draft report.
Senator David M. Wells (Chair) in the chair.
[English]
The Chair: Good morning, everyone. My name is David Wells, senator from Newfoundland and Labrador. I would like to ask my colleagues to introduce themselves.
Senator Simons: Senator Paula Simons, Alberta, Treaty 6 territory.
Senator Wilson: Senator Duncan Wilson, British Columbia.
Senator Mohamed: Good morning. Senator Farah Mohamed, Ontario.
[Translation]
Senator Cormier: René Cormier from New Brunswick.
Senator Arnold: Dawn Arnold from New Brunswick.
[English]
Senator Lewis: Todd Lewis, Saskatchewan.
[Translation]
Senator Miville-Dechêne: Julie Miville-Dechêne from Quebec.
[English]
Senator Dasko: Donna Dasko, Ontario.
The Chair: Thank you, colleagues.
I would like to thank Senator Wilson for chairing last week’s meeting in my absence and that of our deputy chair.
I would also like to welcome everyone with us today, as well as those listening online on the Senate’s website, sencanada.ca.
I would now like to introduce our witnesses from Innovation, Science and Economic Development Canada, or ISED: Samir Chhabra, Director General, Marketplace Framework Policy Branch; Runa Angus, Senior Director, Marketplace Framework Policy Branch; and Sébastien Lauzier, Acting Director, Marketplace Framework Policy Branch. I assume you all know each other because you’re from the same group. Thank you for rejoining us today. I understand you’ve been here before.
A reminder that ISED officials, as I said, came first before the committee at the start of the study on November 25, 2025. With that in mind, they won’t make any opening remarks, and we will dedicate the full time to questions and answers.
I would like to advise senators that you have approximately five minutes for each round. We’ll have a second round of questions if time permits. Should you wish to ask a question of our witnesses, please alert our clerk, who will add your name to the list of questioners.
I am flexible on allowing a conversation to go over time as well.
I invite our deputy chair, Senator Dasko, to ask the first question.
Senator Dasko: Thank you, chair. I was actually looking forward to opening remarks. Anyway, thank you, witnesses, for being here. My question would be this: How far along is the government in its deliberations and its plans for legislation for AI? Where are you and what considerations do you have, especially when it comes to the area we’ve been studying, which is AI and communications? We focused on certain issues. Is there anything you can tell us about those in particular?
Samir Chhabra, Director General, Marketplace Framework Policy Branch, Innovation, Science and Economic Development Canada: Thank you for the question, senator, and the opportunity to be here with the committee. Obviously, we’ve been following the deliberations of this committee quite closely, including the testimony of all the witnesses who have come forward. A number of them, of course, have also been participants in various ways in the government’s own exercise on the development of the AI strategy, which included the task force of 28 members as well as the public consultation that garnered about 11,000 inputs in the fall.
The question, when I think about it, has multiple answers. When we think about AI today, both the opportunities and the risks, obviously, the next milestone that the government’s tracking toward is the release of its AI strategy. It was highlighted in the Spring Economic Statement last week: “AI for All,” identifying six pillars of work to bring forward. This work has been proceeding well. It’s following on the task force work that was undertaken on the inputs brought in during the fall.
I think your question is specifically about legislation, and on that front, I would say there are a number of different pieces of legislation that exist today that touch on AI from a number of different angles. These include the Copyright Act and the Personal Information Protection and Electronic Documents Act, or PIPEDA. Within government, there are a number of rules and norms that apply to how AI is used and applied. There are, I think, multiple layers of activity.
Minister Solomon has been clear about his intent to modernize PIPEDA to better prepare it for action in the current environment, including addressing the risks emerging from AI. I think, depending on the nature of the question and the specific area you’re interested in understanding, we can talk a bit more about what policy efforts might be under way in that domain.
Senator Dasko: Let me just focus on the big picture. When we started studying this and in the readings that I was doing earlier on, I understood — which was maybe incorrect — that the government was trying to focus on innovation and building AI, which sometimes means less regulation. I was reading material that said that was the government’s focus. However, more recently, the readings I’ve been looking at say that the focus now is more on creating a regulatory framework to deal with the, shall we say, harms or downsides of AI. I wonder if you might have any comment about that.
Mr. Chhabra: Thank you again for the question, senator.
When you look at the six pillars of the AI strategy that were updated in the Spring Economic Update last week, the first pillar is about protecting Canadians and safeguarding our democracy. That really reflects some of the issues that have come forward over the past number of years regarding AI, particularly around privacy protection as a key dimension. Online safety is another key dimension that can be significantly influenced by AI tools and how they’re integrated into the tools used on social media and other platforms, for example. There are meaningful steps that the government is working toward taking when it comes to safety and protection.
At the same time, there are five other pillars of that strategy around empowering Canadians regarding skill development, adoption for shared prosperity, sovereignty, scaling Canadian champions and building trusted partnerships. Rather than an either-or, it is a balanced framework that seeks to take into account protection where it needs to be considered and adoption and productivity-enhancing benefits where those are most prevalent.
There are opportunities and risks we might be mindful of, similar to any immense, dual-use-capable technology. We need to be mindful of the risks while also taking steps to better integrate and use the productivity-enhancing benefits.
Senator Dasko: Do you subscribe to the view that Europeans have gone too far in regulation and have stifled innovation?
Mr. Chhabra: Different jurisdictions have taken different approaches to it. The EU was the first out of the gate to have a comprehensive, horizontal framework in place. Other jurisdictions have taken the approach of modifying existing sector-specific rules and laws to deal with the protection issues.
I think that Canada’s approach is now starting to be more akin to the United Kingdom’s sector-specific approaches in dealing with things like privacy and online harms and in addressing AI in the context of transportation. Those are approaches that we’re seeing happen in different jurisdictions as well. I think it’s about recognizing where there is potentially a need for a horizontal or overarching legislative framework that might deal with some gaps in existing frameworks and where to enhance existing frameworks and ensure they’re modernized, up-to-date and deal with the risks presented by AI. Those may be different and require slightly different approaches to capture fully within the existing framework.
We’ve suggested that PIPEDA is a good example of that, as there are opportunities to strengthen it and make it more robust for the current age. This committee has heard from certain stakeholders as well who were very clear that certain laws should not be updated or modernized. There are different perspectives around, for example, the Copyright Act and whether that needs to be updated or should be retained as it is to allow for adjudication of the benefits and the rights and to be taken through the courts as it is right now.
There are different perspectives and approaches being taken by different jurisdictions. I’m not sure we’ve necessarily seen one emerge as the appropriate or right strategy. I think it depends on existing frameworks in the country. I think it depends on a constitutional basis. I think it depends on where they’re at politically as well.
Senator Lewis: To your point, everything is moving so quickly in this space. Things have probably even changed since you were last here. There is something every day, and to the point about court cases and so on, some of those decisions may show some of those gaps, and when you talk about strategy, I’m not going to ask if it’s short-term or long-term because I don’t know if that fits in with AI. It’s probably all short-term. It changes so quickly. Do you feel your focus will be nimble enough to meet these new challenges as they come up?
Mr. Chhabra: Thank you very much for the question, senator. I think the question about agility and nimbleness is about constancy in a way. It’s about constantly monitoring and being alive to what is going on in the marketplace. I will give you a small example.
We ran a consultation on copyright and AI a few years back, in 2023. We had around 1,000 respondents to that. We were asking questions about how training data should be treated and whether the outputs of AI models could be subject to copyright. We had run that exact same consultation with the same three themes and, essentially, the same questions a few years prior to that, and we had a small handful — a few dozen — of responses, most of which said that it’s too early to opine on these questions.
So, exactly to your point, senator, in terms of how things are moving rapidly, we went from “it’s too early to talk about this” in 2021, roughly, to “this is really a significant, burning issue,” where we received 1,000 respondents on these questions. Nobody in 2023 was saying that it was too early or too hard to tell what to do. There were strong opinions by 2023.
That ability to constantly probe, test, consult, engage, listen to the market and watch what is going on is what positions the government to be ready to respond when the moment is right to take action.
Senator Wilson: I have no questions at this time.
Senator Simons: Thank you very much for coming back. It’s amazing how quickly things have evolved in this space and how much has changed since you were last before us.
I want to start with a much more granular question. It derives from a brief we received yesterday from News Media Canada. It delved into an issue about which we didn’t hear a lot of oral testimony, and it’s about competition. News Media Canada is suggesting the Minister of Industry ask the Competition Bureau to conduct a study into the state of competition with respect to search and AI. They say in their brief:
It is in the public interest to have Googlebot split into separate crawlers – one for AI and one for search.
They say that would help level the playing field between publishers and Google. Apparently, they say that Google already does this for nearly 20 other crawlers. They say it’s “. . . not only technically feasible, but also a necessary and proportionate remedy . . .” that empowers website operators to have granular control they currently lack without increasing traffic load from crawlers to their websites.
I wonder if they have made that proposition to you and what you think about it.
Mr. Chhabra: Thank you for the question, senator.
It’s a fascinating space where you see the intersection of AI as a general-purpose and emerging technology with a bunch of existing marketplace frameworks, including with the Competition Act. The bureau itself has been really active in this space for a number of years; in fact, they have conducted studies already in the space of algorithmic pricing and better understanding how the advent and utilization of AI within corporate Canada could enable, for example, collusion without direct human engagement.
Also, a lot of work on this issue has been done at the OECD and G7 levels. It’s something Canada took a strong lead on during our G7 presidency last year — to look at AI and competition impacts.
The specific issue that you’re raising is a fascinating one. It’s actually one that demonstrates the power of the technology and its disruptive potential, as well. If we think back to a few years ago, everybody was using Google pretty much constantly as the way to go and search on the web. There are other search opportunities, but very few —
Senator Simons: Nobody is using Bing. Despite the fact that Bing comes preloaded on my laptop, I am not using Bing.
Mr. Chhabra: There’s another fascinating competition question about pre-loading, as well.
But we’re now in a situation where search queries being run are diversifying for the first time in years; we are starting to see people using AI tools directly, such as ChatGPT, Claude or Gemini — or something along those lines — to search out information.
Of course, you’re also seeing AI being baked into search browsers.
Part of the issue that the group you’re referring to might be flagging is that this is materially changing the way search engine optimization works, the way results are presented and the entire marketing end of that and how that gets dealt with.
It’s important to continue to monitor these things. The bureau has other active work under way in this space, including — I think it’s public knowledge — around ad tech, understanding that stack and how that functions. There have been a number of antitrust cases in the United States around Google, as well, reflecting whether search and Chrome, the browser itself, need to be disaggregated from each other. In the tech space as well, these are active U.S. considerations.
It’s a very active space, and it needs to be thought through clearly and carefully. At each technological advent, you do see some differential effects on competition around access and visibility. Those things need to be monitored pretty carefully.
Senator Simons: It will be interesting to see whether market competition comes back into the space. Having just made fun of Bing on the record, I’m using Google search less and less because the search results that come up are more often filled with AI mistakes, misinformation and false data. I can’t say the title of Corey Doctorow’s book on the record, but we know what it’s called.
I’m seeing the quality of the search return get poorer and poorer because AI is inaccurate most of the time. So, what concerns me from a marketplace perspective is that, while this might add more competition, there is no one to tell you that the AI is returning you nonsense. For consumers who are used to the idea that Google results are, prima facie, accurate — to constantly receive search results that are just fake — is there any way or any capacity to alert consumers to the fact that AI search is fundamentally unreliable?
Have you ever done an AI search that gave you the right answer? I haven’t.
Mr. Chhabra: Maybe I’ll answer in a few ways.
First, we pay close attention to the calibre and fidelity of the products that are available in the market today. I think most observers and most benchmarks used to evaluate these products and tools — and I’m talking about large language models, generally, and their systems — tend to demonstrate a significant increase in capability over the past few years. Since we started this conversation in 2022, or when ChatGPT became largely public in early 2023, what we’re seeing from a scientific perspective is that the existing benchmarks around capability and performance that we used to test these models are actually becoming saturated. That means it’s becoming harder and harder to differentiate between the next generation of a model and the previous one because the scores are so significantly high across the board. The benchmarks themselves aren’t keeping up to the level of capability in some of these tools.
We’re seeing in the market today, at least from my own experience — and I do these things all the time to keep abreast of the things that are going on, and with my kids, as well, in terms of seeing what they’re doing with their searches — it’s pretty typical to find that AI can make mistakes, which is something of a disclaimer they provide with it.
The broader part of your question is this: How do we see the quality, verifiability and robustness of these tools, and how can we rely on them? Those are really important questions. Going back to your earlier points about competition, that is why competition is so is critical to ensuring the market shakes out in a way that results in quality products at affordability prices that people can access. If one of these dominant players stops providing a consistent quality of service and output, I think you’ll see shifts in market preferences that indicate that.
The Chair: Before going on, I would like to welcome Senator Quinn from New Brunswick, who has just joined us.
[Translation]
Senator Miville-Dechêne: Welcome to all three of you.
Yesterday, I met with Dr. Labuz, a Polish expert on online safety for children, as part of his visit to Canada. He noted that we are somewhat behind on these issues.
I am going to ask you a more specific question than some of my colleagues.
Let’s take the two main sources of concern in relation to AI: chatbots and deepfakes. You work in a department that deals with these issues. Beyond your statement that you intend to take action, more specifically, will all forms of deepfakes be addressed? How are you ensuring that children are protected? Chatbots pose a more serious threat, because it involves relationships that children form with bots that can cause them immense harm, all within the privacy of an online relationship.
What can be done, and can it be done quickly? After all, in the meantime, children are being harmed.
Mr. Chhabra: Thank you for the question. I invite my colleagues to also respond.
[English]
I’ll start by noting that the questions you’re raising are significant and broad in nature. They’re not necessarily questions that lend themselves to a single response but rather a multivariate response that forms a web of protections that are effective and robust against a variety of use cases. I’ll offer that the government has already taken steps in this regard, including through Bill C-16. That’s one piece of the puzzle. I don’t mean to suggest that it is the only or a sufficient piece of the puzzle when it comes to dealing with “deepfakes.”
We’ve also been clear that the minister has highlighted his intention to update and modernize PIPEDA, including to take better action against “deepfakes.” “Deepfakes,” when they are using your personal information, are, in fact, a reflection of your personal information, so PIPEDA would apply. That means consent rules and things like appropriate purpose tests would apply.
There is already a framework in place that can help Canadians deal with the risk of “deepfakes.” Part of what we’re working toward in the PIPEDA modernization proposal is to ensure that Canadians have robust access, stronger enforcement and clearer understanding of the rules and how they apply so it can help them take better action in this context.
There is also a scientific angle —
Senator Miville-Dechêne: With respect to “deepfakes,” would it go through a question of consent and sharing of information?
Mr. Chhabra: As I said, senator, there are a number of different approaches that the government is working toward taking in this space. It’s not a singular response. It’s about having a series of tools that can actually help to be agile and effective in a number of different use cases.
The other part of this is really about the science around how this functions in terms of being able to discriminate and identify between genuine content and artificially generated content. There is work under way on that front within the Canadian AI Safety Institute as well as among our partners in the International Network for Advanced AI Measurement, Evaluation, and Science.
Last year, we co-published with Australia a research agenda on synthetic content, meaning content generated by machines rather than genuine content. Again, we’re trying to ensure that it’s not just within our network but broader, with other labs, academics and others around the world who can all look at an agenda and determine that these are the kinds of questions we need to better wrestle with and understand. How do we work on the scientific and technical requirements to help us better understand and discriminate between what’s real and what is artificially generated?
Senator Miville-Dechêne: Because it’s more and more difficult, almost impossible, to distinguish — as this expert was telling us — between reality and a fake.
Mr. Chhabra: Because the fidelity is becoming very good. Absolutely.
The other part of your question is around how interactions with AI are affecting people and affecting kids. Again, that is another space that the government has already leaned into. At Mila in Montreal, through the Canadian AI Safety Institute, we funded the establishment of the AI Safety Studio. It’s doing work right now — it’s active and under way — around what kinds of risks present themselves to Canadians, to kids, when they’re dealing with chatbots, including the psychological effects of a long-form interaction. We are also studying and understanding how chatbots maintain the guardrails that they’re established with through longer-form conversations. There is some evidence that suggests that longer interactions can enable or allow chatbots to start to, in a way — not to anthropomorphize them — almost forget the guardrails. The guardrails potentially become less robust over time during long-form conversations.
That is another piece of work and research that is under way, not just in the case at Mila but around the world. It is an active question about how to better understand this and how to ensure that the safeguards and guardrails that are put on these models are robust over a longer term.
Senator Miville-Dechêne: Are there some safeguards around chatbots? Is there anything in place by platforms to protect children now, or is this the job you will have to do? You’re talking about safeguards in the present term.
Mr. Chhabra: Sure. If you’re talking about chatbots, the broadly publicly available ones, we see a breadth of approaches being taken right now. A number of the larger companies have published their model cards and system information about the safeguards that they’ve implemented — around the kinds of things that you could or could not expect the chatbot to respond to. An example that’s often used in this case is CBRN risk — chemical, biological, radiological and nuclear — asking a bot a question about how to build a device. It will usually return a result saying, “I can’t talk to you about that.” That’s an example of the kinds of safeguards that tend to be built in.
Increasingly, over the past year or so, as psychological harms and risks have emerged, we’ve seen some companies recognizing when there is a potential mental health concern or event and taking steps to refer the user to a mental health professional or service. These are more nascent or emerging, and I do think there are opportunities to continue to work with the industry and think about how to better standardize or encourage these approaches.
Senator Mohamed: Your colleagues are welcome to jump in here, as well. We’ve heard from a couple of witnesses that our adoption rates are lagging behind international numbers. I wonder if you might reflect on why you think that is and what we can do to change that. That’s my first question.
Second, we heard from a few witnesses, and it kind of stuck with me that the department is using AI to develop its AI strategy. I wonder if you wanted to reflect on that. That is out there, so I think we should hear from you directly on it.
Finally, aside from the principle of “do no harm” in coming up with the AI strategy, perhaps you could provide some other top principles guiding the development of the strategy. Thank you.
Mr. Chhabra: Thanks very much for your question, senator. Maybe I’ll invite my colleague Sébastien to talk about adoption rates around the world in terms of what we’re seeing.
Sébastien Lauzier, Acting Director, Marketplace Framework Policy Branch, Innovation, Science and Economic Development Canada: In terms of adoption, it’s true that if you compare with certain other countries, Canada has been lagging. What’s important to flag is that adoption has been increasing quite a lot. According to Statistics Canada, it has doubled from 2024 to 2025.
One of several challenges that Canada faces is that we have many SMEs in our economy, and it’s often more difficult in terms of getting capital, time and resources. You see that in the Canadian economy, the rate of adoption for bigger businesses is quite a bit higher. I think there is a KPMG survey that says businesses making over $100 million in Canada have an over 80% adoption rate, while the actual adoption rate in the economy for all businesses is under 20%, so that poses a real challenge with the makeup of our economy.
That’s kind of core to what the government has been doing in the past few years but also to the six pillars of the upcoming strategy released in the fiscal update. I think you will recognize there is more than one lever that can be pulled to increase adoption. We have great talent, researchers and institutes — you’ve heard that from other witnesses here — but how do we give them the capital? How do we give them the compute infrastructure, which was part of the compute strategy that the government launched in 2024? Those discussions we just had about governance and trust are very important. Can we make a regulatory framework that will make companies confident about adopting AI? I think it’s part of a package.
Senator Mohamed: Is there a percentage that would suggest success as an adoption rate in Canada? Big companies, I understand, are on the path, but for SMEs, where we can alleviate a lot of the pressures they face?
Mr. Lauzier: The overall goal is to increase productivity and economic growth through AI. A high adoption rate of AI by itself is not necessarily a success. It’s more what goes along with the adoption rate that will show success. That is how the government is looking at it: how it will affect productivity and economic growth.
Mr. Chhabra: I’ll note that I think the AI strategy itself will have a bit more to say about that kind of approach. We saw the emergence of the pillars through the Spring Economic Update last week and a little bit of colour around what those pillars are about. When you look at the adoption section, which is the third pillar of the strategy, you’ll get a bit more detail to understand where the priority areas are. As Sébastien pointed out, it’s not just about a top-line number around adoption; it is also about recognizing where we’re adopting in a way that maximizes impact and benefit.
To your question about how AI has been utilized in the development of the strategy itself, the government has been pretty clear that, in going through 28,000 public inputs, there was a very well-monitored and governed approach that was overseen and supported by the policy framework that was established by Treasury Board on automated decision making. That was the approach that was taken to go through all of those submissions and essentially summarize them and bring them to where we could look at them and see the data and the assessments that were emerging.
I think your other question was about the principles that were underpinning the strategy. I think AI for All is what really encapsulates it. The government is really working to ensure that the approach that is taken on AI is one that benefits all Canadians. So there are considerations around skills and development; adoption; growth and economic potential; and, absolutely, around protecting Canadians and safeguarding our democracy as well.
It is really about a rounded-out approach. The pillars are essentially representative of the principles, including things like sovereignty and growing our engagement and partnerships with alliances around the world.
Senator Mohamed: Thank you. I was kind of hoping to hear about things like fairness, non-discrimination or other areas where people have fear in terms of how AI is used — transparency, explainability and so on.
Mr. Chhabra: That is a really important question. I would say that the “Protecting Canadians and Safeguarding our Democracy” pillar does then have a series of other important elements in it. What does it mean to protect Canadians in this context? Well, fairness is a critical piece of it. And safeguarding democracy gets back to issues — I think Senator Miville-Dechêne raised them — regarding “deepfakes,” misinformation and disinformation. Senator Simons highlighted the quality of search returns.
These are all examples of things we need to be mindful of and watch for to ensure that the approach taken is one that actually provides robust coverage but also recognizes the power and potential of AI in certain applications to advance.
Senator Mohamed: Thank you.
Senator Arnold: Thank you for coming back. We appreciate it. We have heard so much testimony now, and it is about that balance between utopian hype versus cataclysmic fatalism and trying to find a mid-ground with so much happening right now. We had a witness here, Jaxson Khan from the Munk School and co-author of Sovereign by Design, and many of us heard Janice Stein talk recently about the goal of protecting against coercion. I’m curious how much that study might have influenced or impacted your work.
Runa Angus, Senior Director, Marketplace Framework Policy Branch, Innovation, Science and Economic Development Canada: I can talk a little bit about how the current legislation promotes digital sovereignty and control over our data. As a reminder, the Personal Information Protection and Electronic Documents Act, or PIPEDA, is about the collection, use and disclosure of personal information in commercial contexts. It is a law that has existed since 2001, so before AI but also before smartphones and social media.
One of the characteristics of this law, though, that makes it adaptable and applicable to even some of the threats that we face today that we maybe did not think about a few years or even a few months ago is that it is tech neutral and it is principles based. Those principles include having control of your information through consent, ensuring that information is used for appropriate purposes and really grounding the act in accountability so you are responsible when that data flows somewhere — to a third party, somewhere abroad and so on. Those are principles that apply in 2026 as much as they did in 2001.
Regarding more specific principles that apply, for example, to cross-border data flows, first, Canadian laws apply wherever there is a real and substantial connection to Canada. The fact that a company is abroad does not subtract it. We talk about Canadian companies or organizations in Canada being subject to foreign laws, but the truth is that foreign companies are also subject to Canadian laws the moment there is a real and substantial connection to Canada.
On top of that, there are very specific rules that require companies to ensure that protection that exists in Canada carries over — usually through contracts — when data flows to another company or organization and particularly across borders. So companies have a responsibility to ensure that the companies they are doing business with are able to provide the same protections that we would expect in Canada.
Further, companies are also required to safeguard information, and they can do that through a number of means. One is technological, and the law actually says that — you should undertake technological safeguards. Those include encryption. That’s another way you can safeguard Canadian data even if it is transiting through foreign companies or foreign networks and located abroad. Again, if it is only Canadian companies or Canadian individuals that have the encryption key, it really does not matter what a foreign law says; that data is not going to get out of Canadian control — again, being subject to Canadian law. So that’s on PIPEDA.
Mr. Chhabra: I will just add to that, senator. I think your question was a little bit broader. The data sovereignty piece is a critical pillar of this, but there is more to it. I think the conversation around digital sovereignty today is a lot more around having the ability to set our own agenda and direction and have control. The data piece that everyone has talked about around personal information is a critical element of that. There is also an element around building the infrastructure that we need here in Canada to be able to have that sense of control over that. It is also about ensuring that we have, throughout the technology stack, whether we are talking about the models, the data or the infrastructure layer and the hardware, that we have the ability to broadly define and set our own destiny. That does not mean isolationism or building every single product here in Canada. It means that, either through a combination of Canadian sourcing or strategic partnerships with like-minded jurisdictions, we have the ability to create a technology stack that works for us and is responsive to our needs, our direction and our values.
Senator Arnold: There is another great study from Gen(Z)AI, a youth assembly, and there are some really great recommendations in there.
Runa, you said, “This will say you are responsible.” When you say that, so much of what they are hoping for is real regulations and putting the responsibility back on the platforms. I wonder if you could speak to that, because that came up a lot. They are looking for more tangible responsibility from the platforms — safety by design, that concept.
Ms. Angus: I can take that one. As Samir said, the minister has publicly stated his intention to modernize PIPEDA. He has also said that it is going to be based on some of the work that we did in Bill C-27, which had a PIPEDA modernization proposal.
If we look at that piece of legislation, a lot of that was based on giving individuals more control over their information, but it also had much more prescriptive requirements for organizations in terms of what exactly they need to do to protect people’s information. Of course, layered on top of that was a robust enforcement framework to buttress all of that. That is something that we will look at again as we think about modernizing PIPEDA.
Senator Arnold: Do we have a date on that?
The Chair: All right. Thank you, Senator Arnold.
[Translation]
Senator Aucoin: I have two questions for our witnesses.
First, we have heard about all the possibilities and everything the government is considering, but one issue has not been discussed much: protecting the development of the French language and francophone identity in Canada. I would like you to tell us what the government has done regarding this issue. How does AI . . . . Have you talked about this? In a few years, we could find ourselves facing consequences we had not anticipated, which could greatly affect the future of the francophone community in Canada.
Mr. Lauzier: That is an excellent question, because I think it demonstrates that this is not just about technology but also about the data used to develop it; all this is a reflection of who we are as a country and as francophone and anglophone communities.
It comes down to sovereignty and the infrastructure we were discussing a few minutes ago. Developing technology that reflects our values is part of all of this.
I think we need to look at it from both sides. Indeed, AI can help keep many languages alive. I understand your example regarding French. However, I would like to quickly give you an example involving certain First Nations languages. The National Research Council launched a project to use AI to help translate less widely spoken languages. We also have companies like Cohere, which have developed very good multilingual generative models.
To get back more specifically to your question, I believe it is up to the government as a whole to determine how we will support creators and communities so that they can generate their own data and make their content accessible online, because ultimately, the models will use online data. They cannot be trained on data that is not accessible or not available online in a specific language. Consequently, this type of data will be less accurately reflected in the final model.
[English]
Mr. Chhabra: I will add that the question is relevant on so many different levels. There is a question there around culture and representation, but there is equally a question around safety and robustness. That’s another area where we tend to see this happening a lot: How robust are the safeguards on any of these models across a variety of languages? How robust are they in languages that are not often spoken? How consistent are their responses in the Canadian context, in English and French?
Part of the work we are doing through the International Network of AI Safety Institutes is this multilingual evaluation and testing work. Part of what we need to ensure is that there is not a situation where a language that is spoken in a minority community or a language that is little spoken or understood in the world — there is a cultural question around whether it ought to be included and represented. That’s a question for the speakers of that language themselves, to determine how they want to be represented.
The other piece is ensuring that, no matter what, if a service is being provided in a certain language, the service is being provided with an equivalent level of quality and safety.
[Translation]
Senator Aucoin: All the witnesses we heard from spoke to us about the benefits of AI. There are many, even in the field of translation. However, there could also be several more subtle negative effects regarding the culture and unique characteristics of the francophone, Quebec, and Acadian communities. No one is really talking about this issue. I have another question to ask, but first, would any witness like to comment on this? I understand that no one does.
My other question concerns the News Media Canada document we received yesterday. It lists all the things the government should be addressing. It discusses the consent of publishers, artists, authors and musicians. In particular, it mentions the need to cite sources, ensure transparency, provide compensation and require consent. Is the government considering legislation in this area? What can you tell us about this?
Mr. Chhabra: It is an important question that is asked frequently.
[English]
I think there are 16 cases right now that have been brought in front of courts on this issue, primarily by Canadian media publishers. There are a significant number of cases in the United States, as well. We’re starting to see some judgments emerging, particularly from the United States — none have come from Canada so far — that actually indicate, even within the same circuit in certain cases, differential judgments and interpretations.
So, there is a really interesting question right now around what the exact state of Canadian law is. As I said earlier, we have heard from a number of stakeholders in the creative community who certainly press on the ART framework — authorization, remuneration and transparency — as you highlighted. Many of them have also highlighted that they would prefer to see no changes to the Copyright Act as it stands because they believe that it currently provides them an avenue to seek the licences they are seeking to pursue.
I think that question is still to be determined as the courts work through these issues. It could take quite some time to do so. For the moment, the preponderance of the creative community in Canada has highlighted that they would prefer that the government not act and give some time for the courts to review and assess.
[Translation]
Senator Cormier: My questions are supplementary to those asked by my colleague.
It is well known that one of the challenges of AI is the issue of protecting the country’s cultural sovereignty. The Coalition for the Diversity of Cultural Expressions and several artists have made it very clear that authorization, compensation and transparency are essential in the use of works that will be used to train AI. You touched on this briefly, but what are the specific legal challenges that AI poses for the protection of freedom of expression and artistic creation? My question addresses a somewhat more specific topic than the general issues.
Mr. Chhabra: It would be easier for me to answer your more technical question in English, if that is acceptable.
Senator Cormier: Go ahead.
Mr. Chhabra: Thank you.
[English]
It’s a fascinating question. Copyright laws are very old; they date —
Senator Cormier: I agree. But the Coalition for the Diversity of Cultural Expressions, or CDCE, is saying that we should not change the Copyright Act, so that’s clear for me.
[Translation]
Mr. Chhabra: Thank you, senator.
[English]
With the way copyright acts tend to be developed — and Canada’s is similar to this — it is about the making of a copy. One of the technical questions that is underpinning the deliberations the courts are going through now is this: Is a copy being made, reviewed and utilized by the machine in its learning process and then referred to again? There is a series of questions that need to be unpacked from a technological perspective as to what is happening.
There is a case in the United States where Anthropic had a significant judgment against it, but the issue in that case — the way the court decided it — was the piracy of the materials that they had used in the first instance. They had stolen the materials, but that’s a different part of the regime; that is not specific to whether they trained an AI system.
We’re seeing the adjudication of a number of related and similar issues, and we have to watch the court cases extraordinarily carefully and read the judgments carefully to understand exactly where liability was assigned and for what purpose. Some of the headlines we see coming out on this say that an AI company was found liable for copyright infringement. Sure, but what aspect of copyright law were they found to be infringing?
How do you get the materials in the first place? Did you pay for them? Then, there is a question about what you are allowed to do with those materials once you have them in your possession and whether you are allowed to use them for training. This is where a number of the stakeholders have highlighted the framework around ART as being critical.
In Canada and other jurisdictions, we are seeing a question around how creators know if their output — their work — was used to train AI. In Europe, there is an approach to doing this, which requires AI companies to publish summaries of the training data that they have used. There is an opt-out framework. We understand that there have been some challenges on both those fronts in terms of companies being able to develop summaries that are detailed enough and usable for people to actually understand what has gone into it, and, on the flip side, some of the training can make it very challenging to be very specific about products, works, articles or books that were ingested.
So, the creation of the summary itself can pose challenges, and the opt-out framework itself is also facing some challenges.
It is not clear that there is a gold standard that we could look to in terms of being a jurisdiction that has this sorted out. A number of jurisdictions are working through this, and we’re doing so in collaboration, to some degree, by sharing knowledge and understanding of what is possible, technically, and what is feasible within different legal frameworks to better understand what is emerging and where this conversation will go.
But, for the moment, what we tend to hear from creators is an emphasis on maintaining the integrity of the Copyright Act as it is. If there is another ask, it is this: Can the government help to create more transparency around how the models themselves are trained? That would help enable the exercise of rights that are encoded in the Copyright Act.
[Translation]
Senator Cormier: Thank you. Is the Canadian data protection framework sufficient to regulate the use of cultural works in training AI models?
Ms. Angus: Could you please repeat the question?
Senator Cormier: Yes. Is the current Canadian data protection framework sufficient to regulate the use of cultural works in the context of what we’ve just discussed?
Ms. Angus: I will begin by discussing the Personal Information Protection and Electronic Documents Act, or PIPEDA, but if my colleagues wish to discuss other frameworks, they are welcome to do so.
This is the framework governing personal information. Personal information is used to train AI. There are two basic rules under PIPEDA.
The first is that data must be used for appropriate purposes. An appropriate purpose is what a person might consider reasonable under the circumstances; so it is somewhat context-dependent. If a company uses AI, it must be for an appropriate purpose. It cannot be for just any reason, so there really needs to be an appropriate purpose — which wouldn’t be profit, for example. The activity must be regulated and the purpose specified; this is consent-based legislation. Companies cannot use personal information unless they obtain consent, because there are very few exceptions to consent under the current legislation.
That said, as I mentioned earlier, this law is over 26 years old. So, it needs to be modernized, and the minister has already said he will do so. The idea would be to ensure that this legislation is now adapted to the digital age.
Senator Cormier: I have one final question. How can Canada — given that our country does not work in isolation but collaborates with the rest of the world on these issues — influence international AI standards to reflect its cultural and linguistic values? Earlier, we were discussing linguistic duality, the two official languages and Indigenous languages. How can we be a leader in promoting a use of AI that takes our cultural and linguistic values into account?
Mr. Lauzier: That is an excellent question. We have talked a lot about legislation today. There are several international standards that are not limited to legislation, and there are standards that the industry must follow — this is very important. It facilitates trade so that the industry can operate, but these standards also reflect our values, since the industry standards for AI are not the same in every country. That is why Canada has been active on this issue for several years, since the second phase of the pan-Canadian artificial intelligence strategy.
We fund the Standards Council of Canada with initiatives aimed at the entire ecosystem. They have a group working to advance two things: developing standards in Canada, but also internationally; there are groups like ISO/IEC 4200 — that is the best known, but there are others. People from different countries participate in these groups, and Canada is active in them as well. We ensure that this reflects our values and the interests of Canadian companies.
It is also very important, as we discussed earlier regarding the European Union, to ensure that our Canadian companies can export abroad. The entire framework of European law will be implemented through standards currently being developed by CEN-CENELEC, a European organization that deals with —
Senator Cormier: Does this happen all the time? Does this conversation always include a discussion of culture, language, and our values? There are many agreements with other countries, but is Canada showing leadership in this regard? That is the point of my question, actually.
Mr. Lauzier: When it comes to standards, it depends. Some standards are very technical. For example, for synthetic content, culture is the most important issue. Standards, I believe, are an example of this, and Canada has been very active in the Global Partnership on Artificial Intelligence.
As for culture, as you probably saw, Minister Solomon and Minister Miller attended a summit in Banff that brought the entire community together.
The question is how to bring these ideas to the international level. For example, UNESCO is examining cultural issues related to AI. Our colleagues at Canadian Heritage are handling this file, but Canada is very actively engaged at UNESCO and in other forums.
Senator Cormier: Thank you.
[English]
The Chair: I have a couple of questions. The whole question of the pipeline of information is international now. We may perform a search on our laptop or phone, but that information may be generated outside Canadian borders, and, obviously, that’s often the case. How much control can Canada have on policy, on regulations or even on laws when most of our interactions now are borderless?
Mr. Chhabra: Thank you for the question, senator. It is an important one. Canada, along with other jurisdictions, maintains a significant amount of control over what happens to information within its jurisdiction. As my colleague Runa pointed out, the Personal Information Protection and Electronic Documents Act, or PIPEDA, continues to apply on a personal information basis. There are opportunities for Canadians to feel protected and covered by PIPEDA. Also, as we highlighted, there is an important rationale to modernize PIPEDA as well, to strengthen the enforcement framework, to put more emphasis on children’s rights and to give Canadians better, more usable and understandable control over their personal information and more transparency about how it is used.
There are frameworks that Canada already has and that can be modernized, and there are approaches that we can use — including through investing in Canadian infrastructure and hardware — to better ensure that we have more control over the data that is resident in Canada as well. There are approaches to be taken both on the frameworks perspective as well as the infrastructure perspective to help buttress sovereignty.
The Chair: Thanks for that. With that, there are seven, eight or more artificial intelligence platforms that all operate within Canada, and they also operate elsewhere. How is that guardrail put up just to affect or protect Canadians? When I do a search on whatever platform I decide to use, does that go through a filter where you are in Canada and can or cannot access that?
Mr. Chhabra: Thanks very much, senator, for the question. You are absolutely right that companies have the ability to tailor experiences on a jurisdiction-by-jurisdiction basis. This is possible. There are differential guardrails that can be applied, and there are different rule sets and norms that can be respected in different jurisdictions.
Certainly, we would expect any company operating in Canada today to respect the rules around personal information protection that are encoded in PIPEDA, the Copyright Act and other frameworks that we have talked about today. So, absolutely, that’s possible, and we expect it. That’s part of the rationale for thinking about how to modernize the frameworks to ensure they are effective in this new age. It is because we certainly expect that they are respected and ought to be effective in this context.
The Chair: Thanks for that. My next question is related to data centres. Tell me if this is not part of your remit so we do not have to spend time on it.
Is there currently competition for electricity in Canada between the data centres that are being developed and our own domestic electricity needs?
Mr. Chhabra: Thanks, senator. I will start out by saying that, because I am on the marketplace framework policy, I’m not leading on the data centre end of this thing. What I can say is, from the conversations and engagements I have had, this is a very seriously thought-through question.
It has been well considered in all aspects, for example, in sovereign compute work, Canada’s AI infrastructure and building out the sovereign compute strategy. These are important questions in terms of the impact of bringing any kind of data centre investment on the environment, on local ratepayers and on infrastructure. It’s part of the consideration that’s being thought through, but it’s certainly a question that we can take back to the department for more details.
The Chair: Is there a section within the department or another department that would deal with that?
Mr. Chhabra: That’s right, and we’re happy to connect the dots —
The Chair: Sorry. Is there a department or section within your department that deals with infrastructure capacity?
Mr. Chhabra: That’s right. Within ISED, there are two units, Innovation Canada and Industry Sector, that are heavily engaged on these issues. I’m happy to connect you if there is a question the committee would like addressed. I’ll bring back a response.
The Chair: Thank you, witnesses. Colleagues, we’ve come to the end of our time on this.
(The committee continued in camera.)