THE STANDING SENATE COMMITTEE ON TRANSPORT AND COMMUNICATIONS
EVIDENCE
OTTAWA, Tuesday, April 14, 2026
The Standing Senate Committee on Transport and Communications met with videoconference this day at 9 a.m. [ET] to examine and report on the opportunities and challenges of artificial intelligence (AI) in the information and communications technology sector.
Senator Larry W. Smith (Chair) in the chair.
[Editor’s Note: Some American Sign Language passages were presented through an interpreter.]
[French]
The Chair: Honourable senators, welcome to this meeting of the Standing Senate Committee on Transport and Communications. Thank you for your cooperation.
My name is Larry Smith. I am a senator from Quebec and chair of the committee. I would like to ask my colleagues to introduce themselves.
Senator Simons: Paula Simons, Alberta. I come from Treaty 6 territory.
Senator Wilson: Duncan Wilson, British Columbia.
[Translation]
Senator Quinn: I am Jim Quinn from New Brunswick.
[English]
Senator Mohamed: Good morning. Farah Mohamed, Ontario.
[Translation]
Senator Cormier: I am René Cormier from New Brunswick.
[English]
Senator Lewis: Good morning. Todd Lewis, Saskatchewan.
[Translation]
Senator Arnold: Good morning. I am Dawn Arnold from New Brunswick.
Senator Aucoin: I am Réjean Aucoin from Nova Scotia.
Senator Miville-Dechêne: I am Julie Miville-Dechêne from Quebec.
[English]
Senator Dasko: Donna Dasko, Ontario.
The Chair: Thank you, colleagues. I would like to welcome everyone with us today, as well as those listening to us online on the Senate’s website, sencanada.ca.
We are meeting today to continue our study on the opportunities and challenges of artificial intelligence, or AI, in the information and communication technology sector.
With that, I would like to now introduce our first panel. Our first invited individuals are Jeffrey Beatty, Chairperson, Deaf Wireless Canada Committee; and Nathan Sanders, Researcher at the Berkman Klein Center for Internet and Society at Harvard University. Nathan is with us on a virtual basis today. Thank you both for joining us today.
Witnesses will provide opening remarks of approximately five minutes, which will be followed by a question-and-answer session with senators.
I will now invite Mr. Beatty to give his opening remarks.
[Interpretation]
Jeffrey Beatty, Chairperson, Deaf Wireless Canada Committee: Thank you, Mr. Chair and honourable senators. My name is Jeffrey Beatty, and I am Chairperson of the Deaf Wireless Canada Committee, or DWCC. DWCC advocates for telecommunications and digital accessibility for deaf, deaf-blind and hard-of-hearing Canadians. Our work sits at the intersection of communications policy, consumer rights, public safety and accessibility. We advocate for approximately 4 million Canadians who are deaf, deaf-blind or hard of hearing across Canada, including an estimated 150,000 to 250,000 sign language users.
As this committee studies artificial intelligence in the information and communication technology sector, I want to emphasize one central point: AI policy will be incomplete if accessibility is treated as secondary rather than as foundational.
For our communities, AI is not just about innovation. It affects whether we can communicate safely across public services, rely on captions and relay functions, participate in digital life and trust the systems that increasingly mediate communication.
Today, much of the national AI discussion is framed around growth, trust, sovereignty and adoption. Those are important goals, but AI in communications cannot be truly trustworthy or truly for all if it is built on speech-first assumptions and only later adjusted for accessibility.
That creates real risks. Systems may perform poorly for sign language users, produce inaccurate captions, overlook non-audio communication needs and fail to reflect the lived realities of deaf, deaf-blind and hard-of-hearing people. Accessibility is too often treated as a feature or accommodation rather than as a core design and governance requirement.
Later this week, I will be attending an international deaf-led summit focused specifically on sign language and AI. That matters because it shows that deaf communities are not waiting on the sidelines of this discussion. Deaf-led expertise already exists on ethics, data sets, translation design and governance. Public policy should recognize that expertise and build it in from the start.
Canada has an opportunity to lead differently. Success in this Senate context would mean AI systems in communications that deaf, deaf-blind and hard-of-hearing Canadians can actually rely on because accessibility expertise, testing, procurement and governance were built into their design from the outset.
First, accessibility must be built into AI governance and information and communication technology, or ITC, policy from the beginning. Second, AI systems used in communications, customer service, public information and emergency-related environments should be subject to accessibility testing and audit requirements. Third, deaf-led and disability-inclusive validation must be part of procurement, standard setting and oversight. The communities affected by these systems must help define whether they are actually safe, accurate and usable.
Accessible AI is not a feature request. It is a communications right and public accountability requirement. If Canada wants trustworthy AI in communications, then accessibility cannot remain a footnote. It must be part of the core public interest framework, from design to deployment to accountability.
Thank you. I look forward to your questions.
[English]
The Chair: Thank you, Mr. Beatty. I will now invite Mr. Sanders to give his opening remarks.
Nathan Sanders, Researcher, Berkman Klein Center for Internet and Society, Harvard University, as an individual: Thank you very much. I greatly appreciate the opportunity to speak to this committee. It is a privilege to follow the powerful remarks of Mr. Beatty.
I’m Nathan Sanders. I am a data scientist based at Harvard University here in Massachusetts. I am the co-author with Bruce Schneier of the book Rewiring Democracy: How AI Will Transform Our Politics, Government, and Citizenship.
Having done so much working looking around the world at examples of how societies and governments are responding to AI, I applaud this committee’s study. Canada is well positioned to be a global leader in applying AI to the public interest, and I hope the work of this committee will galvanize Canada in that leadership role.
Canada should invest purposefully in domestic AI to pursue two goals: first, to develop sovereign AI capacity for human communication by fortifying Canada’s ICT sector with civic technology investment; and, second, to position Canada as an international leader on AI governance by shaping the AI ecosystem to cooperatively serve democratic publics.
For the first purpose of developing sovereign AI capacity for human communication, I think Canada can achieve this by investing in its civic technology ecosystem. Canada has significant assets in this space. It has three national AI institutes. It has the Canadian Sovereign AI Compute Strategy, deliberative democracy expertise at universities like the University of Toronto and McGill and homegrown civic technology platforms like Vancouver-based Ethelo. The opportunity is to make a connective investment that turns these into a coherent public interest AI ecosystem.
Deliberative technologies increasingly use AI to help people make decisions together. Innovations for processing human-generated content and speech in commercial settings can serve democratic use cases, and investments in either can serve both.
These tools can do more than summarize public input at scale; they ensure diverse viewpoint representation and help historically marginalized individuals and communities participate. For example, the deliberation.io platform uses AI to elicit participant opinions through Socratic dialogue, then facilitates social review and voting on those ideas. Similarly, Harvard researchers developed generative social choice, enabling groups to vote on complex issues by using AI to synthesize diverse inputs into a representative set of proposals for voting.
Because AI-assisted coding is accelerating the development of these types of capabilities, even for small teams, even modest grants or government procurement incentives for researchers and civil society organizations can generate a massive impact.
For the second purpose of positioning Canada as an international leader on AI for the public interest, I think Canada can achieve this by orienting AI investments toward open science and interoperable public goods. Open-source tools developed by Canadian civil society in cooperation with international groups will be fundamentally more trustworthy than proprietary corporate technologies from the U.S. or elsewhere. They reduce vendor lock-in by enabling public auditing and creating shared public assets instead of capturing value for foreign investors.
There are numerous models for international cooperation on open-source, interoperable, AI-powered communication tools already. The Scottish government has funded a domestic non-profit called CrownShy to develop an integrated, AI-supported legislative participation platform. In Japan, the legislature has distributed public funding to Team Mirai to develop open-source constituent engagement tools. And Decidim is an open-source framework for digital democracy born in Spain and already used around the world, including in Canada.
There is a robust Canadian civic tech ecosystem that can build cooperatively with international partners like these and thereby leverage global innovations, localize them to Canada, Canadian bilingual and federal-provincial requirements and, at the same time, contribute to the greater democratic good.
Canada should also invest in public alternatives to corporate AI and big tech. Switzerland has demonstrated that mid-size countries can develop publicly funded foundation models. Apertus is an AI model built by Swiss universities on publicly owned data centres, using renewable energy, with appropriately licensed data and released for free as a public good. It’s not jockeying with OpenAI’s latest models for the highest levels of performance. Instead, it provides a compelling alternative for users that prioritizes ethical consumption, sustainable AI and regulatory compliance. With its national AI institutes and emerging sovereign compute infrastructure, Canada has the capacity to do the same.
In summary, Canada can lead globally in applying AI to the public interest by, first, developing sovereign AI capacity for human communication through a flourishing civic tech ecosystem; and, second, by shaping the AI ecosystem to serve democratic publics by international cooperation and building public AI infrastructure.
Thank you again for including me today. I welcome any questions.
The Chair: Thank you very much, Mr. Sanders.
We will now move to questions from senators. I would like to advise senators that you will have approximately five minutes each for the first round — and a second round 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 will start by inviting our deputy chair, Senator Dasko, to ask the first question.
Senator Dasko: Thank you to our witnesses today. Mr. Beatty, thank you especially for being here with us. You have said that we should build accessibility in from the start — that it must be part of core design and governance. I would like to drill down a little to learn how that happens. How does it come about? Who is responsible? Is it government that needs to set the rules or is it the AI systems themselves? What is their responsibility? Or is it the responsibility of the government to put in place rules and regulations that deal with the systems?
I would like to understand that a little better. Thank you.
[Interpretation]
Mr. Beatty: Thank you for the question, senator. Success would mean an AI system for people who are deaf, deaf-blind and hard of hearing that they can rely on. We need expertise to be built into the design and to have it tested during procurement and then during use. The government has a responsibility for any communication that is public facing, and it also a role in accessibility testing during deployment. Deaf experts and deaf‑led expertise should be involved with the innovation and design of AI systems to ensure they are reliable and usable. I envision that Canada has sovereign accessibility and that we are creating it, so we are not doing it as an afterthought but have sovereign accessibility.
We have several deaf, deaf-blind and hard-of-hearing experts in linguistics, research and academics. There are already position papers on AI, and those are available here and now in Canada. As we are developing AI systems to include that expertise on accessibility for our community, that will help to have sovereign accessibility and make sure we have an advisory committee who is also available to make sure that information is available across Canada.
[English]
Senator Dasko: Thank you. I have another question. In terms of existing AI systems, how are they doing? Have any of them built this in, and to what extent have they succeeded in achieving this?
[Interpretation]
Mr. Beatty: That is a question that is very often overlooked in the current ecosystem. I have noticed that there are some gaps in development. I think our impact here at this committee is very worthwhile to make sure that we are being included and to have our expertise to make sure that system designs are deaf-led. Because we want to have an accurate data set and make sure our linguistic requirements, both in American Sign Language and Langue des signes québécoise, or LSQ, are available to users in Canada.
We want to make sure we are involved in the governance of AI creation because it has an impact on our use of sign language and communications. We are looking at that sovereignty within Canada.
[English]
Senator Dasko: Thank you.
Senator Lewis: Thank you both for your comments so far today. Mr. Sanders, you spoke about some of the international players involved. Of course, in so many ways, it is going to be led by actors outside of our country. That’s one way of putting it. As we see this developing so quickly, for a country like Canada, if we don’t get involved at the public level and so on, are we going to miss the opportunity? How soon until we miss that opportunity if we don’t get on board and start down this road?
Mr. Sanders: There are risks to not engaging with AI development, and I think they are on at least two sides. One is on the opportunity to leverage AI responsibly and, by the way, also to mitigate risks on the application of AI in the public sector and in society. The other is on missing out on economic opportunity and value capture that is being created by AI. Today it is being captured, particularly by a small number of companies in places like Silicon Valley.
On the former, I was interested in sharing the example of the investment that the Scottish government has made through their domestic non-profit CrownShy because I think it is an example of a legislature that recognizes that they can improve their engagement with the public, improve their participation and policy making and improve their responsiveness to their own community by leveraging these types of deliberative technologies in their process. They can create value for governance by leveraging the technology and can steer the development of that technology by being actively engaged in the ecosystem by directing their civic technology actors within their country with explicit guidance and by provisioning tools that meet their needs. And they are doing so in a way that is internationally cooperative and is being built in an open-source and interoperable manner that is meant for others to use as well.
I think that is a really exciting model for how democracies around the world, and particularly mid-size countries, can leverage and steer the development of the technology and really band together to make sure that their particular national interests, requirements and sovereignty are maintained in the face of much larger countries that may otherwise have monopolistic power.
Similarly, with respect to value capture, I think the biggest existential risk for democracy that technology is creating today is the fact that a very small number of corporations have grown to multi-trillion-dollar valuations on the basis of investor interest in AI. I think that concentration of wealth and power is a global existential risk, and the only way to put ourselves on a different course is to find alternative methods of capturing and distributing the value created by this new technology. The technology is clearly creating value. I don’t think there is a world where we don’t build and leverage these tools throughout government and society. However, I think it is possible to capture that value in public institutions and distribute that value to the public instead of a small number of very large private corporations. There are examples of that type of public interest, publicly owned and open-source AI development around the world today.
I shared the example of Switzerland because I think it is the most mature example, where, as part of a national AI strategy, a government has invested at the foundational level of building a so-called foundation model as a public good according to their own ethical principles, including environmental sustainability and adherence to EU regulatory requirements. Again, they have provided that as a public good for anyone to use and as a basis for international cooperation.
I am eager to see other governments around the world join in that kind of collaborative effort and invest in that kind of public AI infrastructure.
Senator Lewis: Thank you. I cede the rest of my time to other members.
Senator Wilson: My question is for Mr. Sanders. Just thinking about this civic or public infrastructure approach, how do we compete with what the private sector is doing? I don’t mean in head-to-head commercial terms; I mean more in terms of rapid development and deployment. I’m worried that by the time we get some of the collaboration you are talking about organized, other systems will be so far ahead that we will not be able to keep up. Is that a problem? Is that a solvable situation?
Mr. Sanders: I think that is a great and important question. I would call out that there is an opportunity to compete on different terms than commercial, big tech AI is doing today. I think there is a huge gap in the marketplace that big tech is not filling and is likely to never choose to fill, and that is around the development of trustworthy and ethical AI.
Today, the biggest technology companies in the world are competing mostly for incremental gains in performance. Those of us who follow this field closely are used to seeing new announcements from OpenAI or Anthropic that claim subtle improvements in performance that are very difficult to measure. They are getting just incrementally better at solving tasks related to coding language translation, et cetera, which is valuable and advances the field but for most users it is not truly noticeable except as it compounds over many cycles of releases and competition from these players.
However, neither of these companies is really investing in a fundamentally more trustworthy basis for AI. They do not involve democratic publics in the governance of their models. They do not involve governments or the public in making decisions about how they should be built. They are not taking steps to make sure that the data they are training on is ethically gathered and appropriately licensed. They are not giving control over the future development of those technologies to anyone other than their investors.
So there is a marketplace gap that is not being filled today by the private sector that government can uniquely serve by providing fundamentally more trustworthy AI. It can take time and investment to set that up. I think the example of Switzerland that I shared shows us that it is possible to do that. It is not necessary to compete on a time scale where, as new models are released from the private sector every week, we are trying to outperform them with public models. Rather, we can show that there is a fundamentally alternative path that will be more appropriate and more attractive for many use cases, including those in civil society and the public sector.
I think it is okay if the performance of those models never races to compete with the very best from the private sector as long as they achieve those goals of being under public control, capturing value for the public and being fundamentally more trustworthy. I believe that is very achievable.
Senator Wilson: Thank you.
Senator Simons: My first question will be for Mr. Beatty. I wonder if you could give us some practical examples of the way people from the deaf community make use of AI so that we can understand why this is so important. For example, I produce a monthly podcast. At first, I was delighted when Apple started producing an almost instantaneous transcript. I thought it was great — that people who cannot listen to the podcast could read it.
Then I read the transcript, and a lot of times I was using vocabulary that it did not understand. The robot is doing its best, but it was not a great transcript. If somebody were just relying on that, they would have an inaccurate sense of what actually was said.
I’m sure, at first, members of your community thought, “This is fantastic. Everything will be transcribed.” But the transcription is often unreliable. Could you perhaps give us some practical examples of the pros and cons of AI for your community? I think we would have a better sense of how you want it to work if we can understand how people are using it now.
[Interpretation]
Mr. Beatty: Thank you, senator. With this analogy, you are perfectly right when you are looking at a spoken-word transcript and then something that is signed. We do want to access the transcripts, but you are right that it is not there yet. And it is the same for sign language; the accuracy is not there yet. We are wondering who the audience is and who is checking to see if the signs are right and what is being produced. The Accessible Canada Act recognizes the use of sign language as the primary language for deaf and hard-of-hearing people, for both ASL and LSQ and for Indigenous sign languages as well. Those are the primary languages used here in Canada.
So if you have AI signing, who is looking to see that it is accurate? Who is governing? Who is monitoring what is being produced? It is not there yet. That is why we are saying that we need to talk about accessibility from the onset and not as an afterthought.
At least when you refer to your own podcast and read your own transcript, you know it’s not accurate, but for deaf people and signing, there is not the same luxury or the same access to that communication. It is on parallel with exactly what you were saying, AI and sign language. The data sets right now — there are thousands of minute facial movements, which have grammatical features in how one person may vary in their signing style from another.
This week, there is a sign language summit on AI. There will be 200 delegates who are leading expertise in AI. They are going to talk about exactly what we are talking about right now on sign language and the use of AI. I am really looking forward to gathering more information, which I would be happy to share with you. It is an international summit so we have delegates from all over the world.
There are linguistic and technical experts coming together for the first time this Thursday and Friday. I will have a meeting with them to talk about sign language. Thank you very much for your question.
[English]
Senator Simons: I have a quick question for Mr. Sanders. When we’re talking about setting up a regime that would be more sovereign, can you speak to what the role of copyright law would be in that so that large language models, which steal information and regurgitate it — are there things Canada needs to do from a copyright law or governance perspective to set us up for the kind of sovereignty you’re talking about?
Mr. Sanders: Yes, I think that’s an important consideration as well. I would note that there are two important ways that information is used and is necessary for the development and use of AI models. One is during the initial training of those models, where often enormous amounts of data, trillions of tokens — or think of it as millions or billions of documents and articles — are used to train a foundation model. But it’s also during the use of those models, when a user brings a specific question or task to an AI model that requires specific and often up-to-date information, such as current news reporting or other data sources, to be used by the AI during the performance of those tasks.
Respect for copyright and supporting industries globally — but also, particularly from Canada’s perspective, domestic industries, including publishing — during both training and use examples is very important. This is an area where big tech companies have not been playing by the rules, at least according to my understanding or the traditional understanding of the rules, in countries around the world.
I think making sure to clarify how existing copyright law within countries and internationally is intended to apply to AI models, so that there are not loopholes created by questions about the use of new technologies with traditional content, is very important in this context.
Establishing new mechanisms that allow content creators and copyright holders to be paid and compensated when their content is used either during training or use is very important.
There is real value created for users and for society when AI has access to relevant and grounding information. I think we want to find a way to allow that but not to sacrifice the creative and intellectual work that goes into producing content. These two things must co-exist. I think copyright is an important mechanism for that, and it starts with clarifying how the existing intention of existing law should apply to AI.
[Translation]
Senator Cormier: My first question is for Mr. Beatty.
[English]
In a context where AI systems are increasingly integrated into telecommunication services, do you believe the current legislative framework, particularly regarding accessibility and provider accountability, is sufficient to protect the rights of deaf or hard-of-hearing persons or should obligations specifically related to AI be explicitly incorporated into federal law? I’m thinking particularly about the Accessible Canada Act.
What would be your recommendation? Should we integrate something directly into those laws about the use of AI for your community?
[Interpretation]
Mr. Beatty: Currently, the CRTC regulates telecommunications and broadcasts. They have their own act, which says these should include voice and video. Video is where something becomes accessible for the deaf and hard-of-hearing community. The CRTC needs to have its article 23 that talks about accessibility. As AI is a new technology, that law becomes more consequential because we want to have deaf-led innovation. We want to have advisory input on accessibility and the technology, and if we collaborate with the federal government through that, then we would be protected under the law.
I see it more under policy regulation as the governance model because many innovations right now, including AI, overlook the deaf and hard-of-hearing communities. I am very grateful for the opportunity to be a witness here, to talk about a potential governance model and what can be established, just to make sure that our sovereignty to accessibility is available.
The Accessible Canada Act does recognize ASL, LSQ and Indigenous sign languages as our primary languages of communication. That does need protection under the law. That should include AI procurement, ethics and standards. As Mr. Sanders already discussed, we need to be included in any decisions being made. We would be the prime drivers of any kind of accessible communication.
As I am attending this summit at the end of this week, I’m starting to realize that we do need a governance model, and that would be my recommendation. I completely agree that to protect the use of sign language, it needs some kind of oversight to make sure that it is ethical, reliable and incorporates well into any new technologies and to protect sovereignty across Canada.
[English]
Senator Cormier: Thank you.
Senator Mohamed: First, thank you very much, Mr. Beatty, for joining us. I have to admit, in all the time that I have thought about AI, I haven’t really thought about this angle. It’s incredibly important that you have opened our eyes to some of the challenges and some of the opportunities, quite frankly, for Canada to lead.
I have a very simple question: Have you had the opportunity to meet with the minister or the department as they are putting together their strategy that will lead to investments?
Recently, we have heard about “AI for all.” How do we hold that up as a banner? I’m worried that if we don’t really approach, as you say, inclusion from the beginning, trying to add things on at the end will become very difficult. So I wonder if you might reflect on your engagement with the minister and the department.
I can’t speak for my colleagues, but thank you again for opening up my eyes to something that I really had not been connecting the dots on.
[Interpretation]
Mr. Beatty: I have met with the minister and with their staff. It’s interesting; they had the same response. They had never considered it before either. Many were taken aback because it is so easy just to fall down that rabbit hole of AI innovation and never think about the accessibility lens when it comes to deaf, deaf-blind and hard-of-hearing Canadians. When we saw that, we knew our concern was maintaining that sovereignty over accessibility because that lens did not seem to be present.
We are advocating now so it is not left as an afterthought, because if it is, it becomes far more complicated and impossible to cover. Thank you very much for your question.
[English]
Senator Mohamed: Have you met with Minister Solomon in particular? He is holding the pen on the strategy and comes before the Senate today in a Committee of the Whole.
[Interpretation]
Mr. Beatty: I heard him speak at the Liberal National Conference, and I agree with his message of “AI for all.” I completely agree, but to emphasize, if he means “for all,” it must mean “for all.” We need to be on a level playing field, and that’s the key point.
[English]
Senator Arnold: Thank you. I really appreciate both of you being here today. It’s a real honour to hear both of your perspectives. My question is for Mr. Sanders.
First, I would like to thank you for your positive and upbeat work around democracy. Particularly, I loved your article in The Guardian about the positive aspects of AI and democracy. It’s really inspiring people. In my little community, there is an upcoming municipal election. People have used AI algorithms to put together a robust voting site where people can see what past records are and that sort of thing, so I think it can be really great for democracy.
My question builds on Senator Wilson’s around public infrastructure and how we feel all the time that we must get going and make this happen. It’s a simple question, I think: Do we have to start from scratch to do this? Can you take the systems and just reteach them?
Mr. Sanders: It’s a great question. I appreciate your commenting on the beneficial aspects the technology can bring. This is how I interpret your question. When we put this into the hands of members of our community who are looking to do good with that technology, who are looking to help us all work together and do better as members of a democracy, as well as to make government processes, including elections, work better, I see that potential and value.
When it comes to the question of whether we have to start from scratch, the way I see it is that we don’t because our community is already filled with examples of individuals, civil society organizations, governments and government agencies that already have capacity, expertise and people who are looking to do good. To me, it’s really a matter of putting them in a position to succeed and arming them with resources — including technology, access and funding — to help them do good work.
I think one of the things that AI is changing is the speed, scale and scope at which individuals, small organizations and communities that have been historically underserved can act.
We see examples of remarkable civic technology projects around the world using AI to do more faster. I’ll share an example from a project that I’m a part of here in Massachusetts. We built a legislative communication platform we call MAPLE, which is the Massachusetts Platform for Legislative Engagement. A number of us who are volunteers — mostly academics and researchers from universities around the Boston area, including Harvard and others — have come together to build an open-source technology platform that makes it easier for constituents to submit testimony to our state legislature. We have been doing that work together for about six years now. The pace at which we can build, support and maintain that platform has been increasing as we’re able to leverage AI coding and software development assistance in the work that we do. We are able to serve use cases we were not able to serve before. We can make the platform more accessible on the basis of language in particular by leveraging AI technologies that help us do more with the same amount of resources — a small volunteer team.
I know that there are remarkable projects like that and remarkable groups across Canada that are poised to put the technology to good use. I think what the Canadian federal government can do is make sure to support and foster a thriving technology ecosystem, which includes not only funding but also creating conditions for them to succeed. This includes prioritizing the use of domestic innovations like that and giving them alternatives to big tech technologies, particularly American AI products, that they might otherwise have to rely on to produce products like that.
I know I have mentioned several times open-source and public good AI models, such as the Swiss Apertus model. I think that’s an example of an emerging category of AI models that are more appropriate for civil use cases like that. I hope Canada can help support the development of more models like that.
Senator Arnold: Thank you.
Senator Aucoin: Good morning.
Mr. Sanders, you spoke briefly about ethics in artificial intelligence. Can you elaborate on that? Primarily, what should we as a group look at or what should the government be aware of?
Mr. Sanders: Thank you. I’m happy to respond. I think ethics is an important and frankly immature topic in the AI industry — immature in the sense that there are not yet strong professional standards as there are in so many other industrial domains to give not only practitioners, developers and companies but also the industry as a whole guidance on how to develop AI models that are aligned to societal ethical standards.
Government direction of what is appropriate ethically — both through formal regulation and law, and also by interventions that help steer the ecosystem, such as directing and supporting civic technology investments — is how to build those ethics.
Maybe I can give you a couple examples of where ethics show up in the development of AI models to illustrate that point. One we have spoken about already is that there are choices in selecting training data that are critical and foundational to AI model development — what information you show to an AI model when developing it. There are ethical choices about whether you will violate copyright law or wade into a grey area to use content that is not clearly licensed for your use or may not be licensed across international boundaries for use.
Having stronger guidelines to prevent the inappropriate usage of content and training is critical. But there are also decisions beyond what data to use that are involved in AI models. There are decisions in tuning and guiding the development of the models to make sure that, for example, it’s representative of a particular community’s viewpoint or that it is instructed to follow certain ethical principles or guidelines for the use of the model. In the past, that was a highly technical process. Today, that instruction is largely given to these models in natural language.
If you haven’t seen it, for example, some of the “constitution,” so to speak — a sort of brand name that Anthropic uses — that they have given to their Claude models has been made public in many forums. You can see the choices that the company Anthropic is making in how to instruct its model to behave when it’s being used by users.
This is a long and detailed document. There was a recent New Yorker article that went in-depth on Anthropic’s process in developing it. I think it’s appropriate and interesting that companies are putting that much thought into the instruction and guidance that govern their models’ behaviours with individuals. It shouldn’t be left only to individual executives at companies to make those decisions and guide those ethical principles. There should be public cooperation and control in guiding those ethical principles. It may or may not be appropriate for government to prescribe that behaviour for private sector companies, and that’s another reason why I think the development of public AI models within public institutions, where it’s natural and appropriate to have public control over those ethical decisions during design, should take place.
[Translation]
Senator Miville-Dechêne: I have a question for Mr. Sanders. I’ll ask it in French; Can you hear the interpretation?
I find you very optimistic, and it’s good to hear from you in this debate, but I read recently that the Claude Mythos AI system by the California company Anthropic is so dangerous the company refuses to make it accessible to the public.
So here’s my question: What is being done to ensure these very powerful private companies aren’t controlling the public debate and the debate on democracy and elections? There’s a movement building in Canada in that regard. A while ago, my researchers sent me a petition to impose a moratorium on powerful private AI companies working in these sectors and, as you said, let the public systems operate, as they have the common good at heart.
I know this moratorium idea is radical, but it sounds like you’re saying public systems just have to step up and take the space. That said, isn’t there a rather big risk these large, powerful corporations will remove the democratic aspect from the debate on our fundamental institutions? That would obviously be terrible.
[English]
Mr. Sanders: I appreciate the question, and I fundamentally agree with your framing about the danger of resting these powers in private organizations. I would say three things.
First, when it comes to that example of certain companies not releasing their most recent model developments, I would urge you not to take everything at face value or buy into the hype. I clearly think that AI models can create value and be very useful. I don’t doubt that for a second. I use them a lot in my own work and find them useful. However, I do not see evidence that some of the models recently written about have truly superhuman intelligence or capabilities that go beyond what a company that might employ, for example, a human cybersecurity researcher could do if directed to look for security flaws and foundation software, which is an example written about a lot recently.
Second, I think we should recognize that situation — where private companies have enormous sway over important democratic processes like elections — is, unfortunately, already true and a reality, and that predates AI. The easiest example to point to would be social media companies, which for decades now have had an enormous impact on elections around the world. Speaking here from the United States, I can say, from my government at least, I see shockingly little action to try to control and shape the behaviour of those private companies to align with public interests. I think we need more of that for AI companies, yes, but also for other technology companies and other industries, including the energy industry, for example.
Third, I agree with you — we should urgently take the constraining and concentration of wealth and concentration of corporate power as more of an issue about money than technology. As I said, AI has really remarkable use cases. It can be very powerful and beneficial in its use. But I’m not sure it’s so fundamentally different from the much older mechanism of money in endowing companies with power and capabilities.
I think the reason why most companies that have undue influence on democracy around the world today have that influence is because they have money. They can control large workforces of humans and command them to do work, and they can use their money to influence government and power. The fact that some of those companies now also have remarkable technology capabilities hasn’t changed that equation much. I think it’s a concern about finance more than technology.
[Translation]
Senator Miville-Dechêne: If I understand your subtle answer correctly, a moratorium, as some Canadian researchers have suggested, wouldn’t be the right way to go to maintain a harmonious public debate, one not controlled by private interests?
[English]
Mr. Sanders: I haven’t seen that specific moratorium proposal. I’m not familiar with its details. I know that, in the past, there have been moratoriums proposed on AI development, including in the U.S. In general, my concern is that it’s not practical to limit private behaviour in doing research into new technologies. I’m not sure a moratorium like that would really constrain the ability of the technology to advance it in its capabilities.
There are open-source development and international actors developing and pushing forward AI technologies in countries around the world, but I don’t think they would be constrained by a moratorium like that. However, I do think there is an urgent need for governments to constrain and shape the behaviour of the private sector in the public interest and for all those governments around the world. Again, that has more to do with wealth than technology.
Senator Simons: I have a question for Professor Sanders. One of the challenges of sovereignty in AI is the question of energy and the compute power needed to make these things run. These systems require extraordinary amounts of electricity and water.
Canada is, in many ways, well situated to meet those challenges because we have access to green hydropower here. We have access to a lot of water.
Are there concerns that you see from an environmental perspective in cranking up the capacity to compete on a global scale as a sovereign AI power in terms of how we manage those needs for power and fresh water?
Mr. Sanders: It’s an important question because I think the environmental impacts of the industry as it has been developing are not only large but disproportionately or unequally distributed. Certain communities have been burdened with intensive development of data centres that cause acute environmental impacts, and that’s largely a consequence of the industrial model that has been developed. It need not be that way.
I described earlier how the biggest technology companies in the world today are competing to release new models every few weeks that try to leapfrog by a very small amount the performance of the previous model. That is a very energy-intensive process that is maybe natural in a competitive capitalist system where a small number of companies have trillions of dollars of investment to command to do that. But I don’t think it’s necessary by any means to create value from this technology.
I think there is a better and less energy-intensive way to do that development. Incremental development that is not chasing tiny performance gains but rather some of those fundamentally different bases for creating value, such as more trustworthy AI, would be a much better allocation of our resources.
Also, when it comes to the use of the technology, so many companies today are trying to differentiate themselves in the market by showing that they can use more AI — by showing that every interaction you have with their product can be AI-driven, regardless of whether it really creates value to use it in that context. That is an unintended consequence of the development of the industry that is not a responsible use of energy resources.
I’m not an energy policy expert, but I would look to government to shape the industry to make sure that there is not a disproportionate impact on certain communities through the choices that this industry is making in its development — to make sure that there is internalization of those environmental consequences so these AI companies don’t just externalize the ecological, energy and affordability consequences of their use of energy and water.
There are things that the government can do to steer these private sector companies to be more responsible stewards of our resources. Again, I think there is a role for the government to play in demonstrating and providing an alternative path for development that is fundamentally more sustainable.
Senator Quinn: Thank you to both our witnesses for being here today. Thank you, colleagues, for the questions that you have been asking and the responses that have been given. It’s a real learning opportunity for somebody who is definitely outside their field of expertise.
I wanted to ask about people who are more like me, who don’t know a lot about this but see AI used when they perform a search on Google or elsewhere. Up comes the first answer, and it’s AI generated. Also, we’re getting phone calls that are probably AI‑driven. How concerned should we be with the fraudulent or untoward use of AI? Is that something that should be of concern to the public in general and the very specific communities that both of you come from?
[Interpretation]
Mr. Beatty: We are absolutely concerned about fraudulent activity, particularly for sign language users. That is why we’re making the push for sovereignty in sign language in IT use.
We’re getting closer to seeing more sign language models being used, but within the past year, it’s become more and more perfected. We’re starting to see realistic models appearing in AI, so technology is changing and improving quickly. Our concern is the gap in accessibility governance. How do we monitor or control that? What if that technology is used for fraudulent activity?
We want to make sure that people don’t develop things and think it’s generally okay. We need that expertise to review what those models are doing before looking at setting up ethical standards that have already been proposed. We want to respect the deaf, deaf-blind and hard-of-hearing communities in their use of sign language. That is absolutely key to our community. Be careful of fraudulent activity.
Again, if we put those governance models in now, that will work much better than adding them as an afterthought. If we’re looking at ethical decision-making within these models, we don’t have anything right now. We do have technical experts who are looking at this. The expertise is out there, and now is the time to consult to get going on it.
[English]
The Chair: Do you wish to make a final comment, Mr. Sanders?
Mr. Sanders: I think this is a very important question as well.
Fraud using AI is real, accelerating and a very significant concern. Maybe I can share an example from the U.S. and the insight that I would take away from that example. In the U.S., about a year ago, our Federal Election Commission, or FEC, considered the question of whether existing laws that prohibit what is called fraudulent misrepresentation in campaign advertising should apply with AI. Historically, in the U.S., you have not been allowed to dress up in the costume of another candidate, pretend you’re that candidate, record a TV ad and put that on TV. That would be fraudulent misrepresentation.
Today, of course, we have that happen with AI. People are using “deepfakes,” generated fake videos of other candidates saying things. The FEC had to consider the question: Is that fraudulent misrepresentation? Is it okay because the technology has changed? Is it okay because you are using a computer to do it instead of a costume?
Ultimately, they clarified that the existing law does apply to AI. It is still a fraudulent misrepresentation when you use an AI “deepfake” to do it, and that should not be allowed. To me, that is the obvious conclusion. The insight is that when we have existing laws and prohibitions on activities and behaviours, in general, they should apply as new technologies are developed. Sometimes we — as researchers or in policy — overthink what is new about a new technology. Is it really changing things? Often the answer is no; it may be a different mechanism, but the behaviour is still clearly the same, and the existing law should apply in the same way.
Wherever there are potential loopholes or gaps in the law where it’s not clear that it should apply to the new technology of AI, we urgently need to find and close those loopholes so they cannot be exploited by private actors. Of course, there is enforcement as well. In the U.S., the FEC historically has not had a lot of resources for enforcement; that allows widespread deviations from those rules and norms. The scale at which AI makes it possible to perform that kind of fraud reinforces that more investment and enforcement are often needed.
The Chair: I would like to thank Mr. Sanders and Mr. Beatty for their outstanding job today in sharing information with us and responding to senators’ questions. Based on that, we have reached the end of our time for this panel. I would like to thank you all so much for your feedback and information. It is much appreciated.
I would like to now introduce our next panel. Here with us today we have Martin Waxman, Adjunct Professor, Schulich School of Business, York University. Welcome, Mr. Waxman.
[Translation]
We welcome also Mr. Simon Claus, Director, Public Affairs, from the Association québécoise de l’industrie du disque, du spectacle et de la vidéo.
[English]
Thank you both for joining with us today.
Witnesses will provide opening remarks of approximately five minutes, which will be followed by a question-and-answer session with senators.
I will now invite Mr. Waxman to give his opening remarks.
[Translation]
He will be followed by Mr. Claus.
[English]
Mr. Waxman, you have the floor.
Martin Waxman, Adjunct Professor, Schulich School of Business, York University, as an individual: Thank you. Good morning, Chairman Smith, Deputy Chair Dasko and honourable senators. My name, as you heard, is Martin Waxman, and I want to certify that my remarks today are 100% human and no AI was used to generate them. That may sound like a funny way to begin, but I believe this type of disclosure is one way to maintain trust as we establish guidelines around acceptable AI use cases for content and creativity.
I am an adjunct professor at the York University Schulich School of Business and the associate director of the school’s Future of Marketing Institute. That’s a global think tank that examines and analyzes tech trends and their implications for marketing and communications. I write the AI and Digital Marketing Trends Newsletter that has over 500,000 subscribers and create courses on generative AI for LinkedIn Learning. I also train students and marketing and communications professionals how to use AI beyond simple content generation and how to develop detailed and structured prompts that combine natural language and systems thinking.
In addition to all of that, I have also published two books of fiction, and I am a former film and television writer. My background is in the arts and not coding. I appear before you today highly conflicted about the role AI will play in content production and distribution and its effect on writers and creators.
On the one hand, I am very excited; on the other, I’m unsure how to deal with the unprecedented pace of change and uncertainty we’re all living through every day and that’s causing a lot of people a great deal of anxiety. I can see incredible potential for AI to help provide writers with an endless array of new ideas and perspectives they would not have otherwise seen.
I also know that in the creative industries, like marketing and PR, professionals are under a great deal of strain to produce larger and larger quantities of content and that generative AI has the potential to take a good part of that load off their shoulders.
Yet when I watch how people use AI, including my students, I am often disheartened by how much thinking they outsource to a machine. And that, along with the many dire predictions around job cuts, frightens me.
Rather than having a challenging conversation with a large language model, many people use AI as a shortcut to avoid the difficult task of thinking. They make a generic request, have a brief exchange and get an output that looks good enough but lacks the quality and spark of something unique. They’re unable to see beyond the mediocrity of that output because they’re missing subject matter expertise.
They are not sparring with AI, not engaging in provocative discussions and not asking it to challenge their assumptions and ideas. They accept what it gives them and move on, choosing the easy way out.
I am concerned that many people, and especially new professionals, will miss out on the valuable and difficult experience they get by anguishing over a blank screen or taking the time to reflect on an idea, let it percolate and revise it into a polished piece of work. If we decide to offload our thinking, our ideas and the creative labour involved in content production to a machine, how will young people ever develop their craft, imagination and determination? What role will they have in the workplace?
I also wonder, as was mentioned earlier in one of the questions, who owns the output of a piece of content that’s AI generated. How much of a contribution does a person have to make to claim authorship?
What laws and regulations are needed to pave the way to a promising AI future for everyone? These are just some of the difficult questions we must discuss and debate right now. While I don’t have the answers, I believe we should start by focusing on three areas: research, training and regulation.
First is research into how these systems perform, as well as into their emergent behaviour, that is, when they go off the rails, scheme and present hallucinations as facts. There is also research into how people are using AI and whether their minds are being stimulated or atrophied by assigning too many cognitive tasks to a machine.
We need to develop and implement a formalized AI training curriculum that starts in primary school and continues throughout a person’s education. And we must put an emphasis on teaching students how to think, analyze and make judgments for themselves.
We need to develop policies, guidelines and regulations that are codified into our laws. That might be the most difficult part of all. It was challenging enough to put regulations around social media, and finding a way to regulate AI is even harder because the platforms and capabilities are evolving so rapidly.
By nature, I am an optimist. I believe AI has the potential to make us smarter and more creative and to enhance the quality of the content humans produce, but it is incumbent on us to use it intelligently and guide its outputs rather than letting its outputs guide us.
In my remarks that you will all receive electronically, I am attaching a public resource I created that’s accessible to all of you. It is a list of links to over 10,000 pieces of AI and technology news, trends, research and articles from a variety of reputable sources, and it’s a database that I update all the time. You can access it by visiting the link in the document.
I am happy to answer any questions that you may have or to assist the committee in your study today or in the future. Thank you.
The Chair: Thank you, Mr. Waxman.
[Translation]
Mr. Claus, you have the floor for your opening remarks.
Simon Claus, Director, Public Affaires, Association québécoise de l’industrie du disque, du spectacle et de la vidéo: Mr. Chair, senators, thank you for inviting the Association québécoise de l’industrie du disque, du spectacle et de la vidéo, or ADISQ, to participate in this study. I am Simon Claus, Director of Public Affairs.
Our association represents over 200 members working in the music sector. The association’s mission is to prepare and support its members and ensure the framework in which they work fosters their development. Maintaining a competitive music industry means ensuring our music and shows continue to exert their influence here and abroad.
Let me begin with a fundamental reminder: Music is unlike any other product. It’s with us every day, helping to define who we are. At the heart of all music is the artist who embodies what they sing through their experience. This experience creates a unique relationship with the public, a foundation of trust built over time. It’s not built through a probabilistic inference system.
A whole ecosystem surrounds that artist. They’re entrepreneurs who take risks and invest time and money in the projects they believe in, and generate considerable social, economic and territorial benefits. That ecosystem is made up of small, independent structures. It is therefore fragile, and some AI evolutions make it even more so.
Nowadays, AI is everywhere, from creation to music consumption, production and broadcasting.
We’re not pushing back against this new reality. The various trades in our industry can use AI as a tool, but its development cannot come at the expense of our music or the social and cultural model we’ve collectively adopted.
We support the positions of the Coalition for the Diversity of Cultural Expressions, of which we are a member, namely the ART pillars: authorization, remuneration and transparency.
Big tech has trained their generative AI systems on entire catalogues of copyrighted musical works without authorization, remuneration or transparency, making it impossible to build a fair market.
First, transparency about the music content used to train AI models is a reflection of the rights’ effectiveness. However, things right now are unclear.
This transparency is also necessary downstream when it comes to AI-generated music, known as synthetic music.
In a recent study, the Deezer platform found that about 60,000 AI-generated musical works are released every day, which represents about 39% of all music streamed daily.
Second, rights holders must authorize the use of their content. That’s the principle of copyright. There can be no AI training without consent; that’s the opt-in principle.
Third, any use of musical works and recordings must result in financial compensation.
As I mentioned, AI is at all levels of the value chain and impacts all trades, whether artistic, technical or commercial. However, our companies are up against giants investing heavily in these technologies. They don’t have the same resources to adapt. That’s why it’s important that governments support an active and responsible transition to protect skilled jobs and give our businesses the tangible tools they need to stay competitive.
We’re not opposed to AI, and we understand its potential societal and economic contributions. AI development has to be ethical and must respect the rights of creators whose work is being used as raw material to train generative AI.
Our cultural policies are not a barrier to innovation. They’re a prerequisite of that. They stimulate creativity and entrepreneurship while cultivating and protecting what distinguishes us. AI evolution must be in lockstep with the growth of our music ecosystem while protecting our cultural diversity. It’s with this mindset that we call for the creation of a fair framework that benefits all parties. Thank you for your attention, and I am available to answer your questions.
The Chair: Thank you, Mr. Claus.
[English]
We’ll start questions with our senators and special guest star, Senator Miville-Dechêne.
[Translation]
Senator Miville-Dechêne: Thank you for giving me the floor, since I have to leave a little earlier. My question is for Mr. Claus. I understand this is a very important issue for Quebec musicians. Explain to us how things work right now.
Do they receive any compensation? When a musical work is used to train an AI, do they receive anything? Explain to me what the status quo is. Don’t the creators receive anything?
Mr. Claus: The short answer is very simple: They get nothing. It’s been proven that all music content on the internet has been “harvested” to train generative AI software. This is known as text and data search. There are trials in the U.S. where this harvesting has been acknowledged. We consider this process as unauthorized, uncompensated reproduction, and therefore a copyright infringement.
Our music content is the raw material used to train AI. To generate quality content, what we call output, you need quality raw material, in other words, professional works. We’re competing with an increasing amount of AI-generated music, including on streaming platforms.
That is the current situation, in a nutshell.
Senator Miville-Dechêne: I admit I didn’t know you were competing with so much AI-generated music. Isn’t that perhaps the biggest threat?
Mr. Claus: Absolutely. ADISQ has a database of the various releases on the music charts. There’s an increase in AI-generated music, and even AI-generated music identified as Quebec music. For example, there are many Christmas musical works on the playlists.
Let me give you an example. There’s a rock band called The Velvet Sundown which got very good numbers on Spotify. They’re easy to find, they got a lot of media coverage. The group released more than three albums in less than a month. That’s impossible for an artist. There’s a lot of what we call noise taking up more and more space on the platforms.
That’s also why, on the one hand, we’re asking for some form of transparency regarding the input, meaning what data is used to train the machine, and, on the other hand, we’re asking for transparency regarding the output. Purely generative music, or synthetic music, should be linked to a form of digital marking so it can be identified as such, if only for the public. The public also has a right to know what they’re listening to.
Senator Miville-Dechêne: Music, songs with lyrics . . . . Another issue you haven’t talked about is that French is a minority language on the continent, so the threats are even greater to a minority culture.
Mr. Claus: Absolutely. AI is a system of probabilistic inferences that has led to massive corpora of works. AI doesn’t compose. It mashes and creates works at the centre of the spectrum to increasingly gravitate toward what is statistically dominant, the most common forms, the most depicted aesthetics. That raises a lot of questions about the protection of cultural diversity and minority cultural expressions.
[English]
Senator Dasko: Thank you, witnesses. I am also very interested in the topic of copyright. We just heard about the Quebec situation. I would like to ask Professor Waxman about his understanding of where we are on the copyright issue in the rest of Canada and around the legal status. Has the ship sailed on this? Has all of the content been absorbed by the AI systems such that we cannot regulate? Surely we can still regulate in this area. I really want to understand where we are on copyright. I have another question for you after.
Mr. Waxman: Thank you, senator. It is a great question. Music is one thing; writing is another. We are starting to see it with visuals and more of it with film and video with the synthetic platforms from Google, for example, or from OpenAI.
It seems as if we have not caught up with copyright. The big question in the U.S. is about fair use. How much material can you use to train an AI system before you are breaching copyright? How much credit do you have to give? How much compensation do you have to give to creators? As my colleague said, the amount of compensation is zero. That is one area that needs to be looked at.
A lot of the data that we have created — whether it is text‑based, audiovisual, musical and so on — has been used for training, but we as humans are pretty prolific and keep producing more data. So I don’t think the ship has sailed away yet, but it is about to leave. I think copyright is essential.
In the U.S., about two or three weeks ago, there was a case before the Supreme Court where they said that AI output could not be copyrighted. That is a good thing. But the question is this: What percentage of input do humans have to contribute in order to get that copyright?
Senator Dasko: There is the copyright with AI but also the copyright of data that goes into AI; we are talking about newspapers and vast amounts of information that has been collected and is owned by entities who are now having their contributions used without compensation.
Mr. Waxman: Yes.
Senator Dasko: Do governments need to regulate?
Mr. Waxman: Governments definitely need to regulate. We need to take a leadership position in all this. We need to have more current regulations on questions around the ethical use of AI, data safety, security, privacy and phishing emails where AI systems are tricking human beings. As a former artist and current writer and content creator, I believe creators must be compensated for their work.
Senator Dasko: Thank you. I have another question. I want to go back to your comments about the use of AI outsourcing our thinking. I might object to that a little bit.
Surely we can look at the output of AI as an input into our own human activities, right? We all do research, and research is incredibly valuable for us in our jobs. As long as we have control over what we do in the end, isn’t that the most important thing?
You use AI, you are writing a speech or something, and you are in charge at the end of what is produced — the outcome of your use of AI. Wouldn’t you say that is a really important thing?
Mr. Waxman: It is a very important thing. I agree that AI is amazing for research.
Senator Dasko: Absolutely.
Mr. Waxman: But if all we are doing is reading these machine-generated summaries and never going to the source — that is an extra step, when you get the results of research — we are missing a key piece of human-created content and nuance.
I see this especially with my students. I encourage them to use AI in their assignments but to disclose it and use it in a smart way. They have to produce content. What many of them do is give AI a fairly simple prompt, not a sophisticated prompt. They will receive content that looks really good on paper. However, they are not subject matter experts, so it is new to them and not new to the world. Creating something that is new to the world or has an interesting perspective comes with experience and expertise.
This is my concern. I use AI all the time, although not for personal writing. That’s where I draw the line. If we don’t teach people how to think, how to go to original sources and how to put in more detail and more of their thinking and ideas into the systems, then we’re going to have a generation of people who just ask it things, accept the response and move on. That’s not what I believe we should have.
Senator Lewis: Just to build on what you spoke about, there is obviously a spot for early education on this, like in K to 12 education. Is Canada doing enough now, as the technology is just emerging? How do we ensure that it’s part of our education system so that students can learn how to use AI but not be used by it?
Mr. Waxman: I believe this is a huge conversation between the federal government and all the provincial governments.
As all of you know, changing any system like the curriculum in K to 12 schools is an enormous task. We need to teach students how to use these systems because they are using them in the same way that social media was used by younger people when it came in, regardless of whether they were prohibited from using it.
Younger people will start using AI systems. We must give them the type of training that helps them become, in essence, coders or programmers who can work with natural language and can work in English to develop a program. If you write or look at a sophisticated prompt, it’s very much like a program. It’s written step by step. You have things called delimiters, which are little similar symbols that the machines treat as punctuation, and if you just say to an AI system, “Give me this,” you’re not getting it to pull from the right data sets to give you something really specific and nuanced.
That is a big challenge. On the university side, it’s a huge challenge. A lot of it has been pushed to individual instructors or professors to deal with. There are some bigger policies. That is a challenge, but the government needs to lead that discussion because, again, as you heard from previous panellists, industry is not going to do it.
Senator Lewis: I have a question also for Mr. Claus. We talked about other jurisdictions throughout this whole conversation. Are you aware of an example of another jurisdiction outside Canada that is doing a better job of protecting copyright, musical content and the creators of that content when it’s used in AI?
[Translation]
Mr. Claus: I don’t think we’re respecting the Copyright Act. That’s why we’re simply asking that the Copyright Act be respected.
Actually, to summarize our requests, new exceptions for text and data searches must not be introduced to the Copyright Act. What we’re asking for and where the government can be proactive — we talked about this in the context of Bill C-27, which died on the Order Paper — is transparency: A little more transparency on the corpora of data used to train these AI systems.
Here’s a good example: The EU has adopted AI legislation, which I’m sure you’ve heard about or will hear about, that now requires AI providers to provide a sufficiently detailed summary of the content used to train their algorithm.
There’s another example that came out last week: The French senate passed a bill that goes a little further, saying generative AI models are presumed to be using protected content. What does that mean? It means they’re reversing the burden of proof. In cases of litigation, these platforms will have to prove they haven’t illegally used content.
So, yes, I think Canada is playing a leadership role. For example, the principle of fair dealing is much stricter than in the United States, but I think we can go even further by drawing on legislation similar to ours.
Senator Simons: I have a question for Mr. Claus and Mr. Waxman.
[English]
When I make roast chicken with a beautiful chicken I bought from the farmer’s market, it has a different flavour and texture than the chicken nuggets that I always refused to buy my daughter when she was a child.
I very much worry that we are creating a market in which people prefer the taste of chicken nuggets to the taste of the real thing. It doesn’t matter whether we’re talking about music, video, images or text: AI can give us a simulation of the real thing. I’m existentially distressed by the fact that people can’t seem to recognize pablum and the chicken slurry of the art world.
I wanted to ask each of you, in English and in French, what to do to make sure that people have the aesthetic taste and skill to differentiate between slurry and delicious organic content?
[Translation]
Actually, I don’t think it’s the same thing.
Mr. Claus: Transparency is one of the principles we’re defending. Can we at least have transparency on what you call garbage content or bad chicken nuggets?
There are principles, but there are also solutions and digital makers for synthetic content. It’s the choice Deezer has made, for example. The platform said, “We’re going to develop software — so technology reacting to technology; that’s the world we live in — that allows us to identify all synthetic content, because the content used leaves a trace in the works created by generative AI. We’ve also decided to rely on human creativity, so we’re deregulating this music content, meaning they’re not going to benefit from the platforms’ promotional tools.” Bandcamp, another platform, went further and banned them completely.
Bill C-11, a bill you know well, was passed at the time. The CRTC is now implementing it. One of the demands of the cultural community, whether in audiovisual or music, is that cultural content, meaning music or audiovisual content, that will benefit from recommendations, promotions or quotas should be produced by humans. Humans can use artificial intelligence — I’ve already said this — but according to the definition of Canadian content, we must ensure humans are behind it. That seems to be the direction the CRTC has taken with its latest decision on audiovisual.
[English]
Senator Simons: Watermarks are a great idea. Transparency is a great idea. However, if people prefer garbage, I don’t know that watermarking solves the problem.
Mr. Waxman: That’s a really challenging question. If you take the chicken nugget example, which I love, and apply it to everything else we do — even just taking it literally, many people like to eat chicken nuggets and junk food. There is a proliferation of that everywhere you look.
I’m sure your roast chicken tastes much better. If you bring in samples, I would be happy to do a taste test.
However, I think the question comes down to what you said: judgment. How do we teach younger people to have that kind of judgment, those aesthetic values, to tell the difference between junk and something that is thoughtful and has humanity and heart poured into it? Does that mean a human can’t work alongside AI to produce something? I don’t think so. I think there is a middle ground here.
However, I think the danger of all this AI slop, which is what it’s called, proliferating is something we have to watch out for. And regarding disclosure, if people aren’t paying attention, it doesn’t matter. It doesn’t matter what some of the fast-food chains share about the fat content or the fact that their food is really unhealthy for you. People who want to eat it will just eat it and ignore that.
So I think it goes further; it goes to teaching people how to use these tools but also how to think.
[Translation]
Senator Cormier: My first two questions are for Mr. Claus. I’d like to come back to what you said about France and the presumption that AI systems are using cultural works. Actually, you said the burden of proof was reversed and companies would have to prove they didn’t violate copyrights. How do you think that could be applied in Canada? Do you think it’s a promising measure for Canada? That’s my first question.
Also, at the Artificial Intelligence and Culture Summit that was held in Banff, cultural organizations talked a lot about the need for digital literacy training. What do you think are the main shortcomings for musicians and arts organizations in that regard? What would be the main support needs?
Those are my first two questions. If I have time, I’ll have a question for Mr. Waxman.
Mr. Claus: I’ll try to give you time for your last question. First off, I’m not a lawyer. This is something brand new; it came out last week. I think the idea of presuming AI models are using protected content and companies having to prove in litigation they haven’t misused that content is very interesting.
From what I understand of this bill, the idea is not to increase the number of trials, but to create a disincentive to encourage AI actors to abandon certain predatory behaviours related to our cultural content and rather negotiate with the cultural creation sector. It’s possible to put a system in place. In fact, there are major discussions going on between the main players in the sound and audio sectors regarding the implementation of licences.
We aren’t part of the discussion once again, obviously. We don’t have the catalogues or the clout to wield the bargaining power to bring these giants to their knees. However, I do think that this type of incentive and bill will help to encourage these players to better comply with our Copyright Act. I think that we really need to explore this. Local music producers, no matter how talented and no matter how extensive their catalogues, have a hard time proving that their music has been used to train AI systems. The entire package is embedded in an extremely large volume of protected content. Yet here, we’re doing the opposite. I find this worth noting. We don’t necessarily have the technological capacity to prove it, but the opposite can be done.
Regarding your second question, it boils down to the idea that yes, artificial intelligence also provides opportunities. We’re a professional association ourselves. We’re currently developing training programs for our members. We’re trying to move past our initial sense of bewilderment. Opportunities are available at every stage of the process. We can talk about creativity. AI can help us explore new versions. We can initiate new creative processes by using AI as a training partner and by saying that we want to try out a certain version of a certain song. It offers productivity gains for certain tedious or routine tasks, such as post-production, mastering or mixing. It’s also useful for all aspects of promotion and distribution. We can explore ways to make our promotion and distribution campaigns more effective.
Senator Cormier: Are there any training programs planned for these different aspects of creation, distribution and production?
Mr. Claus: We’re in the process of identifying this. We’ll start by identifying the needs of our members. We’ll then develop training programs.
Senator Cormier: Okay. Thank you. If I have the time, I would like to ask Mr. Waxman a question.
[English]
From a communications and reputation perspective, what are the main challenges that AI poses for cultural organizations regarding the authenticity of their relationship with audiences, particularly in sectors where trust and human experience are central? It’s a bit of a philosophical question, but we’re talking about authenticity here, human relationships with audiences, so what can you tell us about that?
Mr. Waxman: I think that’s a really important question. I actually did my master’s — I graduated in 2019, so not that long ago. That was the topic of my thesis, which was called “My BFF is a chatbot.” It examined human-AI agent relationships and how they affect trust and the way we communicate.
I think for organizations that use AI and don’t disclose it, there is a very large risk of harming their reputation. I think there is a risk of reputational harm from “deepfake” videos that bad actors produce maligning organizations. I think there is an incredible risk of disinformation that proliferates about organizations, and organizations don’t have the mechanisms to combat it or to restore their reputation. That’s a question, again, with no easy answers. Unfortunately, none of these questions have easy answers.
This is, by coincidence, Disinformation Awareness Month in the U.S., and the Institute for Public Relations is hosting a series of webinars and a master class on the topic that you mentioned: How do you navigate and keep your reputation and trust with your stakeholders or audiences?
I think one way is to be transparent. That transparency in how you use AI is probably going to be a moving target, because right now, I think it’s very important to disclose that something was created with AI or AI was used to do something. If it’s a customer service chatbot, for example, disclosure is essential. But as we use the tools more, that disclosure is probably going to be diminished. For instance, a great example of AI that we all use is spell-check. Does anybody here disclose when you use spell-check in a document? No.
So right now, I think disclosure is important, but we need to know that’s going to evolve over time.
Trust is something that, as all of you know, breaks in a second and is really hard to build.
Senator Cormier: Thank you.
Senator Arnold: There is a lot here. I’m just trying to grapple with the chicken slurry versus — that authentic theatre experience you have when you are an audience member and you see something that is so totally unique.
I’m also grappling with the idea that it is Disinformation Awareness Month and someone in the U.S. has just posted an AI-generated picture pretending they are someone they are not. There is a lot happening right now.
Professor Waxman, you started with the disclosure that no AI was involved. I didn’t know that there is no output copyright on AI. I was happy to learn that.
Watermarks just seem so weak to me in all of this. I’m trying to put it all together and understand what lessons we learn from this. This is maybe an unfair question, Professor Waxman, but I feel as if you have been around in this ecosystem for a bit: What lessons should we have learned, that we maybe didn’t learn, from social media that could be applied here?
Mr. Waxman: Wow, that’s a great question. I think one of the biggest lessons is that if the tools are there, people will use them whether there are rules banning their use or not. Knowing that people are going to be using the tools, we need to show them how to use them in a way that builds trust and transparency and that goes beyond chicken nugget-type content or processed content. I really like that expression, so I may borrow that: the whole idea of processed content.
Senator Simons: Just because you said you like it, it’s yours.
Mr. Waxman: Thank you.
Let’s talk about AI-generated music. Any of us in this room could go to a number of free music-generation platforms, put in a prompt and get a summary in the form of a song — a country song or whatever genre you want — about the proceedings here today. We would listen to it and go, “That’s pretty good.” But the thing is, we’re not songwriters, at least I’m not, so what a songwriter might produce by knowing how to use these tools would be of much better quality.
It really comes down to quality and judgment. There are some cases where generic AI slop, if I can call it that, is okay, like for instructions, say, to put together a piece of furniture. We don’t need that to be brilliant. We just need it to explain things to us in a way that makes us understand. For anything aesthetic, and that goes to the content that we’re producing, whether musical, video or written, we need to think about how we train people to have the judgment to know the difference between good and bad and between something that is worthwhile and something that is just AI slop. The thing is that human beings have been producing slop for a long time too. The systems are getting better. We have to get better than the systems.
[Translation]
Senator Aucoin: I have two questions. My first question is for Mr. Waxman.
Senator Arnold basically talked about social media. I’ll be talking about cellphones.
Since the advent of cellphones, an entire generation has lost its grasp of interpersonal relationships. We’ve lost that skill. With artificial intelligence, aren’t we running the risk that, in 10 or 20 years, we’ll realize — as you said — that the young people of the new generation won’t know how to think anymore, because they’ll just need to press a button and say that they want this information and they’ll accept it? At that point, we’ll realize, 10 years too late, that entire generations of people can no longer think.
[English]
They don’t have any critical thinking.
[Translation]
What are your thoughts on this?
[English]
Mr. Waxman: Thank you. I wish I could respond in French. Your point about cellphones is a really good one. That is a discussion that my wife has with me all the time when I’m on my cellphone, maybe a little more than I should be when I’m with her.
I think there is a big danger of building relationships with these systems because they appear to be empathetic in a synthetic way because they are always supportive. They are built to be supportive and help us out. If a generation of young people are always asking these systems questions like, “What should I do?” and, “How should I behave?” and the AI system tells them, “You’re really good. You’re doing a great job,” they won’t develop the critical skills needed to step back and know that maybe that’s not such a great idea. We have seen examples of younger people who have committed suicide or been pushed toward psychiatric issues because of advice from these systems.
Again, it comes down to the fact that they are here. We can’t get rid of them. We can’t ban them. We need to start thinking about how we use them and when we use them. That takes time. A lot of us are in a hurry and want a quick response. We need to teach people that it’s okay not to have the answer in a second.
[Translation]
Senator Aucoin: My second question is for Mr. Claus.
Who will have jurisdiction over artificial intelligence for all our artists, for example? Will it be the provincial governments or the federal government? We’re talking about us at the federal level. However, you also talked about their management of artist training, video content, music, lyrics and writers. Can you comment briefly on this?
Mr. Claus: I think that we all share a responsibility. That would be my short answer.
Some steps can be taken at the federal level. I talked about what’s been done with the Broadcasting Act and about what the CRTC is doing. Copyright falls under federal jurisdiction, so steps can be taken at that level. In addition, Bill C-27 introduced some useful aspects in terms of transparency.
Steps must also be taken at the provincial level. For example, if we’re talking about deepfakes, this concerns the Quebec civil code. We don’t have the right to use a person’s name, image or voice. This constitutes an invasion of privacy. Different legislative responses are possible depending on the legislation.
Our broadcasters also have a responsibility. We encourage radio stations to play artists because that’s what people want to hear. The platforms also have an individual responsibility.
I would say that we need to make a collective effort to decide what kind of world we want to design around this AI in terms of music, but also on a broader scale.
[English]
Senator Mohamed: I used to be on the board of Music Canada. We grappled a lot with the issue of copyright. I understand the transparency and all the reporting. I think in the Copyright Act, you have to disclose, within three years, whether there has been an infringement on your material. There are civil and criminal remedies that can be taken.
Because of the prevalence of this and the concern regarding how widespread this could become, do you feel it would be advantageous to increase the penalties in relation to AI being part of copyright infringement? Should we just leave them where they are? Because voluntary disclosure is voluntary disclosure, and in a world where this moves so quickly, how do you make sure that you’re saying to people that this is not okay? Do you change the three-year window and make it longer, or do you say there will be more severe civil or criminal penalties? How do you handle the other end? You can try to dissuade people, but once they have done it, how do you make sure that doesn’t become the norm where people just get away with it?
Mr. Waxman: I’m not an expert in this area.
Senator Mohamed: What’s your view?
Mr. Waxman: I wouldn’t mind reframing the issue. Often when there is a penalty, if the company that is being sued is, say, one of the big tech companies, they have so many more resources, so the penalty doesn’t really matter to them. I would rather reframe it as a way of compensating creators and build in compensation rules and regulations early so that they don’t have to get these penalties. For Google and OpenAI and Meta, the penalties don’t mean anything to them. They just pay them and move on.
Senator Mohamed: Mr. Claus, do you have a response?
[Translation]
The Chair: You have 30 seconds, Mr. Claus.
Mr. Claus: We need to impose penalties commensurate with the capital of these companies. We live in a world where certain companies have the same capital as certain countries. I think that we can set a precedent here, because we aren’t alone.
I’ll refer to an age-old formula. I think that middle powers, such as Canada and other countries, have a vested interest in implementing penalties when people fail to adhere to the cultural and political model developed to address the issues at stake. A small penalty multiplied by a given number of countries starts to add up to a big penalty.
[English]
The Chair: We have reached the end of our time for this panel. Thank you for appearing.
[Translation]
Thank you both for sharing this valuable information with us.
[English]
I would like to thank our entire support team, those in the forefront of the room as well as those behind the scenes who are not visible. Thank you all for your work, which contributes enormously to the success of our work as senators.
(The committee adjourned.)