THE STANDING SENATE COMMITTEE ON SOCIAL AFFAIRS, SCIENCE AND TECHNOLOGY
EVIDENCE
OTTAWA, Wednesday, March 25, 2026
The Standing Senate Committee on Social Affairs, Science and Technology met with videoconference this day at 4:16 p.m. [ET] to examine and report on matters related to the impact of artificial intelligence in Canada.
Senator Rosemary Moodie (Chair) in the chair.
[English]
The Chair: I would like to call to order this meeting of the Standing Senate Committee on Social Affairs, Science and Technology. My name Rosemary Moodie. I’m a senator from Ontario and the chair of this committee. Before we begin, I would like to have senators introduce themselves.
Senator Burey: Sharon Burey, Ontario.
Senator McPhedran: Marilou McPhedran, independent senator from Manitoba.
[Translation]
Senator Boudreau: Good afternoon. Victor Boudreau from New Brunswick.
[English]
Senator Hay: Hello. Katherine Hay, Ontario.
Senator Arnold: Good afternoon. Dawn Arnold, New Brunswick.
[Translation]
Senator Petitclerc: Chantal Petitclerc from Quebec.
[English]
Senator Cuzner: Welcome. Rodger Cuzner, Nova Scotia.
Senator Muggli: Tracy Muggli, Saskatchewan.
Senator Senior: Paulette Senior from Ontario.
The Chair: Today, the committee continues its study on matters relating to the impact of artificial intelligence in Canada.
This study will examine issues including data governance, sovereignty, ethics, privacy, safety and the risks, benefits, and social impacts of artificial intelligence in Canada.
Joining us today for the first panel, we welcome by video conference, from Vector Institute, Glenda Crisp, President and Chief Executive Officer; and from the Alberta Machine Intelligence Institute, Stephanie Enders, Chief Delivery Officer. Thank you both for joining us today.
For your opening statements, you will have five minutes each, followed by questions from committee members.
Glenda Crisp, President and Chief Executive Officer, Vector Institute: Thank you, Madam Chair, and good afternoon, honourable senators.
Thank you for the opportunity to contribute to this critical study on the impact of AI in Canada.
My name is Glenda Crisp, and I am the CEO of the Vector Institute. I have over 35 years of experience in the technology, data and AI adoption space, primarily in the financial services industry.
The Vector Institute is one of Canada’s three national AI institutes, established in 2017 as an anchor of the Pan-Canadian AI Strategy.
Vector is headquartered in Toronto with over 960 affiliated researchers at universities across Canada. Vector-recognized AI master’s programs in Ontario also collectively produce over 1,000 graduates annually — over 90% of which stay and work in Ontario.
The easiest way to summarize Vector’s mission is this: taking cutting-edge AI research and enabling organizations to adopt and deploy it faster and more easily across a range of different industries.
Vector has more than 300 partners, from start-ups to global enterprises and broader, public sector institutions.
Vector’s industry sponsorship model is second globally only to MIT’s, as measured by the number of participating companies.
Canada has successfully built a world-class AI research base. Our three national AI institutes anchor one of the world’s most concentrated pools of AI researchers and graduate talent. The national AI institutes have each developed approaches that accelerate the safe, responsible and economically productive deployment of AI technologies across multiple sectors.
However, Canada’s private and public adoption of AI lags behind its international peers. In 2019, Canada was fourth in the Tortoise Global AI Index, thanks to being the first country with an AI strategy. However, since then, Canada’s position has been slipping. As of 2024, Canada now ranks eighth overall.
As noted, this decline is not uniform. Canada still ranks third globally in AI research. Where Canada is weak is in infrastructure, at sixteenth; operating environment, at eighteenth; and commercial strategy, at tenth.
This isn’t just something reported by global think tanks; I hear it on the ground as well. The message is consistent: Canada is excellent at producing AI talent and research, but Canada lags in adoption and deployment.
I have three main reflections to share with you today. First, Canada’s adoption problem is fundamentally a trust problem, and trust comes from understanding. Recent research shows 69% of regular AI users trust the technology, compared to just 5% of non-users. More concerning is that only 20% of Canadians feel well prepared for the changes AI is bringing.
Second, Canada’s $2-billion Sovereign AI Compute Strategy, announced in 2024, was absolutely critical but is two years behind schedule. The lack of commitment to spend is not the issue; the speed at which it is happening in real life is. In contrast, the U.K. — a middle power like Canada — has committed to similar resources but is ahead in operational capacity.
It only took the U.K. three weeks to procure its national AI compute infrastructure. Without faster deployment of compute, our world-class researchers will leave for institutions with better infrastructure and our start-ups will scale in the U.S. rather than here.
Third, increasing productivity through AI is genuinely difficult and requires a fundamental rethinking of business processes, upskilling the workforce and sustained organizational commitment. In other words, if you’re not rewiring your business processes around AI, you’re essentially buying a Ferrari and only running it in first gear.
Canadian AI small- and medium-sized enterprises, or SMEs, face risk aversion in procurement. Canadian companies looking for AI solutions from our SMEs often require a U.S. “reference customer” before considering them. This is backwards. We should be the first adopters of Canadian innovation, not the last.
Senators, Canada has world-leading research talent and a growing start-up ecosystem. Our banks are global leaders in AI adoption. However, without deploying compute faster and addressing risk aversion, we risk watching this competitive edge diminish further.
Thank you for your time. I welcome your questions.
The Chair: Thank you, Ms. Crisp.
Stephanie Enders, Chief Delivery Officer, Alberta Machine Intelligence Institute: Wonderful. Thank you, Madam Chair.
Good afternoon, senators. Thank you for the opportunity to speak today:
I’m Stephanie Enders, the Chief Delivery Officer at the Alberta Machine Intelligence Institute, or Amii, also one of Canada’s three national AI institutes.
We were also established in 2017 as part of the initial Pan‑Canadian AI Strategy, building on more than 20 years of foundational AI research excellence anchored at the University of Alberta, with Mr. Richard Sutton, a 2024 Turing Award laureate, as our chief scientific adviser, who really has laid the groundwork in a field called “reinforcement learning.”
We support world-leading research in artificial intelligence and machine learning and then translate that scientific advancement into industry adoption.
We really focus on three big pillars; invention, pushing the AI frontiers forward; innovation, advancing domain-specific AI and connecting to industry impact, creating new business models; and finally, we really focus on diffusion — how we can spread AI’s impact as wide as possible, creating receptor capacity for companies and people to leverage AI responsibly and with full agency regarding its use in their lives and businesses.
To bring these three key pillars forward in our daily lives, we focus on four key activities to drive forward Amii’s leadership.
The first is research. AI is still in the early stages. It is a scientific field, and there is much science still to be done. There is more to discover about its fundamentals and how the field can drive other key domains in Canada. Research creates the knowledge and talent needed to advance the frontiers of AI for Canadian benefit.
The second is literacy. We truly believe AI is the defining knowledge of the coming decades. It prepares people for an AI‑driven world and creates companies ready to advance and adopt AI in their operations and business models.
The third is adoption. It’s how we see the commercial impact of this technology and how it’s realized in society.
Fourth and last but not least are start-ups. Scalable start-ups providing AI tool options to Canadian businesses are the best way for AI to see positive disruptive impact in the world.
Our mission for Amii is “AI for Good and for All.” What that means is we think and act through a Canadian sovereign lens, where we must build Canadian’s capacity to leverage AI for long-term social and economic benefit. This includes strategic collaborations with trusted peers nationally, like our sister institutes, as well as international partners.
I’d like to highlight some of our work in each of the key areas so you have a better sense of what Amii does day in and day out.
First, with research, in our most recent expansion of Canada’s CIFAR AI Chairs, we focused on growing our AI and X models, which means we’re emphasizing interdisciplinary research and exploring how artificial intelligence can push the boundaries of research in fields like health, energy, physics and the humanities.
We see this as a new kind of bilingualism in which researchers advance both their domain and the frontiers of AI, creating a new generation of talent with these same skills.
On the literacy front, Amii has trained more than 250,000 individuals. Through our work in the K-12 education space and our AI workforce readiness program with post-secondary institutions, our impact has grown dramatically. Core to that work is the foundational understanding of this technology and the role it plays in building societal trust with organizations like our own and the technology we’re creating.
For that, our work, especially in workforce training, embeds a triple-matrix approach to all learning objectives: the UNESCO AI competency framework for students, a user AI persona aligned with fluency in given job roles, and the potential for a return on investment from AI adoption.
When it comes to adoption and start-ups, you can think of the way AI drives business impact in two ways. The first is the way that AI can be put to use: It can project, produce, prevent, promote or personalize.
The second is how it can drive a return on investment. Think about it not only for economic benefit but for productivity and efficiency; experience and engagement; resilience and optimization; and scalability and business transformation. In the past year alone, Amii has engaged with over 286 industry partners, including 102 scaling start-ups.
We provide advice, we support AI strategy and de-risk investments and we build AI to create talent pipelines all while finding tactics and actions to bring our mission to life.
In closing, echoing my peers from previous sessions, we know AI will underpin future prosperity and Canada must continue to advance it while remaining diligent about societal understanding, trust, agency and opportunity.
So thank you, and I look forward to your questions.
The Chair: Thank you. We will now proceed to questions from committee members. For this panel, senators will have four minutes for your questions, and that includes the answer. Please indicate if your question is directed to a particular witness or witnesses.
Senator Burey: Welcome. Thank you so much for being with us today. This question is for both our esteemed witnesses. We’ve heard from a number of witnesses before, as you know, consisting of Canada’s talent.
I’m on your side, Ms. Crisp, and on the Vector side. A new study reveals AI’s $100-billion economic impact across Canada, with Ontario leading the charge. I’m from Ontario.
We heard about the promise, the perils and the safeguards, and we are hearing today about trust and the lack of adoption. I want you to expand on that. What do you think we could do as legislators to move this forward while also keeping in mind the perils? So the promise and the perils — can you speak about that?
Ms. Crisp: I would like to double down on something Stephanie mentioned. I believe AI literacy is key. Trust is at the root of the problem in terms of adoption. I think rolling out that K to 12 education curriculum that Amii has developed across the country would have a meaningful impact. I think it would be a nice sideways to help drive adoption in business. As the kids come home and start talking about it, parents get curious and, potentially, that drives adoption in the workforce as well.
Ms. Enders: On the promise and perils, the first piece is knowing there are trusted partners with whom you can have open conversations, especially as business leaders, where you’re setting AI strategy. It’s understanding you have trusted organizations and collaborators that can help you understand what those risks are and navigate them with confidence.
On the literacy front, if folks are unfamiliar with the UNESCO framework, there are four main categories. The first is a human‑centred mindset, followed by ethics, tools and techniques of AI and, finally, AI systems. I think the conversation is dominated by that third element: AI techniques and tools. When we think about literacy, it’s about everyone having fluency in the entry level of all four of those competencies so that safety and the kind of understanding of what builds trust aren’t removed from the technology. It’s part of a cohesive conversation. We see these handshakes between what happens at the kitchen table, what happens in people’s post-secondary education, as they enter the workforce and the executive leadership room.
They are essentially the same questions: What should I be asking? How do I think critically about this technology? What are the opportunities for deployment that are most meaningful for the impact I want to have? Productivity and prosperity could be two of those levers you’re looking to drive.
Senator Hay: Glenda, nice to see you again. Thank you for being here.
My question is very similar to Senator Burey’s. I just want to pull at the thread a bit.
We know AI is in full play pretty much everywhere in our lives — such as work and health care — yet we talk about trust being quite low. We heard from “the godfathers of AI,” Professor Hinton and Yoshua Bengio. That created a sense that the pace is too fast or it is a little frightening on various levels. There are some legitimate concerns in there, when you see Senator Bernie Sanders and Claude have a dialogue together about privacy.
When you think of it not just economically but more broadly, how do we balance this fear, our natural instinct to say we have to be protective and stop — and real risk too — with the reality that we must lean in and lead in Canada? We have to.
I ask both of you that question.
Ms. Crisp: They are legitimate concerns. That being said, all three national AI institutes have researchers working in the space of safety in AI. Vector, particularly, is in the applied research space. We have created privacy-enhancing technologies, and we continue to advance those.
One of the things that we need to do is get the message out that the national institutes actually have worked in this space, and we are here to help businesses adopt in a safe way.
I think the bigger risk for Canada at this point is the fear that we are going to be left behind the rest of the world because we don’t even adopt more traditional AI-like predictive modelling, et cetera.
Ms. Enders: From Amii’s perspective, the three institutes collaborating in the Canadian AI Safety Institute allows that research capacity on trust and safety to continue to grow at a rapid pace. Then, it relies on the translational skills at the national institutes to bring it in.
Practically, what we have seen is providing frameworks and ways to have conversations about risk versus reward. We do that in a lot of ways.
At Amii, for our own work, when we decide which kinds of projects with industry to advance, we frame it under the sustainable development goals for 2030 and the opportunities for positive or negative impacts if this were to go to full deployment.
It’s not a decision-making framework. It’s a way to give language to the conversations.
From there, as we work on projects, I think managing risk and being open about it help build confidence. We use tools like algorithmic impact assessments, data sheets for data sets and model cards to really document those conversations that technical experts and domain experts are having as they work on AI so that they can be shared openly with, potentially, a team or organization and, in some cases, the public. That way, there’s transparency on the thoughts behind how we are deploying these private-by-design or technology-based elements related to guardrails.
Senator Hay: I ask this question to both of you. I love that the three institutes are working well together. It makes me feel safe.
On a hope scale, from 1 to 10, regarding the future of AI and how it’s shaping up, where are you?
Ms. Enders: I’m a hopeless optimist, so I’m definitely a 10 out of 10. I’m non-technical myself, and I have faith and great optimism because I see the passion and diversity of the people building this technology and how they want to serve the public good.
Ms. Crisp: I’m always a hard barker. I would go 9 out of 10.
Senator Hay: Thank you.
Senator McPhedran: Thank you both for being with us today. What types of AI deployment today would you consider to be premature and/or unsafe?
Ms. Enders: I can take this one first. For me, it’s less specific technology and more the process of how that technology has come to the market. The elements of technology that I would have reservations about are ones where there haven’t been open conversations about risk versus reward, and where there isn’t fulsome dialogue around how risks have been mitigated.
I think the conversation around AI is often dominated by generative AI, and I think we are seeing huge risks related to misinformation. People have questions about how to navigate that, but I think the bigger risk with any of the AI technologies is when there is not a fulsome discussion about risk.
Ms. Crisp: I would concur with Stephanie. My concern is more in the application of AI to inappropriate use. For example, I don’t think asking ChatGPT for medical advice is wise. It is not something I would recommend people do. I don’t think the large language models are at the point where they should be doing that.
That is not to say that there couldn’t eventually be some kind of a medical diagnostic tool, but I don’t believe it’s ready at this point. What we really need to see is how it is applied and governed and if there is a human at the helm.
Senator McPhedran: Thank you. Can I ask you if there is an assumption here that the benefits of AI will materialize while you’re treating the risks, essentially, as manageable? I’m interested to know, and I think that your answers were on quite an individualistic level. My question was geared more to national safety for our country, so let me just take it there and be clearer in my question.
What are the policy implications? What are the human and national implications if you’re wrong in this assumption?
Ms. Crisp: I will let Stephanie speak for herself, but my assumption is that there must be governance and human help. So if that is false, if we take our hands off the steering wheel and let AI do whatever, then there are major risks, but I don’t believe that’s what we’re suggesting. I don’t believe that’s what we’re trying to move toward in Canada.
Therefore, the policy implications would be that there has to be governance, transparency and oversight. Frankly, companies that are adopting AI, whether buying it or building it, are accountable for the actions that AI takes, whether it’s providing a recommendation or actually taking action. There needs to be accountability.
Ms. Enders: Looking at it from a national perspective, when it comes to governance, policy or regulation, I really think about something around ownership. From my perspective, there are three big pillars of ownership when it comes to AI.
The first is the data that goes into training different models, so the responsibilities of ensuring data provenance and the ethics and guidance related to the data. That is one piece.
The next is the output of the models. As Glenda said, that is the responsibility of the institution, the person or the business putting that model into the world in a variety of ways to understand its outputs and be able to describe them and have accountability for them.
The next layer is the ownership of the governance. Unlike the first two, this is where there’s shared ownership between industry, between government, and then the research coming into it to inform those two kinds of elements between formal systems and then internal systems of governance for that shared ownership piece. I wish there were a simple answer, but this gives you more insight into the layers that come into this conversation.
Senator Senior: Thank you very much. I really appreciate your testimony and presentations. I thought they were excellent.
I want to go back to the trust issue mentioned earlier. I think, Ms. Crisp, you said 20% of people are prepared for AI change. That leaves 80%, which is a significant number. What got us to 20%? Also, what are the things that people do not trust about AI? I’m assuming that you’ve done some research. You have a sense of this. You have some statistics and notes about the 20%. Did that research reveal anything in terms of the issue of trust here?
Ms. Crisp: We can provide the full study later, but one of the key points is whether Canadians are using it regularly. Only 45% of Canadians use AI routinely and regularly, versus 55% who never engage with it. When you never engage with it, you are basically in that 5% who don’t trust it at all, which then brings down numbers in terms of people feeling ready.
We don’t have a consistent rollout of AI literacy programs in schools across the country, and that is also driving this. We can also provide more data on that and pull the study for you.
Ms. Enders: There are kind of two big pieces when it comes to AI and its general use. There’s AI that people interact with, and that becomes shorthand for what AI is. There’s a lot of AI that is within industrial systems: There might be a human in the loop or it might be an autonomous system that a member of the general public might not understand is made possible or better or has a risk in it related to AI.
One of the things we’re seeing — again, not to be the broken wheel on literacy — is that the shorthand around people who have greater trust is they are folks who are using it. They are using one very specific kind of AI that has come onto the market in a huge tidal wave. I think that’s the piece of why AI literacy is so important: there are many ways that this technology shows up in people’s lives. Some of it is very explicit, like interacting with a ChatGPT kind of tool, and some of it has been in their lives for a number of years, but people can’t identify it as AI.
Senator Senior: I want to pull on that a little more because my focus has been on public education and the seeming lack of it. I haven’t seen any. Could you address that? If trust is an issue and take-up is an issue, what is being done in terms of public education? Can I start with you, Ms. Crisp, and maybe go to the second round if there’s not enough time?
The Chair: Very short answers, please.
Ms. Crisp: Actually, I will refer to Stephanie because I think Amii has the best program.
Ms. Enders: I can start from the littlest citizen to the oldest citizen. So what we do on the K to 12 space is run a two-year pilot where we integrate AI literacy, again, based on those world-leading practices, into existing curriculums on the ground for teacher resources. So rather than focusing on what should be a coming big push on developing the curriculum, it is how we have AI literacy available in all subjects. It’s not for encouraging students to use AI for everything. It’s asking this question: What are the critical learning objectives appropriate for their age related to those four competencies? We’ve done that directly with teachers and school boards for the past two years. We have about 6,000 teachers and 95,000 students impacted by that.
Next comes the PSI layer, formal PSI for post-secondary, and wonderful diploma programs, undergraduate programs and master’s programs in computing science. We look at all of those other faculties where we know there will be a lag around how AI impacts those domains. So, with private funding, we use a national consortium model, where we’re working with 54 PSIs and really bring our machine learning expertise to new readings, activities, assessments and labs in those areas —
The Chair: Thank you, Ms. Enders.
Senator Arnold: Thank you both for being here. We keep talking about literacy, and you are the most literate. I think that your communication skills on trying to explain what you’re doing are absolutely excellent.
Thank you, Stephanie, for that clarity around the 55% who never engage. I would argue there are people who are engaging but, as you said, don’t even know that they’re engaging. I used to run a city, and we used it in our parking lots. It’s everywhere, all around us, and we’re not always aware.
Have either of you done studies on how other countries have increased literacy around this? It sounds great, what you are doing at Amii, but it sounds as if it’s a much larger issue. How did we get to a point where it’s not being accepted in the same way as it is, perhaps, in other countries?
Ms. Enders: There are a number of different models. In some of the places where we’re seeing, maybe, faster rollout of some elements of AI literacy, especially for children — we see it in Singapore, Malaysia, et cetera — it is related to a national approach to education. So there are structural challenges or differences in Canada that make that different.
The other is that for K to 12 education, international standardized testing is a big driver. I believe we’ll see the first international testing related to AI literacy coming out in 2029, with results to follow in 2030-31. I believe that’s Grade 10 literacy on identifying AI-generated content. So, again, it’s not the fulsome conversation about AI literacy, but it’s the pace that education works at.
On the workforce side, there are a number of different models. France and the Netherlands have kind of different approaches based on their needs. There is a massive open online course. There are good things and hard things about generalized education that way, but that’s what we’re seeing.
Then the other piece is just the way that we talk about AI literacy. There are some discrepancies across the reports. Sometimes that means how adept you are at using a tool versus how well you understand the technology. There is still research to be done on both those fronts.
Senator Arnold: I am curious if some of the fear for Canadians comes from the fact that so much of this isn’t developed in our country. Do either of you have opinions or have you done any research on AI as public infrastructure and what we could do to own our sovereignty more? Would that impact things?
Ms. Crisp: We haven’t looked at that from a research perspective. I’m not sure I would agree with the hypothesis because a lot of AI is actually invented here in Canada. As I said, we rank third in the world, so I think it is more a case of how to translate that so that everyday people like me understand it and feel more trusting of it.
Senator Muggli: Thank you to both of you for being here. I will start with Ms. Crisp. Changing gears a little bit, can you tell me a little about how start-ups can balance speed and innovation with the need to build responsible and compliant AI systems from day one? From your experience — if you have experience in the start-up area — are there areas where AI start-ups are actually ahead of larger organizations or countries in adopting responsible AI practices? If so, what might we be able to learn from these responsible practices in terms of putting guardrails in place? I will start with you, Ms. Crisp.
Ms. Crisp: We actually have a specific program at Vector called FastLane, which is specifically for start-ups, scale-ups, and small- and medium-sized enterprises. In some of those cases, we work directly with them; our AI engineers will work with the start-ups. We’ll also, in some cases, second resources to them so that we actually help them oversee how the product is getting built so that we can be assured it is built safely, responsibly and ethically. Vector has its own AI trust and safety principles that were among the first in the world to be developed.
So we will actively and directly engage with start-ups. We provide a fair amount of content generally to the start-up community so that they can find ways to do this, but in some cases, we actually dig in and do hands-on work with them to ensure this happens.
Senator Muggli: Are we leading in the world in this regard with some of our start-ups?
Ms. Crisp: I don’t know if I would say we are leading. We have a growing start-up ecosystem. I think capital becomes an issue, as well as compute, as I was saying. So we certainly do have large tech partners that will give compute credits, if you will, to start-ups, which is helping, but we really need to bolster that infrastructure so that we maintain those start-ups here and they don’t leave for the south.
Senator Muggli: How could the government help start-ups in this regard?
Ms. Crisp: It would be really great if we could actually spend the money that was promised on compute.
Senator Muggli: Thank you. Ms. Enders, do you wish to add anything?
Ms. Enders: Sure. I will echo that. We have a program called Level Up. It’s different than FastLane. It’s specifically chief technical officer, or CTO, coaching. So the most advanced AI person at a start-up has dedicated time with machine learning scientists to tackle really hard questions like the ones around responsibility.
One of the things that’s super important when it comes to creating many more AI start-ups — and also identifying start-up champions who can scale to the world — is compute capital and customers. Responsible AI practices built into the technology side make Canadian start-ups more attractive to capital and make it easier for them to secure large-scale customers.
Procurement within Canada to Canadian agencies has very high standards regarding compliance. So when you’re asking how to help start-ups be the best in the world regarding these trust and safety principles, the way that you describe that to a start-up founder is, “This makes you more desirable to customers and to capital.”
The Chair: Thank you.
Senator Cuzner: I had a couple of questions, but your response to the last question sort of prompted a change in mine. Could you expand on the inability to spend the money they had allocated for compute?
Ms. Crisp: So the $2-billion sovereign compute that was allocated in 2024 — there hasn’t been meaningful spending in that space. It is going, but it is going extremely slowly. The challenge we have is that this compute is becoming increasingly difficult to get your hands on, and not only that, it is increasing in cost. For example, in the course of roughly three months, some of the infrastructure we wanted to purchase went up in cost by 30%. I’m not sure where the sticking point is, but the request for proposals, or RFPs, are not moving fast enough. The compute is not getting in the hands of researchers fast enough.
Senator Cuzner: Do you have anything to add to that, Ms. Enders?
Ms. Enders: No. I think Glenda summed it up well.
Senator Cuzner: You mentioned trust, and trust has been a common thread that has run through a lot of the testimony of people who have presented. Is it mainly because of the lack of a rigorous governance structure around the sector? Is that where the lack of trust comes from for investors?
Ms. Enders: When we talk about trust in AI, there are a few things. Governance plays a part. The sort of pace of guidance from formal structures is part of it.
I think the other piece is that this technology is many different things and can have a lot of different impacts. Honestly, the way that most people have interacted with AI has been through cultural references. This is one of the first sort of general‑purpose technologies where people have been exposed to what they think this technology is for most of their life before they interact with it.
On formal governance structures, there are a lot of questions like, “Will this lack of structure slow down my business or slow down my start-up?” Those are legitimate questions on that side. But on the general adoption side of AI, it is questions like, “What are my protections?” and, “What are my courses for action?” and, “How do I start to understand this technology to advocate and have agency in my life around it?”
Ms. Crisp: I would point to the banks, who were early adopters in AI; they are global leaders in AI adoption. However, the banks also had well-established model governance practices because their quantitative models had been under scrutiny for decades. They had some of the building blocks already in place, so when AI started coming a dozen years ago, when I was still in banking, it was more a question of how to adapt model governance for AI and not how to build AI governance from scratch.
Senator Cuzner: Thank you.
Senator Boudreau: Thank you both for being here. Maybe I am the downer in the bunch, but I’m still having a hard time embracing AI, though I know it is coming rapidly. I hear both of you talk about how we are world leaders in AI research and have the godfathers of AI in Canada, yet we are dropping globally from fourth to eighth, we lag in adoption and deployment, our strategy is two years behind schedule, and we’re not investing in compute as we’re supposed to. What does it say about the need for robust regulations and guardrails? That’s the piece I find we are not talking enough about.
Everybody wants to lead the charge with AI, but I feel there is not enough talk about what we should do to put proper guardrails and regulations in place. Can we even, as a government, keep up with an industry that’s growing and developing so quickly? Can government even keep up in terms of trying to put regulations in place to protect citizens?
Ms. Crisp: On the regulation front, let me go to the basics. AI is based on data. Without good data, you don’t have good AI. I would say a good place for us to start is to really refresh our data privacy laws. I could be wrong, but I believe our Privacy Act dates back to 1983. The Personal Information Protection and Electronic Documents Act, or PIPEDA, is from the early 2000s. Clearly, a lot has changed in that time. We absolutely need to look at our data privacy laws and data use because the way data is used with AI was not contemplated back in 1983. If I were to start anywhere, I would start there.
In terms of regulation, generally, we have to balance safeguards with innovation. You have to be careful on both sides, but I want to leave Stephanie a few moments to respond.
Ms. Enders: Earlier, I said I’m a hopeless optimist and a 10 out of 10 on opportunity. Part of the conversation that gets left out regarding wide-scale AI literacy and AI fluency, as well as the pace, is that we are focused on making sure people have the understanding so they know when to say no and when AI is not the right tool.
With regulations and formal systems, this space does move rapidly. However, I think the things that don’t move rapidly are the elements related to critical thinking, data and our understanding that a big part of AI literacy is knowing when to say “not AI.” It is not a sneaky tool or a lever we’re trying to bring into society so people just say yes all the time. It’s really around building the capacity of people to have informed conversations, so the balance between regulation and innovation is a really fulsome discussion around concepts that are well understood by everyone around the table.
Senator Cuzner: Very quickly, if there were one country you would take regulations and guardrails from and apply to Canada, which would it be?
Ms. Enders: It would be challenging to say that there is a one-blanket solution. The most important thing around regulation in Canada is that it should reflect Canadian values.
Senator Petitclerc: Thank you so much for being here and helping us with the study.
I’ve been asking questions about our youth since we started this study. Usually, I’m asking questions on safeguards; I’m a little worried. However, I want to spin it in a different direction on the aspect of youth, opportunities and a possible lack of equity in those opportunities.
Ms. Enders, you were talking about literacy, competency and fluency. My short question is this: Is there a risk, if this literacy is not offered to all our youth, that later on, when they are old enough to be active and working, there will be socio-economic inequity and then a difference in opportunities? Is that something we should worry about?
Ms. Enders: We think deeply about this, and we think equitable access to AI literacy is very important. We also think that making sure there are pathways for cultural and regional localization to have that literacy show up in classrooms is vitally important. It’s something we do daily.
We think about how we serve equity-deserving groups as we move forward and about our lens check on that as we develop pieces. We think about our pathways so we can really hear from the communities we serve. What is that unique element related to this that can bring those literacy fundamentals to life in meaningful ways?
We’re also seeing that the nation has a very robust work‑integrated learning program at the post-secondary level. That has definitely informed the work we do related to making sure there are equal opportunities for students across different parts of that AI stack.
Senator Petitclerc: Thank you.
Ms. Crisp, did you want to add to that? Do we have time for that?
The Chair: Yes, we do.
Senator Petitclerc: Quite plainly, if someone doesn’t get that literacy, are their chances in life later on different from those of others? Is that a risk?
Ms. Crisp: That is a risk. Increasingly, the jobs of the future will go to people who understand AI and how to use it.
It absolutely is a risk, which is why I should be on the sales team of Amii, because I keep pushing their K to 12 education everywhere I can.
Senator Petitclerc: Thank you.
Senator Osler: My apologies for coming in late.
My question may be a little out there, but I’m still going to ask it. I recently had a meeting with the Deaf Wireless Canada Committee, and we talked about AI. This committee represents the deaf community, the deaf-blind community and the hard‑of‑hearing community. These communities are culturally and linguistically distinct. They see AI systems and technologies as a real opportunity for better inclusion in society as a whole.
They are particularly interested in how their languages would be represented, how their identities would be perceived and how their rights would be respected.
My question is for both of you. In terms of what you’ve seen and experienced, how are these different communities — deaf, deaf-blind and hard of hearing — being involved in the co‑creation and co-development of AI technologies and systems?
Ms. Crisp, I’ll start with you.
Ms. Crisp: I’m blanking on a name, but I believe there are some start-ups that have been in the deaf, visually impaired and blind spaces, but I’m not personally aware of large-scale involvement. I do think it would be valuable. I actually believe that the more diverse the group of people who are building AI and creating AI, the better the AI will be. To put a finer point on it, I have consistently said that I don’t want AI that is built by all White men. That doesn’t represent me or my neighbours, et cetera. The more diverse we can get in terms of the people building and creating it, the better off we will be.
I’m certainly going to take a note to go back to the branch here and see how we’re engaging those communities.
I will leave time for Stephanie.
Ms. Enders: Similarly, I don’t have a wonderful use case that I can pull from, but I can speak a bit about what we have worked really hard at over the past five years at Amii, and that is helping build interdisciplinary teams. All the work we do at Amii involves ML experts and team members who have built up their muscle around being translators between different communities and finding the pathways for meaningful dialogue when building AI tools.
One piece when it comes to huge opportunities but also areas of risk is real conversations between domain experts who have access to data about what the data means and how it came to be. What is the cultural legacy of this data, and what are we comfortable with it being used for?
Also, what is my ownership of that such that I might have new business models that might benefit my community, town, city or family?
Then, it’s finding pathways to change the challenges we face into ML problems, which are highly structured problems that have all these different inputs and outputs.
So, our work — not necessarily with the deaf or hard-of-hearing communities but across other communities — has been incredibly rewarding and impactful. I think of a researcher on our team named Patrick Pilarski, who is working on Bionic Limbs for Improved Natural Control, or BLINC. We look at how we can work with a specific agency to do that.
Senator Hay: What do you want from the upcoming AI strategy that would make a difference?
Ms. Crisp, I’ll start with you.
Ms. Crisp: I’d like to see continued funding of the Canadian Institute for Advanced Research, or CIFAR, AI Chairs, and at a greater rate than we have in the past, because we are competing with Europe, not just the U.S. I would like to see increased funding for the work we do to drive adoption with small- and medium-sized businesses.
Those would be my top two things. I will leave time for Stephanie.
Ms. Enders: I echo Glenda. Also, we saw eight things in the task force and very wide public engagement. I would like the report to show pathways and organizations like ours that are doing this work so the interest in the strategy can follow on with whom you talk to if you are interested in AI adoption, have an idea for an AI-first start-up or are looking at bringing literacy into your community. There are trusted partners, and we have learned a lot along the way. I think the strategy including those eight themes and what is working well and whom you can go to is important. I think the public is hungry for the strategy to direct the work but also give them something tangible right away on where they could go.
Senator Senior: What would be the top one or two guardrails each of you would recommend for Canada?
Ms. Crisp: I’m going to repeat that I would focus first on data privacy and data rights because it underpins AI.
Senator Senior: Data privacy and data rights. Thank you.
Ms. Enders: I echo Glenda.
Senator Senior: Thank you.
Senator McPhedran: I want to close with this question: If we don’t follow the path that you’ve both so enthusiastically proposed and promoted, what are the implications for Canada and for Canadians?
Ms. Crisp: The World Economic Forum’s study last year on jobs said there are going to be 170 million new jobs related to AI and 92 million displaced. That is a net gain of 78 million.
I want to see Canada be on the winning, net-gain side of that. If we don’t get busy adopting, we are going to be left behind and will have more displaced jobs than gained jobs.
Senator McPhedran: Thank you.
Ms. Enders?
Ms. Enders: On my side, it is a similar piece, especially on scaling start-ups. That is one of the places where we could have a mass impact. If those are not AI-first start-ups, they will not be able to scale to the world to offer Canadian AI built on Canadian values to a global partnership of middle power.
The Chair: Thank you very much. This brings us to the end of our first panel. I would like to thank our witnesses for your testimony today.
Senators, joining us today for the second panel, we welcome via video conference, from Abundant Intelligences, Jason Lewis, Professor and Principal Investigator; and Gideon Christian, Research Chair in Artificial Intelligence and Law, University of Calgary.
Thank you both for joining us today. For your opening statements, you will have five minutes, followed by questions from committee members.
Jason Lewis, Professor and Principal Investigator, Abundant Intelligences: Thank you, Madam Chair, and good afternoon, honourable senators. Thank you for the opportunity to speak with you today.
I am Professor of Design and Computation Arts at Concordia University in Montreal. There, I co-direct Aboriginal Territories in Cyberspace, a research group that explores how Indigenous Peoples use computational technologies. I myself am Hawaiian and Samoan.
I am here today as the Principal Investigator for the partnership for Abundant Intelligences. Abundant Intelligences is a Canada-based, international and Indigenous-led research project that connects researchers, Indigenous communities, technologists and policy leaders across Canada, the United States and New Zealand to reimagine artificial intelligence through the lens of relationship, reciprocity and responsibility.
The project considers how AI innovation can support diverse ways of knowing and living. It does this by exploring how to integrate AI within Indigenous knowledge systems to support Indigenous communities. It then applies its research findings to help mainstream AI research innovate and improve AI for everyone.
I want to be clear that I am not representing any specific Indigenous community. I am here to represent our research activities and the results.
Mr. Hinton was asked by this committee about his call to foster a maternal instinct in AI. Back in 2018, several of us now in Abundant Intelligences wrote an article called “Making Kin with the Machines” that made a similar point. We drew on Indigenous practices of making kin with non-humans — animals, trees, stones and rivers — to argue the urgent need for us to think about AI in relational terms. What kinds of relationships do we have now and what kinds of relationships do we want to have in the future as AI becomes smarter, more capable and more autonomous?
My colleagues and I believe that Canada’s next AI strategy must move beyond a narrow focus on computational power and toward relational leadership. We imagine AI technology that strengthens self-determination, community well-being and environmental stewardship. This includes extending the notion of national AI sovereignty to include knowledge self‑determination and the ability to define what “intelligence” means within Indigenous and Canadian contexts.
This entails building AI that reflects Indigenous and Canadian values, languages and world views; supporting Indigenous data governance that ensures communities retain control over their data, then extending those frameworks Canada-wide; and embedding cultural and ethical sovereignty into AI infrastructure so that public investments in computational resources reinforce democratic control.
This approach draws on the lessons Indigenous Peoples have learned in our long struggles for self-determination to demonstrate how sovereignty can include computational innovation, technological independence and cultural agency.
The majority of our research team is Indigenous and comes from communities whose intelligence was denied for centuries. Our ancestors were labelled as less than intelligent and, therefore, also less than human. Because of this history, we do not take the definition of “intelligence” in the term “artificial intelligence” lightly.
We fear that the lack of critical consideration of the full spectrum of human intelligence will, once again, elevate some communities’ knowledge systems over others.
Our core research questions include these: Who gets to define intelligence? Who gets to validate knowledge? Who gets to use that knowledge?
We work with our collaborators in Haudenosaunee, Anishinaabe and Niitistapi communities here in Canada and with Lakota, Hawaiian and Maori communities abroad to answer those questions.
The questions they ask include the following.
How can we take advantage of the enormous power of AI technologies while protecting our cultural knowledge from the AI industry’s infinite appetite for stolen data?
How can we implement data sovereignty principles like OCAP — ownership, control, access and possession — into the core fabric of our AI systems?
How do we train our youth so they can be active participants in this technological revolution?
How do we reconcile the enormous environmental cost of AI with our roles as territorial stewards?
Because of our research, we think that Canada has a deeply transformative opportunity to take a lead from Indigenous communities to think about AI in a relational manner.
We must ask these deeper questions of how we want to live alongside this technology as it continues to grow in power and impact.
Thank you.
The Chair: Thank you, Professor Lewis.
Gideon Christian, Research Chair in Artificial Intelligence and Law, University of Calgary, as an individual: Madam Chair and distinguished senators, thank you very much for the invitation to appear before your committee today to contribute to your study on the impact of artificial intelligence in Canada.
My presentation is going to focus on the impact of AI on the Black community. I will start my presentation with a real-life example.
In late 2022, a young Black man by the name Randal Reid was pulled over by the police in Atlanta and arrested. He was detained for six days. He was allegedly involved in the theft of handbags from a luxury store in Louisiana. But there was a problem here: Mr. Randal Reid has never been to Louisiana and could not have stolen luxury handbags in a place he has never been to.
Why, then, was he arrested? He was arrested because it later emerged that AI facial recognition technology erroneously matched his face to a suspect. That error by the AI system was sufficient to deprive Randal of his liberty.
While it could have been worse, Randal’s experience is not an isolated incident. In the United States, there are well‑documented incidents of wrongful arrest arising from the use of AI facial recognition technology by the police. The overwhelming majority of the victims are Black individuals. This is not coincidental.
Facial recognition technology has been shown to have its highest error rates in identifying Black faces, particularly Black women, with some studies indicating error rates as high as 34% for Black women compared to 0.8% for White males. This is not merely a technical issue; it is an equality and civil rights issue.
Over the past few years, Madam Chair, my research has focused on identifying how racial bias is embedded in AI systems deployed across multiple sectors.
In the immigration context, I have studied Federal Court of Canada litigation challenging the use of evidence from AI facial recognition technology in the revocation of refugee status of successful refugee claimants in Canada. Almost all the cases involve Black people.
In the criminal justice system, I have researched the use of AI in recidivism risk assessment, which is to determine the likelihood that individuals in the criminal justice system will reoffend. This determination is important for the purpose of determining bail or certain sentences for convicted criminals. My research in this area has consistently shown that AI tools used in assessing this risk overpredict the risk of Black defendants, doing so often at twice the rate as with their White counterparts.
In predictive policing, AI systems are used to predict the geographical location where the next crime is likely to occur. However, AI tools used in this area are trained on historical policing data shaped by decades of over-policing of Black and Indigenous communities.
As a result, these AI tools learned patterns of police presence as opposed to actual patterns of crime in the community, inevitably predicting the next crime to be in those marginalized communities that are disproportionately represented by the training data. This results in further policing of already over‑policed Black and Indigenous communities.
In the medical field, AI tools train on historical health care data generated within systems shaped by long-standing racial inequality, including practices such as “race norming.” This eventually replicates the historical bias buried in the training data, resulting in systemic underdiagnosis of health conditions for Black patients.
Racial bias has also been evidenced in tools used in human resources and hiring, which screen out résumés of Black applicants based on proxies of race such as names and area codes. Similar race-based bias has also been shown in AI tools used in the financial services sector to assess the creditworthiness of loan or mortgage applicants.
Many AI systems are trained on data generated within societies marked by past historical and ongoing inequality. Without deliberate intervention, these AI systems will rather insidiously encode or amplify these problems rather than correct them.
It is imperative, therefore, that we critically examine the AI tools we adopt and their impact on marginalized communities in Canada. This is because, as a country, Canada has come a long way in seeking to rectify the racial injustices of the past. The lessons from the past must guide our path forward.
We must ensure that our technological advancement contributes to a more just and equitable society. To this effect, I will offer three recommendations.
First, Canada should adopt a precautionary approach to the deployment of AI systems in high-risk environments where the consequences of error are severe and would further marginalize communities that are already disproportionately impacted.
Second, AI systems that have been shown to exhibit differential error rates along racial lines must be subject to rigorous risk impact assessment prior to their deployment.
Third, Canadians must be guaranteed the right to transparency and meaningful recourse, including the ability to know when AI is used in public sector decision-making processes that affect them to understand how the decisions are made and to challenge those decisions.
The question before us today is not whether Canada should continue to adopt AI technology, Madam Chair, as it must —
The Chair: Thank you, Mr. Christian.
Mr. Christian: Thank you.
The Chair: We are going to move now to questions from committee members.
For this panel, senators will have four minutes for the question, and that includes the answer. Please indicate if your question is directed to a particular witness or all witnesses.
Senator Burey: Welcome to our witnesses, both here and virtually. This testimony is so compelling. Thank you for coming and sharing it with us today.
I’m going to start with you, Professor Christian, but this is for both our witnesses. You talked about how we have to be deliberate in our interventions. What are the legislative and regulatory guardrails we need to put in place?
Mr. Christian: Right now, we don’t have specific legislation that governs artificial intelligence in Canada. The regulation of AI in Canada today is guided by legislation that was drafted even before AI. We are basically overstretching that to try to cover this situation. That’s not the best way to regulate AI.
“AI” is a very broad term that covers various aspects of this technology. When we talk about regulation, it is often important to understand what aspect of AI we are seeking to regulate. Regulation should start with high-risk AI tools or AI tools that are deployed in high-risk environments. Some AI tools are basically cautious in their use, for example, the one I use to unlock my phone. The adverse impact of using that is not the same as AI tools being used in the criminal justice system by police or correctional facilities.
Having Canadian legislation that regulates the use of AI is overdue.
Mr. Lewis: I am not a legal scholar, so I can’t speak specifically to the existing legislation in place. I can speak to the great interest on the part of the people with whom we are working on the research team in thinking through how we take principles like OCAP and create standards out of them — so that those standards can be used to both shape and evaluate the AI systems we’re encountering, particularly if they will be used in and with Indigenous Peoples and Indigenous communities.
Senator Burey: Thank you.
Senator Hay: Thank you both for being here. It’s very compelling and important testimony. This is for both of you. Yesterday, the Minister of Artificial Intelligence, Evan Solomon, made a comment around AI, referencing it as artificial intelligence but also ancestral intelligence. He thought of it as both. I loved hearing that; however, I want to know what that means to both of you in the work that you do. Also, in the build for the pending AI strategy, there has been, I heard, record consultation, though I don’t know the specifics. I would love to know what the consultation has been with First Nation, Métis and Inuit communities as well as Black, African and Caribbean communities in Canada.
Mr. Lewis: I am very interested to hear that the minister said that. This is a term that’s used in many Indigenous communities we work with to talk about knowledge passed down to us by our ancestors and also ways that knowledge is expressed, which are not necessarily ways that are particularly legible to Western science — for example, through various cultural practices that actually encode and embed data about things that have happened in the past to the landscape, the territory, the people, et cetera.
One way people we work with who work on Indigenous AI talk about that is as ancestral intelligence. I am curious. Maybe that’s where he picked that up. What’s really important there is an understanding that there is really no readily agreed-upon definition of “intelligence.” Certainly, from a lot of research, there are many kinds of intelligence. It makes a difference who is asking the question and who is evaluating what an intelligent action is. I say that because what one culture or context might deem to be highly intelligent — for example, amassing hundreds of billions of dollars of wealth — may be considered in other cultural contexts and other communities to be the height of selfishness and stupidity.
Mr. Christian: When it comes to my work in this area, I started initially as a technology lawyer with the Department of Justice. I was deploying this tool in providing legal services in high-profile litigation involving the Government of Canada. I had a change of heart when I read a report by a U.S. think tank by ProPublica entitled Machine Bias. For the first time in my working in this area, it allowed me to understand that AI could be racially biased. I had initially thought it was objective and barrier-free. That was what led to my change in career from legal practice to academia.
I joined the University of Calgary in 2019 as an assistant professor and was promoted to associate professor and Research Chair in AI and Law in 2014, enabling me to research this issue. In almost every sector, I have looked at where AI tools have been deployed, basically, searching for racial bias, and you always find it: in health, financial services, credit or mortgages. To me, that is disturbing.
With regard to consultation and the attempt to regulate, I don’t know how much time I have so please let me know.
The Chair: Please wrap up as quickly as you can. You have a few more seconds.
Mr. Christian: Bill C-27 was the bill that was meant to regulate AI in Canada. That bill went through committee here, and there was not a single mention of racial bias. I was so concerned that I had to write a letter to the chair of the committee during that to get their attention, but that bill died because of the prorogation.
Now we have a new Minister of Artificial Intelligence who has now appointed the AI Strategy Taskforce. The original members of the task force were 27 experts. When that committee was appointed, there was not a single Black person on it. I coordinated with 60 Black scholars across Canada to write a letter to the Prime Minister and the Minister of Artificial Intelligence bringing that to their attention. After our letter was written, a Black student with no background in AI was appointed to represent the Black community on that committee.
The Chair: Thank you.
Senator McPhedran: I will precede my questions to both of you with an observation that we are very grateful that you are here and bringing these perspectives. It was a little upsetting when one of our previous witnesses said she would start looking at who was involved in the work they were doing. We are hearing from you that this is definitely a pattern.
My question is this: If each of you had to choose one first step from where we are today, would it be to ban certain uses of AI? Would it be to legally define and enforce a duty of care? Would it be a liability or accountability regime?
Mr. Christian: That’s a very interesting question, and it’s a big one too. Almost everything you have said so far is part of the solution. It is hard to say which one to start with because everything there is needed.
But when it comes to regulation, I always emphasize starting with high-risk environments because those are the most concerning environments for now. I would strongly argue for regulations starting with high-risk environments, and when I talk about regulations for high-risk environments, I mean what we use in a legal context as a strict liability.
Why do I say this? I don’t know how much time I have, but I will illustrate my answer with an example. AI tools are designed. The designers of these tools are good-faith designers, putting in their best to make sure things check out in terms of the tools’ safety before they’re deployed. But the thing with AI tools is that the ones that leave the factory are often different from the ones that actually start working. That’s because AI learns from data, so after you take all the precautionary steps and design a marvellous tool, once it goes into operation and starts to learn, it may start exhibiting characteristics that were not in evidence when designed. I will give you a very good example.
Senator McPhedran: I am very interested, but I also want to make sure we have time for the other witness to respond, please. I would love to circle back.
Mr. Lewis: I agree with what Mr. Christian said in terms of needing a whole bundle of those things.
But one area where I diverge is that I do not have faith that the designers of these systems are acting in good faith. Even after they were shown the biases that were being built into the systems and that they were training their systems on stolen data, they did not change the way they operated.
Again, as Mr. Hinton said, capitalism is the spectre behind all of this that is driving uptake and development, and it’s not particularly compatible with safety or the concerns for the duty of care that we’re talking about.
We are focused on building capacity in the Indigenous community to create AI scientists and engineers. We want to have Indigenous people who are part of the creation of these technologies and innovation around them. We want to start getting our viewpoints and our ways of thinking about these technologies in at the beginning, instead of having to be in this constant game of whack-a-mole where we have to react to what other people build for us.
Senator Muggli: Thank you for being here. I will carry on with Mr. Lewis, and it might segue into what you were going to talk about.
You are now about halfway through your research, I think. Since the short time ago you started, in 2023, how have you seen Indigenous use of AI change or AI adapt to Indigenous ways of knowing?
Mr. Lewis: A big part of what we did in the first part of the project was to thicken the relationships with the Indigenous communities with whom we went into the project and the relationship.
It was very interesting listening to the previous panel. A huge amount is about literacy, talking to them about what AI is and isn’t, but within an Indigenous cultural and historical context, with a lot of history of it being done badly by Western technology; so there is that increased critical lens.
A lot of the movement so far has been in establishing basic literacy around that and making space with them to dream and think about what kind of technology and AI they want, instead of having to react to what is given to them.
One big change from when we started and now is this: When we started, the questions were, “What is AI?” People heard bad things about it and were not sure they wanted to give it their cultural knowledge.
We got through that with our communities, but now it’s very much like, “Wait a second. We’re supposed to be stewards of our territory.” What they’re hearing now is that AI is wrecking their environment. They are asking how they can they participate in the creation of this technology if this technology will wreck their territory.
The environmental questions are really central and on their minds because of our particular obligations to the lands on which we live.
Senator Muggli: That is very interesting. Thank you.
I have a quick question for Mr. Christian. I was just made aware recently of an article that you wrote about boycotting the government’s AI Strategy Taskforce. You may have already answered this in various answers, but is there anything you would like to share with this committee that you would have wanted to share during consultations with the task force?
Mr. Christian: Yes, there are a lot of things.
Senator Muggli: Could you maybe share your top two?
Mr. Christian: My top two? As a Black person, I think I would probably start from home first. Racial bias in AI is a very big deal because it’s not something that is openly discussed out there. It’s not even something that is openly evident yet.
I was working in this field with a PhD as a technology lawyer, and I was not aware of the fact that AI can exhibit racial bias until I read the particular article I mentioned.
So this is not knowledge that is common out there, but it is something that really happens. Black and Indigenous communities especially have been disproportionately impacted by our justice system, and data obtained from that disproportionate policing and criminal justice conduct is used to train AI systems.
It basically regurgitates the historical biases we are seeking to overcome. We’ve had the Truth and Reconciliation Commission, and we’ve had the Black Justice Strategy basically trying to address this ugly past. To now have AI bring it back in an insidious way that we cannot see is something we should try to avoid.
Senator Muggli: Thank you.
Senator Senior: I would like to pick up where you left off, Mr. Christian. What do we do? I understand where Professor Lewis is. We can’t stop what is happening in terms of AI. The opportunity to fix it is a longer-term strategy. What do we do in the interim?
Mr. Christian: In the interim, we need to design regulations and policies with everyone impacted by these tools on board.
AI tools are designed predominantly by White males, and that’s why we talk about facial recognition technology having a 0.8% error rate when it comes to recognizing that particular racial and gender group; it is 34% for dark-skinned females because they are not represented.
So when it comes to regulating these tools or drafting policies and legislation, it is good to hear from everyone impacted.
And, of course, this means —
Senator Senior: What if they’re not doing that? I’m particularly concerned about the three institutes that are doing research, representing, consulting, being heard and being funded. Is there any connection with these research institutes as it concerns the issue of facial recognition?
Mr. Christian: There is very little research being conducted in Canada on this issue of the impact of AI on racial minorities.
We have many Black women and Black technology researchers in the U.S. working in this area, but Canada seems to be different.
What is of even more concern is that most of the people working in the U.S. on this are actually Canadians. Some of them actually went to school here. When they finished here, they had the expertise and were then poached over.
Senator Senior: That is because they were not hired here. Is that correct?
Mr. Christian: They started here. Deborah Raji has appeared before this committee. Deborah is from Canada; she went to school here in Ottawa and at the University of Toronto.
So we don’t really have that research being conducted here on this particular issue. I happen to be among the very few people researching racial bias in AI here.
So we need to conduct more research, fund more research in that area and identify the issues. It is when these issues are identified that we are able to strategize on how to solve the problem. Research in this area is very important.
Senator Senior: I want to continue with that. Is there funding coming to the Black community to do research? Because I’m not very hopeful that these three institutes — similar to Senator Osler’s question around what’s happening with deaf and blind community members — will have our priorities at the top of their list.
Should that be a strategy?
Mr. Christian: I have not really been successful in getting funding from the three major federal research bodies in Canada with regard to my research in this area. Most of the funding for my research in this area has come from other sources.
Aside from the Tri-Council agencies, I’ve been privileged to get some research funding from the Office of the Privacy Commissioner, which funded my research on facial recognition technology.
I am about to start research on predictive policing, as well as facial recognition technology, because the Edmonton Police Service has just launched a pilot project that basically embeds facial recognition technology in police body-worn cameras, without even conducting a privacy impact assessment. I have been fortunate to get a grant from the Canada Race Relations Foundation to start research in that area.
In terms of research funding in this area, it has not been forthcoming.
Senator Arnold: Oh, my goodness. I’d like to keep building on that. Have you had any success in getting people to listen to you on this topic?
Mr. Christian: Thank you so much for that question. Let me be sincere: I was trained as a technology lawyer, but when it comes to this particular issue, advocating on this issue, I have had to develop and train myself in a new skill, and that is advocacy.
I have performed research in this area. As a researcher, I am really worried and concerned about what I have discovered with respect to this tool and the Black community.
It is kind of disturbing when you voice that concern and people kind of aren’t listening. That is why I really appreciate the opportunity to appear before this committee. I feel this is a very good opportunity for me to advocate around this issue. We can probably find a solution to it, but it is a big problem.
My concern is that, while we have made progress in racial equity as a country, AI is going to reverse it if we don’t address these problems.
Senator Arnold: My goodness. Your third issue on this was the ability to know when AI is being used and the regulations around it. What did you have in mind on that?
Mr. Christian: So government agencies — and the public sector, especially, in Canada — are increasingly deploying artificial intelligence technology. The problem is that most of this is being done with secrecy. There has been a lack of transparency, and this is problematic for many reasons. First, AI itself is a black box. It is a black box in the sense that the developers of these tools don’t even have detailed knowledge as to how the tools operate, which is concerning.
Second, the developers of the tools protect their proprietary information from the public, so the public does not even have the opportunity to scrutinize the black box. That is another layer of secrecy.
Third, when government departments deploy these tools, you don’t even know that they are using them. My research into facial recognition technology started when I got to know, through court litigation, that the immigration department was actually using evidence from facial recognition technology in refugee status revocation. I work in this field, and even I didn’t know. I came to know that through that litigation.
So when you have these layers of secrecy, it doesn’t really build public trust and confidence in using the tools. When government departments or the public sector deploy these tools, it is important for the public, especially the people impacted by decisions made using these tools, to be able to know that decisions impacting them were made by AI. That gives them the opportunity to know whether to challenge that decision and how to challenge it, and that is important. That is part of what we call in administrative law “procedural fairness,” knowing what decision was made, how that decision was made and how you can challenge that decision by a government department, if you have to.
Senator Arnold: Thank you.
Senator Cuzner: I just want to come around on the story that you lead with about Mr. Randal, who was wrongfully accused. I’m sure he had to lawyer up. He went to court and it cost him some money. He was found not guilty. Was he able to press for court costs or damages? Was there any liability on the part of the facial recognition company?
Mr. Christian: Thanks very much for that question. He wasn’t actually charged. He was detained for six days. During the detention, arrangements were being made for him to be extradited to Louisiana before it became evident that this guy had never even been there. That was where the case ended.
But Randal is just one example. There have been many others, including the case of Porcha Woodruff, an eight-months-pregnant woman who was arrested for carjacking.
In most of these cases, with organizations like the American Civil Liberties Union, or ACLU, they have actually sued police departments. Some of that litigation is ongoing; some has been settled out of court. However, none of the technology companies that developed these tools have had any costs awarded against them.
Senator Cuzner: You would think the police departments would be taking that business to the facial recognition companies with which they’ve contracted and pushing them.
Mr. Christian: Can you say that again?
Senator Cuzner: I would think that the police departments should be taking this to the facial recognition companies and pressing them.
Mr. Christian: Do you mean taking them to court?
Senator Cuzner: Yes, exactly.
Mr. Christian: Well, that hasn’t happened, and it is not likely to. In most cases, these tools have been offered to police departments like, “This tool is amazing. We will first provide it to you to use for free.” When that use starts, then the service is that — I’m sorry to use this term, but it’s like a crack dealer who first offers the drug for free and then gets the individual addicted and begins to sell to them. That’s the way the relationship between technology communities and police works. There was the Clearview AI issue, which was a big scandal here in Canada. That is also how that started. Clearview was offering their services for free, saying, “Go test and use it, and then come back.”
That’s why I have advocated for strict liability when it comes to high-risk deployment of these tools, so that not just the user but also the developer of the tool — when you impose this kind of obligation to make them start on their feet, they will monitor this tool and ensure that it doesn’t wreak havoc on the communities that are being impacted.
Senator Cuzner: Great. Thank you.
Senator McPhedran: I will sound a little bit like an echo chamber because I will start where I ended. So the first choice for action — since that’s a big part of what we have to pay attention to — was around a liability regime.
Mr. Christian: A liability regime is important.
Senator McPhedran: Is that the first step?
Mr. Christian: Yes.
Senator McPhedran: You said to do everything possible, but not everything is possible. So would the first step be a liability regime?
Mr. Christian: In high-risk environments, yes.
Senator McPhedran: Mr. Lewis, please — a liability regime or something else as a crucial first step?
Mr. Lewis: I agree about a liability regime. Coming from the Indigenous context, we have to find ways to get things like OCAP written into legislation that would deal with things like that. Because it is not just about the liability on how it’s used on the other end. On the point that Mr. Christian is making really well, it is also about the data. It really starts with the data that’s problematic, yes, and that’s under the control of the developer. It is their responsibility to figure out what the data they are going to use is. How are they going to clean it? How are they going to validate it? There are all these things where there are kind of incentives in the systems to accelerate through those processes.
Again, I’m not a legal scholar, so I am following Mr. Christian’s lead. From my understanding of strict liability, they are not going to pay attention to that stuff on the back end. As we saw in the examples he gave, Clearview and the other technology companies providing technologies that are causing problems with policing are not being held liable at the end of the day. We absolutely need something that holds them liable. Part of that liability is around where they obtained their data. Particularly, if they are dealing with Indigenous communities, did they have permission to obtain that data? Who validated that data? Are they working off a century-old data set that is riddled with biases and prejudices from missionaries, anthropologists or government officials?
Senator McPhedran: Thank you.
Senator Senior: Professor Lewis, I didn’t have time to come to you, but the approach that you are taking is one I truly admire in terms of getting the work done from the ground and engaging the community and the First Nations and others in this. However, when it comes to the use of various AI tools, are you concerned about the same things the Black community would be concerned about?
Mr. Lewis: My colleagues and I are absolutely concerned. It is a huge driver of the project of Abundant Intelligences. In fact, in the essay that I mentioned in my opening statement, “Making Kin with the Machines,” we began writing that in 2017 specifically because of the research that was coming out about the biases that Mr. Christian went into more detail about.
So a huge driver of this whole thing has been saying, “Here comes another technology that will be used against us.” How do we get in front of that? In the kind of research that I conduct, which tends to be long-term research into cultural adaptation of technologies, we recognize that there is a need for legislation now. There is a need for things like strict liability now. There is a need for ways to contain the damage being done now. We sort of take that as a background, then take a step forward and ask, “How do we want this technology to work?” Instead of trying to keep fixing the problems with it, how is it that we want it to work?
In general, we are very focused on that, even though we’re working or in conversation with groups like the First Nations Information Governance Centre, or FNIGC — Jonathan Dewar was here a couple weeks ago — who are very focused on, first of all, really clear political mandates from Indigenous communities. We are not a policy or political shop. We are not mandated in that way. We are Indigenous researchers who are exploring something that we think is of interest to our communities.
But that sort of policy work absolutely needs to be happening at hyper speed. The sort of work that, for instance, the FNIGC does needs to be funded 100 times more than what it is being funded because of, again, the representation issue. These tables get put together and then there are no Indigenous or Black people included. It is the same old faces — the same people asked again and again for their opinions. Frankly, these are the same people who, in some ways, made the mess that we are trying to get ourselves out of.
Senator Senior: Thank you.
Senator Hay: I was writing my question around racial profiling while you were speaking, Mr. Christian, and I have 17 other questions. I’m going to start with a mix of what I heard around racial profiling. I want to maybe add in a bit about data and data governance, which we have all been talking about, as there is no question that AI has unconscious and full-on conscious bias built into it. It is fuelled by data, and that is human data they are collecting. So where are they collecting it from? It starts with data.
Then it’s interpreted, and I’m thinking of the case example you gave. There’s that racial profiling — that individual going about his day and he gets racially profiled, gets called up, and then it is interpreted by humans again.
I guess where I’m going to go is — it starts with human data. I’m asking for your opinion on this because I just don’t know. AI builds in bias, spits out its output and is interpreted by humans who may themselves have bias — in this case, I’m assuming they did.
So I think the problem is more than AI in racial profiling and bias. It starts with data, data governance and our human frailty on racism and discrimination. That’s maybe not the question for today, but it kind of is.
What would you say to us as legislators as we are building laws? We know we need to regulate with high-risk populations in mind first; that is what I heard. But I think it is more than that; it is a bigger problem.
Mr. Christian: It is a bigger problem. It is a problem that starts with data. Data is like fuel that propels AI. So AI is basically fuelled by data. But AI is not a laundry machine; AI is like a sponge. It absorbs the data; it absorbs the beauty and ugliness of the data and basically replicates it by way of its output.
Now, it depends on which sector. Take the criminal justice sector, for example. We have a history of racial minority communities and Indigenous communities being over-policed and overrepresented in the criminal justice system.
Now, the data being used to train AI in this context, in this sector, is from this history of over-policing and overrepresentation in the criminal justice system. When you use it to train an AI tool today, it’s basically going to replicate that past, even though we have tried to sever from the past by developing policies and legislation that will reform that past. But this change has not generated data that will supersede the previous data, so AI is basically replicating that bias —
Senator Hay: Just because of time, what do you recommend putting into an AI strategy or legislation to help work on that?
Mr. Christian: When AI tools are being designed, where the tool or the data relates to previous or historical practices where individuals from racial minority communities have been overrepresented, the designer of the tool should account for that overrepresentation: How does this tool address the overrepresentation? When you take that into consideration, it is not going to absolutely solve the problem, but at least it gives you the fact that the designer is conscious and aware of the problem and is taking steps to address it. We also need to know how they are addressing it so that we don’t have the past being replicated.
Senator Hay: Thank you.
Mr. Lewis: Can I jump in quickly?
The Chair: We are past that question, but I will give you a minute to answer, professor.
Mr. Lewis: I wanted to jump in and say that it is very important to understand a bit of the history of the development of AI. Particularly in the past 20 years, it has come up mainly through what is considered computer science. Depending on the faculty, it might be part of engineering or might be separate.
But we have part of a system in place already: Many of those engineering fields require certain kinds of certification to practise as an engineer. You don’t come out of university with an undergraduate degree in engineering certified to go practise as a civil engineer building bridges. You have to go through additional certification in order to prove that you are doing the sorts of things that Mr. Christian is talking about — that you are actually handling the power that has been given to you in a responsible manner.
Computer science is much more kind of a Wild West. It doesn’t have those kinds of standards or guidelines on a professional level. So something we can think about is what happens in the educational process that reinforces that part of their job as they build these systems is doing that sort of work because, at the moment, that is missing almost entirely.
Senator Osler: My question is for Jason Lewis. I am on the Abundance Intelligences website and see that your first listed research question is:
How can we develop new approaches to AI, based in Indigenous Knowledge Systems, that support the flourishing of Indigenous communities?
To give you some of my background, I come from the world of health care. So many of my questions have been related to AI in health care, its use and how it will be used to help close gaps and prevent harm. As you know, Indigenous health outcomes in Canada have multiple gaps.
We’re hearing about the data inputs that lead to biased outputs. And AI is in use in health care, and I see more Indigenous governments and communities taking control of their health care systems, the delivery and administration, and hence will be using more AI tools in their health care.
In your experience, are these types of discussions about bias, the data and the liability being held by those Indigenous governments and communities as they start to adopt these new technologies?
Mr. Lewis: That is a really good question, and I don’t have a lot of visibility into whether this conversation is happening in Indigenous governance. There are a couple of places where we work with official governmental bodies. However, in a lot of places, we work with non-profit organizations and things like that, so I can’t speak to what is happening in those conversations.
What I can speak to is that we have a couple of researchers in our network who are focused on exactly this question — though none of them are in Canada, unfortunately. We have Professor Keolu Fox, who is Hawaiian but at the University of California, San Diego; and Melanie Cheung, who is a Maori researcher in New Zealand. What they are looking at is an extension of this data problem, meaning that in the same way that a lot of health care products are not tested extensively on a diverse community, and so in some ways the data is not a good match if you are not from the test community. That sort of thing is happening at scale.
Mr. Fox is coming out of genomics and looking at the problematic history in genomics around looking at the genomes from particular populations and then extrapolating that across an entire human population. And Ms. Cheung is looking at how particular neurological diseases that are more endemic in the Maori population are both overdiagnosed and underdiagnosed because the data that the systems they are using were built on comes from a different population.
So I think there is awareness at multiple levels. But, again, I don’t know, and in talking to both of them, it does not feel like there are incentives for companies to rectify that. This is because, essentially, they believe it works for a huge population, and they are not incentivized to collect the right kind of data for smaller populations. Going back to earlier conversations, this is an ongoing issue: Incentives are aligned with collecting data on a vast scale from populations that can provide you data on a vast scale. If you don’t fit within that, you have a problem with the current large language model-driven aspect of AI.
This is also true of language. We have these things called low‑density languages — like Hawaiian, Mohawk and different Indigenous languages — where there is not enough data to use the techniques they use for large-scale systems, so we end up with really bad translation services. Now imagine that but for health care.
The Chair: Thank you very much.
Senator Osler: Thank you.
The Chair: This brings us to the end of the meeting.
I would like to thank Mr. Lewis and Mr. Christian for being with us today and for the time you spent sharing your incredible knowledge. We are very grateful for that.
(The committee adjourned.)