Canada’s AI minister to Big Tech: ‘We Won’t be a Digital Colony’ (Evan Solomon)

Either we invest in building the infrastructure that is needed to this, or we're going to have to rent it from someone else who's built it. Either we are going to invest in Canadians to provide this innovation, or we're going to just have to buy it from someone else. And either we set the rules here about our privacy and our kids and our data, or someone else is going to do it for us. AI for all. That's what the Canadian government is calling its national AI strategy. That is the mission. Trust. Opportunity. Sovereignty. When it first came out, I was a bit skeptical. It made the case for building AI with precise details. A homegrown supercomputer. 60% adoption. All within the next decade. But it said very little about how it was going to govern this technology. But now, the government has tabled two landmark bills that together give Canada something it's never had. A single digital regulator to hold social media and AI companies accountable. I'm talking about the Safe Social Media Act, which regulates social media and AI chatbots, and the Protecting Privacy and Consumer Data Act, which brings our 25-year-old privacy laws into the AI era. Now I need to make something clear. I am not a neutral observer to this. Far from it. I served on the government's national AI strategy task force, and on its expert advisory group on online safety. Which means I've had a hand in some of the thinking behind these policies. And I broadly think the government has got a lot of things right. But not everyone agrees. Go on X and you'll see a wide range of voices. From anonymous bots, to real experts and tech CEOs who say that the government has gone too far. The regulator's power, its design, and its independence are all fair and open questions. And they're what I wanted to ask the person driving this agenda. Canada's first minister of AI, Evan Sullivan. I've known him for a while now. He actually hired me four years ago to make AI explainer videos for GZERO Media. And my own views on AI have shifted a lot since then. I no longer doubt the capacity of this technology. It is going to be transformative. But I still worry whether AI, or the people building it, are trustworthy. This conversation goes pretty in the weeds. I sat through enough of these policy cycles to know that principles are the easy part, and that everything that matters lies in the details. Whether you can force a platform to delete a deepfake. Whether your kid's data gets treated with the same sensitivity as health data. Whether an algorithm can deny you a mortgage. I'm Taylor Owen. From the Globe and Mail, this is Machines Like Us. Here's my conversation with Evan Sullivan. Minister Sullivan, welcome to Machines Like Us. Taylor, great to be here. Great to chat, as always. One of the challenges with this current moment is, I think, the where some of this debate is occurring. And I don't know, you probably don't have a lot of time to open up your XFeed anymore. But let's just say it's not, like, hugely supportive of this new regulation. I open it. I open my XFeed. So tell me what you see when you see the discourse about establishing a new regulator. And I'm not opening up my XFeed to get a representative sample. I'm using it as a data point. X is a data point. I get it. There's a constituency that's very vocal on X. I've got 230 or I don't know how many followers. But it's a platform that I know well. And so I don't dismiss it. I listen. But what you've got to do, in my view, is you've got to really get real data. Like, our social media is based on data. Our approach to AI chatbots, we don't have great data. So we made an approach based on what the data we've got. On privacy, what I am taking care of, the Protecting Privacy and Consumer Data Act, we are trying as hard as we can to consolidate the data from multiple different groups, to balance safety, security, and trust, and to make sure that Canada is open to innovation. Because this is not, and you've heard me say this, and this is not cheerleaders, pom-poms and pitchforks, cheerleaders and pitchforks saying, no, this is a pragmatic thing. I've seen the birth of technology before. I chronicled the internet. That was my job for almost a decade. And it was the very same dynamic. People say, well, this is terrible. We don't need this. And some people are saying, it was neither. And that pragmatism is important. Yeah, that's interesting. I mean, as you know, I don't love that term, the pom-pom and pitchforks line, in part because, and it gets back to this Twitter element or this X element, which is, I actually think it ends up delegitimizing really legitimate views that sit on the extremes of this debate, that get perceived as extremes. And I wonder when you're looking out, I agree, broadly looking at the discourse is critical at the moment. But how do you value views that may be extreme, but in a moment of AI where there's so much uncertainty and we're moving so fast, might end up being right. It might be the case that this bubble bursts. It might be the case that there's an existential risk. It might be the case that this is totally overblown and it's just totally normal thing and you're intervening to curtailing innovation. Those are all relatively extreme views that might be put in this pom-pom or pitchfork framing, but the middle ground might not be right. They might be right. So how do you balance those things when you're looking at all these views? Right. And it's not about dismissing. It's about not picking a side. My job isn't to be a cheerleader or my job isn't to say AI is bad. My job is to recognize pragmatically that this technology is here and we've got to build it productively. Now you are right. How you manage and how you calibrate inputs matters. For example, on questions about the existential question about AI as does it become human? Is it a general intelligence? What happens with the singularity, all these AI moments? And are you concerned about the existential threat? Of course, in the same way that we're thinking about climate change or nuclear weapons, you have to absorb a certain tabletopping of scenarios that are scenario X, Y, or Z. And on the same token, you have to tabletop exercise of what happens if it really does cure cancer. And what if it does? So you're in a tabletop. Or if it doesn't work at all, like if this is just a bubble, which is also possible, right? But this goes to the long view of scenario planning. We are not predicting. This is a question that I get the most, Taylor, about what jobs will be there in the future. Because the jobs question, in a sense, captures both the optimism and the pessimism. Optimism, like, oh, I hope there's a job. Let me just pick the right sector. And on the pessimism, what happens if there's no jobs? Well, to be clear, I mean, 70-80% of Canadians think AI could threaten their job. I mean, that is a real challenge to adopt. So I get, listen, and just to go down from our theoretical discussion, which is why in AI for All, the trust, opportunity, sovereignty, our first investments are, one, in literacy, right? A national AI literacy course. Two, access to a national AI tool for post-secondary students. Three, an investment in young people, creating 90,000 AI spaces with organizations like Mitacs, so young people get paid spaces, job placements, and training. So training, because our view on that is I'm not going to predict what sector is going to have jobs or not jobs. A politician that's in the prediction game is, you know, that's not what we do. We're not a prediction market. We believe that the best hedge against change is education, right? And so we're going to invest in education and training. Education is a passport. Skills are a visa that you're probably going to have to change, and you may. All of us have had skill changes over our lifetimes, and lifelong learning becomes intimately more important. But on that, but just on that point, just to put a little bow on it, you know, this is why we're not here to predict what's going to happen. We are here to invest in skills, education, Canadian tools, and build a framework that we have some kind of sovereign control over, so this stuff works for us, not against us. I mean, I get it, and I think it's, I get the practicality, and I think it's reasonable, and I think it's probably broadly the right approach. I guess I just wonder how you sit with it, right? And I talk to a lot of these people too, some of whom think there is real existential risk, and some of these are very prominent Canadians who know a lot more about this technology than I do, and some are world-famous economists who think we're on the bridge of a collapse, and I just, but I'm not the one making these decisions. I wonder how you sit with that, and do you worry that your practicality might not be the right lane here, and that you're making the wrong bet, that we're, as Canada, making the wrong bet on AI? So we're not betting. I mean, I just, like the terminology matters. We're not making a bet. This is not what we're doing. We are legislating pragmatically to make something that's happening useful for Canadians. You know, this is not about me putting my chips on something. You know, we are investing in education. We're investing in training. We're going to invest in literacy. That's a core. We're going to make sure we protect our information and our kids, and we talked a lot about that, and we're going to invest in our small, medium-sized businesses because we believe that's where innovation happens. I was an entrepreneur for 10 years. I know that's how it works, and we have made a huge bet on making sure that Canadian companies foundationally have access, and the way you asked how it sits with me, and I'm going to answer that because, you know, I talked to Geoffrey Hinton, and he was the godfather of the Nobel Prize winner, and he's worried about AI. When you ask him, he's specifically worried about autonomous weapons that are controlled by AI, so that's a worry. If you talk to Rich Sutton, who's the Turing Prize winner, the other godfather of AI, and at the, at AAMI, the Alberta Machine Learning Centre, one of our national institutes, he's excited. He is a total optimist about the future, so the way I sit with it, and this is my frame. Either we invest in building the infrastructure that is needed to this, or we're going to have to rent it from someone else who's built it. Either we are going to invest in Canadians to provide this innovation, or we're going to just have to buy it from someone else, and either we set the rules here about our privacy, and our So when I talk about pragmatism, all these worries that, you know, you reposit, we just have a really simple choice here, and my choice, and the Prime Minister's choice, our government's choice, is pretty clear. We're going to build here, so that's what this strategy does. We're going to invest to make sure we build, because we could build sustainably, and with our values, and transparently, so we don't have to buy it from someone else. This is not, we are not going a digital colony to someone else. We're going to invest in our innovators, and our IPs, which we're doing. I mean, our strategy has $2.3 billion essentially invested directly into small, medium-sized innovators in this country, through direct investments, and access to compute, to make sure that that innovation is built here, so we don't have to buy it from someone else. And finally, we're going to make the rules here to protect our kids, and our privacy, which is what the legislation is all about, so we don't have to follow someone else's. That's sovereign control. Now, of course, the last thing I'll say is, we have an ambitious AI for all strategy, but I would say I approach this with a pragmatic humility that things are going to change every day, and that this is a framework that there are going to be new opportunities we're going to have to go zig for, and we might have to zag for new challenges, new threats. Cybersecurity is an ongoing issue that is key. People think about, hey, what about Mythos, and we have access to Glasswing, the project that gives governments access to these powerful models, but what happens if there's new powerful models? And I can assure you that in our strategy, sovereign control is very much about that, and cybersecurity, and security, and giving us a sense of control is at the core of this, and again, the strategy is laying that out, and just as we've done with our legislation, I think the public will see that we are taking steps to enforce our own sovereign control and protection against the power of those models in the near term. I think we have a good sense, the public has a good sense, of what the dangers around our privacy are with social media, that we landed on this business model, because we didn't govern effectively, maybe, where our data was used for commercial gain, and that caused a range of harms in the original internet, in the social media internet. What's the concern around AI and data, and how is that different than what we thought of a social media, and why might it need a slightly different approach? I just want to frame this for people, because even as you're listening to this new thing, new chapters are unfolding. You got to have a little humility here, because these are new technologies, and this is not an on-off switch on technology, this is about creating safe environments for people to live, so that was the first thing. We want to give some power to individuals back, so the power to fight back against deepfakes, so if you have a deepfake, you should have the right to delete that, and to say to the provider, please delete that deepfake, so that's called the right to deletion, that's in the bill. We want to make sure that personal information is not used to price gouge on what's called surveillance pricing. Look, Canadians listening are going to say, wait, my personal information is owned by a store, but I like it because I'm getting discounts through my loyalty program, great, but we don't want price gouging, so there's against surveillance pricing. Kids' information should be treated sensitively, like if you're collecting data on my kids, it should be treated like healthcare data or biometric data, and four, we believe that algorithms that are making decisions about folks that, you know, if you're applying for a mortgage or applying for a job and a computer, an algorithm is saying, you get it, you don't, we want to make sure that there's no bias in there, like it doesn't say no women or no people of color, so you should have the right to have transparency on automated decision systems. Those are some of the things, but the key element is, if you're going to do those things, you got to have an enforcement mechanism, so both bills propose a new regulator that, for the first time, will have powers to levy fines and what's called order-making power, so essentially, finally, we are getting a regulator with teeth. So I want to talk about the regulator because I think there has been a lot of debate about whether the Privacy Commissioner should get more power or whether the Office of the Privacy Commissioner, or we need something new, and I thought we could maybe ground this in a practical example here. One of the most striking things that's happened to the country this year, I think, is the capacity embedded in the XFeed by Grok to undress people. Grok AI has been used to perform what some are calling digital undressing, where a user asks the AI to remove a photo subject's clothing, including photos of children. I know there's been lots of events that have kind of shaken your engagement with AI, whether it's Tumblr Ridge, but I think this one is really emblematic because it affected so many people, and it was done with impunity, and we really couldn't do anything about it, right? Like, we didn't have a capacity to push back, we didn't have either of these two bills that gave us the mechanisms to do it, and so the Privacy Commissioner investigated and issued a report that essentially, I mean, my read of it is threw up his hands, and he was like, look, this is still happening, it is still illegal, and there's very little we can do about it, other than to ask them to stop, and they've said no. One, I don't have the ability to find them, and two, I don't have the ability to issue an order. So one of the recommendations that we made was that they suspend the tool until they can put in place all of the appropriate safeguards. So press pause, fix the issue given the harms, and that gives an incentive to fix the issues quickly so that you can reopen the tool. And they've not accepted to do this, and I can't force them to do it. So if you agree that that's a harm that shouldn't exist, and I think it is captured under a number of the provisions inside either of these two bills, how does this new regulator work? Like, how would this solve this problem that we can all agree, I think, is a problem? So let's just first outline for folks the kind of the landscape of how this works, because I think people generally want some kind of enforcement. I mean, what are rules without enforcement? So you want some kind of enforcement mechanism. So the principle of having an enforcement mechanism is something that opposition parties have called for, stakeholder groups have called for. And it's a necessary condition to do anything, right? Like, that's the first principle. And once my other countries can act, we haven't been able to. So the current privacy rules, so this has to do with how commercial entities handle privacy. And I just want to remind your listeners that there are two kinds of acts. What's called PIPEDA, which is the Privacy Act that I'm trying to renew after 25 years with the Protecting Privacy and Consumers Act, that specifically has to do with how commercial entities treat your data. So when you go to a store and you use your credit card, or you sign up for a membership, or any time your data is collected in a commercial, how it's stored, how they can share it, their responsibility and their duties around that. We do have something called the Privacy Act. Now, this is really important. The Privacy Act dictates how governments can also treat your data. Government is a big collector of data. We collect data for, for example, taxes through CRA. So the Office of the Privacy Commissioner is an independent officer of parliament for good reason, because that office's job, first and foremost, and was when it started, to enforce the Privacy Act. Because the government shouldn't be the one enforcing and looking at and regulating how it treats data. We thought the best idea is to have an independent officer of parliament to look at how government handles citizens' data. Which means parliament chooses the person and oversees the body, not cabinet or government, correct? That's right. That's right. And so that's called an independent officer of parliament. And so over the years, we have a number of independent officers of parliament, but the Office of the Privacy Commissioner is that. And that rule is very important, and the Office of the Privacy Commissioner still will regulate how governments treat data. We're not changing that. That remains. That should be an independent officer of parliament. Over the years, as data has become significantly more important, we've asked this office to do more stuff. It just was the only place available to do it. So it started to look at how businesses treat your information. Like Grok and X. Like Grok. In this case. But it didn't really have, as an office, the authorities for enforcement, and it didn't have really the budget to do it. And that said, the Office of the Privacy Commissioner has done extraordinary work investigating. And by the way, even with no enforcement powers, has worked with many companies to come to terms because there's a kind of name and shame element, which matters. And they've really had played an important role. So I'm not diminishing it. But they've been very frustrated, to be clear, with their lack of power. I mean, the Privacy Commissioner has said this monthly going back years now, right? Why don't you just give him more power? Give the office more power. Well, I'm glad to say that the current Privacy Commissioner, when we tabled our Protecting Privacy and Consumer Data Act, was very, very positive. Said, this act builds on my work and my recommendations to do a number of things. First, name privacy as a fundamental right, to treat children's information with extra sensitivity and responsibility, which we've done, to give an enforcement mechanism to enforce, which we've done, to make sure that businesses have privacy assessment frameworks. And so we have built on the Office of the Privacy Commissioner's work. However, the world has changed dramatically. We feel that there should be an independent commission, same way that there's a competition commission, right? Businesses understand that. And in this way, what you'll have when we set this up is a new regulator that over time will gain not just the experience to regulate social media, because there'll be multiple commissioners within this regulator, one specifically dealing with protecting privacy and consumer data. So one of the commissioners will be a Privacy Commissioner, essentially. One of them will be the Privacy Commissioner. So again, the Privacy Commissioner, the Office of the Privacy Commissioner will still regulate the government, which matters a great deal. We may have to come up with different names here. That's right. There's a lot of commissioners swirling around here. There's a lot of commissioners in Ottawa. Yeah. So I just call it the regulator. This regulatory body will have a specific commissioner on privacy and consumer data. But we hope the goal here is to set this commission up so it has the budget and the finances to have a robust regulatory function over how businesses treat data, but also in different areas, social media, AI chatbots, as this space is evolving so quickly and data slash people's personal information as being treated is so important. We think that we need to set up a very robust independent commission structure to regulate that. I guess just to put one more question on this, because I think it's been the subject of a lot of discussion. What if a sort of Trump-esque leader comes into power and decides that the commission should exclude or include certain platforms over others? Are there protections? Are you worried about something like that, like that kind of political interference on what I agree really needs to be an independent body? And are there protections against that sort of level of political engagement as opposed to just oversight? Yeah, I think you're asking exactly the right questions. I mean, there's lots of questions to be asked. There's about the independence. There's about what their mandate will be. There's about how they will rule. There's an appeals process. There's a timing, how fast will this get set up and when will their enforceable powers be? There are questions about, you know, the fines are not insignificant. They're $10 million to 3% of global revenue. So enforcement and how you could collect and how big and therefore, and then what would the appeals be? So, you know, you're asking a lot of the questions. The first question is, as we are in the process of setting up the commission, part of the critical role is making sure that the mandate preserves the independence of the commission structure. And that will be a function of setting this up. I will just, just to remind people, we haven't set this up yet. We literally tabled the bill, but we are now in a really critical process. And this is why I think talking to you about this is so important to make sure that as we set up a regulator, as we set up a regulator to do things that we haven't done before, which is have this kind of enforcement, we do it preserving independence, making sure that things happen in a speedy manner, as speedy as we can. And at the same time as transparent as we can, two things we don't want, we don't want a regulator that's not independent, but we also don't want a regulator that has no mandate. So we don't know what the regulator is going to do because everyone wants us some criteria of how rulings are going to be made, right? So we want a clear mandate on all sides of this. Well, look, I really appreciate you talking to me about it. And I, any sense of righteous certainty on what this is going to be ends up being foolhardy, I think. And I'm reassured to hear you're approaching this with humility. And I hope the government continues to. Yeah, let me thank you, first of all. Your work has been, truly Taylor, has been great. And, and by the way, not just so folks know, the conversations I have with Taylor are usually Taylor criticizing something and then me resisting and then realizing maybe he's got something there and we have to go back, but these are constructive dialogues. We got to, we got to urgently keep up. We got to invest, we got to build, but we got to do it at the speed of trust and citizens and with that humility. So we're trying to make hay slowly here. Well, on that, thanks so much for talking to us. Thanks Taylor. Machines Like Us is supported by the Max Bell School of Public Policy at McGill University. Learn more at mcgill.ca slash maxbellschool. Machines Like Us is also supported by CIFAR, a global research organization based in Canada. Explore their work at cifar.ca.