GC Data Conference 2026 Fireside Chat: Navigating AI and Data Sovereignty, Stewardship and Trust (DDN3-V16)

Transcript: GC Data Conference 2026 Fireside Chat: Navigating AI and Data Sovereignty, Stewardship and Trust

[00:00:00 Animated CSPS logo appears on screen. Text on screen: Welcome.]

[00:00:07 Animated title page, text on screen: 10 Years, GC Data Conference 2026, Navigating Canada's Data Landscape.]

[00:00:19 Vanessa Vermette appears, speaking from a lecturn on stage. Text on screen: Vanessa Vermette, Vice President, Canada School of Public Service.]

Vanessa Vermette: I would now like to welcome our speakers to the stage for our opening session. A very warm welcome to Taki Sarantakis, President of the Canada School of Public Service, Nick Frosst, the co-founder of Cohere,

[00:00:35 The panelists walk on stage as they are announced and take their places in chairs on stage. A large screen is seen above them, showing their names and photos.]

Vanessa Vermette: Bailey Kacsmar, Assistant Professor in Computing Science at the University of Alberta, and Alberta Machine Intelligence Institute Fellow, and Abdi Aidid, Assistant Professor at the University of Toronto Faculty of Law. While they come up on stage and get settled, I'll share a few highlights of their bios.

[00:00:50 Vanessa Vermette speaks from the lecturn on stage. The camera shows brief views of the panelists on stage as each presenter is introduced.]

Vanessa Vermette: First, we have Nick Frosst. Prior to founding Cohere alongside Aidan Gomez and Ivan Zhang, Nick collaborated with Geoffrey Hinton, becoming Professor Hinton's first employee at Google Brain's Toronto lab. And yes, Nick is also the lead vocalist of the indie pop rock band, Good Kid. I hope Taki asks him about that.

Next, joining us all the way from Edmonton, Bailey is a fellow at the Alberta Machine Intelligence Institute, and her research explores the intersection of privacy, human trust, and public perception. She aims to develop human-centred privacy technology through the parallel study of technical solutions for privacy and machine learning, alongside the corresponding user perceptions, concerns, and comprehension of these developments.

And last, but not least, Abdi Aidid, a leading voice in the evolving relationship between law and emerging technologies. Among his many accomplishments, he holds a Canada Research Chair in Artificial Intelligence and Access to Justice and is co-author of Legal Singularity, a book about the transformative potential of AI in the legal field. Abdi is also the first to serve here at the School as the Ian D. Shugart Visiting Scholar.

Please join me in welcoming Taki, Nick, Bailey, and Abdi.

[00:01:57 The audience applauds. Vanessa Vermette exits the stage. The camera switches back and forth from various camera angles throughout the discussion, including close-ups as a presenter is speaking, and wide views of the panelists on stage as the discussion proceeds.]

[00:02:11 Overlaid text on screen: Taki Sarantakis, President, Canada School of Public Service.]

Taki Sarantakis: Thank you so much, Vanessa. Nick, I'm not going to ask you about whatever it is Vanessa said, but I'm sure you're really, really good at it. And you're really, really good at a bunch of other stuff that we'll spend our time talking about here over the next little while.

So, as Vanessa said, this is the 10th data conference in the Government of Canada. The first one, I think, was attended by a couple of hundred people. I think we've been peaking the last few years at between 10,000 and 11,000 people, just to show you the degree to which this issue has had increased salience over the last decade. And also, today we have between 6,000 and 7,000 people online, in addition to the people that we have in the room here today.

So, Nick, start us off a little bit. Why do we care about this stuff? Why is data important? Why is AI important and why is the world kind of shifting under our feet so quickly?

[00:03:23 Overlaid text on screen: Nick Frosst, Co-founder, Cohere.]

Nick Frosst: Why is data important? It seems self-evident. I remember when I first found out that evidence-based policy was like a buzzword. My mother worked in the government for a little bit and she, at some point said, Oh, everybody's excited about evidence-based decision-making.

Taki Sarantakis: As opposed to bias-based decision-making.

Nick Frosst: What the hell have you guys been doing before?

Taki Sarantakis: Yes, exactly.

Nick Frosst: Yes, so I was pretty surprised about that. I think when you guys started this 10 years ago, 10 years ago, what was going on? AI was developing, we were making neural networks. I was working with neural networks that are still fundamentally the same neural networks that are going on today, but it wasn't as accessible. But there was still a realization that, Hey, we should be making decisions by collecting a bunch of data and then analyzing it and then figuring out what is optimal based on that.

Yes, that seems like a very good idea. And independent of machine learning, independent of transformers, of language models, of like modern AI, there's still a huge amount to get done there. It seems like – and I'm not in government, so I don't know this – but it seems like evidence-based or data-based policymaking is still a bit of a buzzword. That's still not the default.

Taki Sarantakis: It doesn't just seem that way, it kind of is that way.

Nick Frosst: Well, I'm trying to be nice.

Taki Sarantakis: So, Bailey, if this is kind of self-evident and something that it sounds like, from what Nick said, it's an intrinsic good, why are there some of us that are a little uneasy? Why are there some of us that are worried about a thing or two about this kind of rapidly emerging new world?

[00:05:14 Overlaid text on screen: Bailey Kacsmar, Assistant Professor, University of Alberta and AMMI.]

Bailey Kacsmar: So, one of the first things that comes to mind is always when we say data, especially those of us who are technologists or writing legal documents, we treat it as this abstract concept. But in reality, most of the data that's going into these systems are being used for decisions is information about people. And that means that it has this connection, but also there's consequences to how we use, or potentially misuse this information, for anything that we make based on it.

And so, one of the things that most contributes to that is people don't understand how it connects to them, what it impacts in terms of their day-to-day life, and how these pieces of information about them came to be collected. And so, there's all this unwieldiness that tends to be a bit much to understand, or to get under wraps because you shouldn't need to be an expert in this field to understand how these things affect you.

But the reality is all the communications that come out about these things do assume a certain level of expertise, or that you should just trust in whatever a particular expert says. But experts don't agree with each other. And so, then people are back in their cycle of, I don't know what to do, and why are you collecting this information about me, despite the fact that data is an intrinsic part of our society and it's not even a new one.

I don't know how long we've had a census, but I'm thinking it's pretty long. It's just now our data is largely digital, which also means how it flows has completely changed. You can't just take a physical document and lock it up. It can travel across the internet throughout the entire world. And so, you can no longer visualize the constraints on that information, even if it's been collected about you by someone you trust.

Taki Sarantakis: It's really interesting that there is queasiness vis-à-vis the state, because from the period even before the state, like villages, towns, royal courts, a big feature of that was data, like registries. Who lives here? Who doesn't live here? Who owns property? Who doesn't own property?

The state historically has been a data machine, and the state historically has been the repository of a type of data that we call the truth. It's like I own this little property in the city of Ottawa with this coordinate and that coordinate, and if somebody in the City of Ottawa didn't have that data, and assert my ownership right to that data, I couldn't have that property.

Now, Abdi, talk to us a little bit about— I want to stay on the fear a little bit, because I think we're going to be talking about both, but I want you to talk to us a little bit about the fear of data, and I'll give you a kickoff. I think as Bailey kind of said, we're not just talking about data because if you're talking about where I live, where I work, what my heartbeat is, what my cholesterol level is, who I'm married to, who my children are, it seems to me you're not talking about data, you're talking about me.

[00:08:42 Overlaid text on screen: Abdi Aidid, Assistant professor of Law and Visiting Scholar, U of T, Canada School of Public Service.]

Abdi Aidid: I think Bailey's right that we talk about data as this disembodied thing, but really it's about who we are. And actually, data becomes more problematic the more it tells about who we are. And I think that kind of goes to the central challenge of AI too, because once upon a time you had to follow me around to know what I was up to in the course of a day. Now you could synthesize a bunch of things.

Taki Sarantakis: And we did, just so you know, the government did follow you around constantly.

Abdi Aidid: I felt it today walking in the building. But I think that's a key thing is that we have this, in part because of our remarkable data science capacity and computing power, we're able to not only ingest more information about people, we're able to make predictions on the basis of innocuous information. So, think about it. You're going to hear me beat the drum of like the law is not enough a bunch of times, but here's how it's not enough.

The law contemplates private information, personal information, a piece of data or datum, I don't know what the singular for data is, that reveals something about you. Well, what about this? Now, because of the advent of big data, we're able to take a bunch of different innocuous pieces of information about you, synthesize them, and then make an intrusive insight about you.

Taki Sarantakis: i.e. you're pregnant.

Abdi Aidid: Yes, maybe, yes.

Taki Sarantakis: i.e., you have diabetes.

Abdi Aidid: Exactly.

Taki Sarantakis: i.e., you're having heart problems.

Abdi Aidid: And so, your legal interest in the privacy of that information doesn't attach to any constituent part of that computation, but the end result is something that's maybe more revelatory about you than any personal information that you were already guarding. And so, that's kind of what technology enables now.

You've heard me say we have insufficient legal and regulatory responses to that problem, but the other one is that people can't really imagine that happening all the time. They can't quite figure out what to steel themselves against. I know that if I'm trying to hide information about my family, that I'm going to draw my curtains and I don't want you to see what's happening inside my house.

But how do I steel myself against an advanced computation that's making a prediction about me? Am I not just going to engage in a bunch of different innocuous behaviours? That's part of the challenge. It's not just the fact of data being used to make insights about you, but it's also the fact that the traditional ways we protect ourselves against it have been abstracted away from.

Taki Sarantakis: Yes. Now, I'm sitting here with 3 super accomplished people, 3 geniuses, and I'm like the token dummy on this panel. But here is my real big contribution to this panel. And it's as follows:

An American dies and goes to the place before you get to go up or down. And the person looks over their life and says, "Congratulations, Mr. Smith, you get to go to the good place. You get to go up". Mr. Smith is thrilled. And then a European dies and goes through the same process. Somebody looks and says, "Congratulations, Madame Labelle, you get to go to the good place." And she's relieved. A Canadian dies and goes to the same place. They look over their life and, thankfully, they get to go to the good place too, but the Canadian is not happy. The Canadian is outraged and the Canadian says, "How did you get my information? Who told you about me?" We have some of the highest rates of, let's call it distrust, of technology, of AI systems, of registries in the OECD.

Who wants to start us off on that? Because that's empirically true. Survey after survey after survey says Canadians trust this stuff much, much less than our counterparts around the world do.

Nick Frosst: Mistrust what? Mistrust technology, as a concept?

Taki Sarantakis: Yes.

Nick Frosst: Mistrust the organizations that use that technology, just all of it?

Taki Sarantakis: Check, check.

Nick Frosst: That's interesting. Canada's got an interesting history. There's a lot of people here who have pretty good reason to mistrust the government, I would say. So, it's not surprising that that has ended up that there's a cultural memory of that.

I hope that Canadians can sort out the difference between a mistrust in a technology, and a mistrust in an organization using that technology. I think these days, especially as technology is rapidly advancing, it's very difficult to tease those two things apart. It's very difficult to say, Okay, this is a technology that I have sovereignty over, that I have control over, or this is a technology that is being given to me, and I don't have an option in using it or something. I don't know what would need to change in order to get that delineation.

Taki Sarantakis: He knows. Abdi knows.

Nick Frosst: I do see here a lot of technologists. There's a lot of people who are interested in pushing science forward, who are interested in pushing technology forward, so that mistrust, at least, is not resulting in a pure rejection of it.

Taki Sarantakis: Right. So, you've nicely caught the distinction between the people working on the thing who are like, "let's keep going, let's keep moving forward, let's make it better, let's optimize." And the people on the other side of the thing going, "I don't know about the thing."

Abdi Aidid: I think the high level of distrust is maybe not even necessarily because there's a distrust of what the organizations developing AI are up to, or a distrust of government. I think it has to do with the fact that people don't know what they would do if something went wrong. So, I think about how we're fundamentally incurious about technology that we trust.

Most people probably don't know how their car works; except they know that they can call a mechanic to fix it. Most people would be surprised to know that most of what operates in your car are actually lines of code, and when you depress the brake pedal, it doesn't clamp the wheel anymore necessarily. That's not what's slowing down your car. Surprise.

But you don't care, because you know that you have someone to call if something goes wrong. We get on airplanes, and we know there's a pilot in the cockpit, and we know there's a hotline you can call if something goes awry with your ticket, and you know that there's Transport Canada that's checking the nuts and bolts. And so, you understand that you operate in an ecosystem of constraint, so if something was to go awry, that you'd have some recourse.

I think part of the challenge right now is we're so curious about the technology that underlies AI. Everyone wants to look under the hood, whether they're capable technically or not. And part of that, I think, has to do with the fact that that's kind of all that we have in the world of AI. I think if there was some enabling rules, some constraining rules, something akin to recourse if people had issues.

And that recourse need not be legal recourse. I'm not talking about a cause of action in case something goes wrong at a new AI startup. I'm talking about actually an institution that can help mediate your experience with the technology. I think you'd probably see higher levels of trust. I will say also the low level of adoption in Canada – stats lie a little bit sometimes, so Canada is dead last in adoption.

Taki Sarantakis:You know, 28.794% of all statistics are made up on the spot.

Abdi Aidid: Made up, yes. Well, here's one that's not made up. Canada is last in the OECD and G7 in AI adoption at the firm level, and that is probably more—

Taki Sarantakis: We're around 12%.

Abdi Aidid: 12%. That's probably more due to the fact that we have a lot of legacy industries, let's say old industries, extractive resources, telecom, AKA money printers, where there's not a significant incentive to really change. They might be two technological revolutions behind on some systems because all of the incremental R&D money is going to better extractive technology.

And so, I think part of that has to do with the fact that AI adoption is not bespoke. The adoption strategies are not bespoke for the Canadian marketplace.

Taki Sarantakis: Absolutely. Now, Bailey, one of the things you study on this theme is technology, AI, and privacy, which is really salient to this conversation because I think what I heard Abdi talk about in part was privacy, in part was rules.

Zuckerberg, Mr. Zuckerberg, a long time ago said something famous or infamous. He said, "Privacy? Who cares about privacy? It's like, get over it. Nobody has any privacy. And if you're asserting privacy, what do you have to hide?" I'm paraphrasing but talk to us a little bit about that big disconnect between maybe the providers and maybe the users.

Bailey Kacsmar: Well, since you started off with Facebook/Meta, I point out their whole thing needs people to share information. So, of course you're going to want to say the nothing to hide argument, but it's an absolute fallacy to start from there. It's an oversimplification of privacy to one concept that it's somehow just about hiding things.

When, if you start thinking about privacy as an actual concept and in your reality, everyone in this room has an intuitive definition of privacy. They're not all the same because privacy is also incredibly complex. It depends on your life, so it has a lot of individualistic and cultural components.

But the reality is privacy is not about hiding; it's about existing within a public sphere but still retaining being yourself. If we don't have a society, if you don't have interactions, we don't need this notion of privacy. It's very, very human and very, very social. And it's about what does and does not flow to different people.

Like, you tell things to certain things to family members that you maybe don't tell to your boss. Or you say things to your friends, and you don't say them to your niece or nephew. All of these things are you enacting a form of privacy at a very simple level. So, when people are seeing these large tech companies taking all their information or kind of hiding all the ways they're collecting their information, it can become very, very hard to control it.

And so, if you think about just your day-to-day life: you opened up your banking website because you wanted to pay a bill and then you had to get down here on time before everything started. You had to click some buttons. Do I, do I not agree with these cookies? You're like, I don't know, do these matter? And you're in a hurry, so you just click the fastest one to get that annoying window out of the way. But as soon as you did that, you technically consented – not really, but that's what we model it as – to that information being collected by that company and then being used in different ways that are in their giant, legalese privacy document.

So, when I hear people say things like the "nothing to hide", or "people don't care about privacy", I always come back to, have you spent even just a few hours seeing how much time and effort it takes within the way our technical systems work to have any actual active control over when you do or do not send information to someone?

And so, that issue of trust comes really into play, as was being emphasized. And some of it is things like you were comparing to Europe and to the States. Europe has way tighter data protection laws. The US is kind of like, nah, man, we'll just have health data and children data, and the rest is probably fine. And so, we have these complete extremes and then there's Canada's version where we have these privacy laws, but we have these emerging technologies, like our existing privacy law hasn't been updated in a while, and people don't know what that means in the digital world. It's not like a car where we know not all cars are built the same, but they all have to make a certain safety standard for when they crash.

What happens when the technology systems crash? In a certain setting, there should be standards of what is an okay fail state. What [are] the requirements of these systems? If we're trying to build in trust, we need to be cognizant of the domains we're putting all these systems in. The data being collected from them, being sent back, and all of it, the same way we've been doing for so many things.

Taki Sarantakis: Now, Nick, it seems to me that if I listen to your two colleagues on your left, it seems to me you can make a strong argument that your data should be your data. And, Bailey, you should own your data. And, Abdi, your data should accrue to you as ownership. Is that the right way to think about this, that my data is my data, and you shouldn't be able to use my data without my consent, or no?

Nick Frosst: That's obviously the right way to think about it.

Taki Sarantakis: Really?

Nick Frosst: Yes.

Taki Sarantakis: Absolutely the right way to think about it?

Nick Frosst: Yes. I mean, of course, there are moments in which it is better to pool data. There are moments in which we, as people, decide that it's best to take data, aggregate it, put it available in some capacity. Like people who are writing things on the internet, hosting websites that anybody can visit. There are decisions that people make when they say this is the data, this is the stuff I wish to put out there.

Taki Sarantakis: Okay.

Nick Frosst: But you should be making that decision. Or like at a hospital, Health data is of the most important, some of the highest, most important, most personal data that we have. And yet, many of the scientific breakthroughs that have happened, and that have yet to happen, happen as a result of the aggregation of data.

So, there are times when we definitely want to do that. But I think it's pretty clear that it's a good idea for people to be aware of that and to have more agency in the process.

Taki Sarantakis: So how come it isn't thus?

Nick Frosst: How come it isn't? Because there are like large for-profit entities that exist to try to...

Taki Sarantakis: Oh, interesting.

Abdi Aidid: There's another challenge, which is ownership being the wrong analogy.

Nick Frosst: Okay.

Abdi Aidid: So, my iPhone's in my pocket. For the AV people, it's in the pocket where the mic pack isn't. That's why you're not hearing feedback right now. So, my iPhone's in my pocket. How do I know that I own my iPhone? How do you know that I own my iPhone? You got it? How do you know that I own my iPhone? Well, maybe my name is on a receipt somewhere when I bought it at the Apple Store, but the principal way that you know that it's mine is that I have it, and you don't.

Taki Sarantakis: So, you physically have something.

Abdi Aidid: There's excludability. It's like a non-fungible good in that same way. There's excludability.

Taki Sarantakis: If you have it, I cannot, and vice versa.

Abdi Aidid: Which is not true of data, because if I send you an email with my personal information, now you have it and I still have it, and whoever you forward it to has it. And so, ownership's the wrong framework. I know, as public servants, some of you will send the recall email and think you got the stuff back. Like, it's out there. It's out there.

Taki Sarantakis: So, that makes people even more curious.

Abdi Aidid: Yes, yes.

Taki Sarantakis: Like when you get a recall message...

Abdi Aidid: What was the first message?

Taki Sarantakis:…run to that message.

Abdi Aidid: Yes, lawyers, we do a thing at the bottom of our email signature.

Taki Sarantakis: It's 8 pages. Yes. If you get this by accident, please send it back to us.

Abdi Aidid: Exactly. It's super long. And again, that's a lie. It's all a lie.

And so, I think part of why we're struggling right now and sort of blustering through this is the concept of ownership is a bit of a poor fit for how we might think about data.

Taki Sarantakis: So, what's the right word, if it's not ownership?

Abdi Aidid: It's something more akin to what Bailey was saying, which is a more dynamic notion of control. Agency, Yes. And the other challenge with ownership is that it's kind of a, I don't know how to say it, like it's an agnostic idea.

So, for example, we say things like "own your data". Well, if privacy is a virtue, if it's a value, if it's a thing that we want because we want people to participate socially as their whole selves, then if they own it, then you could just say they can sell it to whoever they want to, including the highest bidder. Then you create a secondary market where, if I need some money, why don't I sell my data?

Taki Sarantakis: Anybody want to say a last word on privacy before we move on?

Nick Frosst: Privacy, I think we're talking about this, and the kind of framing of this conversation is that it's a thing that people want and not a thing that organizations want, and that's not the reality at all.

Taki Sarantakis: That's because I'm the token dummy on the panel.

Nick Frosst: Yes, at Cohere we make models, we give them to our customers and then our customers, those businesses, they see what goes in and out of that model, in and out of that AI, and we do not. And that is a huge value add for us. We work with companies who have the highest level of privacy and security requirements because they're working in regulated industries, because they're working on healthcare or something, or because they just think it's a strategic advantage. And we take the AI that we build, give it to them, and then we do not see what goes in and out, and they can keep it completely secure and private. They can run it disconnected from the internet. They can run it up in the Arctic if they want.

Taki Sarantakis: Excellent.

Nick Frosst: And that's a big value. There's a lot of companies that are interested in doing that for themselves and for their customers. So, I don't think we need to frame it.

Taki Sarantakis: What I'm hearing you say, I think, is there are other models than the kind of ad-supported, everything's open, everybody come. There's presumably a type of a subscription model or a payment model where you're not doing a trade-off between privacy, data, security. Things are controlled to the way you want to control them, as the holder of the data. Is that correct?

Nick Frosst: Absolutely.

Taki Sarantakis: All right. Any last words on privacy?

Bailey Kacsmar: Just as a kind of conclusion, while it's great when we do have organizations providing these services that are cognizant of privacy and the impact of the data they collect or send or whether or not they see it, one, we can't rely on all companies to be benevolent, otherwise we wouldn't really need any of our regulations. Two, you do run into a risk where it starts to be the companies who are for-profit organizations making the rules about what is and isn't available in terms of protections or visibility into how data flows.

So, you start to get into that like pay for privacy model, which isn't necessarily a world you want to shift towards, even though, yes, we could argue it's great that this is available, but it means that now you've divided it into classes. Who is able to pay to participate in these more protected or privacy-respecting systems?

Abdi Aidid: <inaudible> will do the work of developing internal compliance processes and procedures because data loss is high stakes for Cohere. I'm also worried about the rinky-dink operations that, nevertheless, are able to get significant amounts of your data. And so, if you leave it entirely to them, then you create that additional risk. I'm not worried about Nick, I'm worried about the copycat.

Taki Sarantakis: Yes, Bob's Corner AI. So, let's shift now to a word that very few people used to talk about in Canada. Now you can't not talk about it. Sovereignty.

So, let's talk a little bit about, first of all, we'll talk about data sovereignty. We'll talk about compute sovereignty. We'll talk about a few other types of sovereignty. But do any of you have a definition of sovereignty that you'd like to share with those of us watching? You have a kind of an evil smile, Bailey, so let's start with you.

Bailey Kacsmar: No, I feel like I should throw it to the expert on terminology. Sorry, it's just you have to write formal text all the time, given your expertise.

Abdi Aidid: Yes, I'm at a loss for this one. I could tell you; it's being used in so many different ways.

Taki Sarantakis: Give us one way you think people think about sovereignty or you think about sovereignty.

Abdi Aidid: Yes, when we're talking about sovereignty in the data context, we're often talking about having exclusive jurisdictional control over the data, which means—

Taki Sarantakis: So, jurisdiction is one thing, I think that's a fair—

Abdi Aidid: So, one of the things, and this goes to the point I was making earlier, about data being the fungible kind of thing that can bounce around the entire world without it ever reducing, is that it's very hard for any domestic legal apparatus to constrain data and impose rules on it. In part because if you're a Canadian company, let's say, you can route your data in a bunch of different places, and it would limit Canada's territorial reach, but it would also create jurisdictional anchors from all the other places, all the other countries in which you are.

And so, if you're a nefarious actor, you can start to play the regulatory arbitrage and play games and look for the place that has, say, the lowest privacy protections. So, one reason that you might want data sovereignty—

Taki Sarantakis: Companies never do that. Never. They never incorporate in Delaware or Ireland.

Abdi Aidid: Yes, Yes, Yes. I'll let that pitch go by. I'll let that one go by. But the point here is it's the idea about having more domestic control and relatedly also being able to capture the economic –

Taki Sarantakis: Control, another word.

Abdi Aidid: – and to capture the economic value of more of that. If you hear the Prime Minister talk about data sovereignty, it's also an economic growth strategy. It's the idea of repatriating the information such that it could support a domestic industry of everything from technologists to data centres, to technicians, to mechanics, to construction workers, to whomever it may be.

Taki Sarantakis: So, Nick, I want to come to you on this one because you're with Cohere, you are, I think, a co-founder of Cohere. And you hear a lot of people, including in the very highest echelons of our world, talk about Cohere as our sovereign hope, Cohere as our possibility of playing on this stage.

And I've read a couple of things, interviews you've done over the years in different fora, and you and other people at Cohere have been very clear – we're Canadian. We can go somewhere else; we have chosen to stay here. We have chosen to be Canadian. And part of that has now become relevant to governmental players, which is like, Hey, we have one of these that's based here and can help us with our sovereignty. Talk to us a little bit about that.

Nick Frosst: Yes. Yes, there's a lot there. Let me start by just sketching a little bit of the landscape of AI right now. So, when we talk about AI, mostly we're talking about language models. We're talking about the chatbots that we use, and the things that power any time you're using language on a computer these days. Those models are powered, they are called foundational models. There's 10 companies in the world that can make them. It's 10, maybe 11, maybe 9, it changes every now and again, but it's been about 10 for about 3 years. There are 4 countries in the world that can build that technology. There's China, there's the States, there's us in Canada, there's Cohere here, and there's one in France.

Taki Sarantakis: And there's a big gulf between one and two and three and four.

Nick Frosst: Each of those companies, of those 10, they've all become pretty large. But yes, there's <inaudible> some of the sizes that big. But that's it. There's 10 companies in the world that make this technology.

The reason is it's really difficult to make, it's really capital intensive to make, it requires a bunch of data, a bunch of compute, a bunch of talent. It's just we've ended up in this situation. That's not the situation we thought we were going to end up in, but the way the technology has developed, that's the way it is.

We have stayed in Canada because we are Canadian and because I grew up here in Ottawa. Ivan grew up in Toronto. We have never—

Taki Sarantakis: So did he. Did you guys meet when you were kids?

Nick Frosst: We go way back. Yes. So, thinking about leaving as a strategic decision is a weird thing to think about. We're Canadian. This is where – we're here. We're a global business now. We have offices all over the world. But this is our roots, and this is where we want to build the business.

I think we are a sovereign AI company. And I think the word of sovereignty is very resonant at many different spatial scales right now. We, as a nation, are thinking about sovereignty, and honestly, when people say sovereignty, I just substitute autonomy or agency for that word. It just means, are you in control of what's happening to you. We think of that, as a nation, a lot.

In today's market, companies are thinking about that a lot. Like, is my company in control of what's happening to us? And people, we are thinking about that as individuals a lot. Like, am I in control of what's happening to me? In particular, in our digital lives. Personal sovereignty, personal agency is not experienced that much right now in our digital lives.

So, when we think about making AI, and how we give it to our customers, and how we give it to people, we always think about, Hey, is this technology being put into the hands of somebody who is then in control of it and who can decide how to use it and who can do it privately, securely, safely, do all these things that need to get done in order to make it useful? And of the 10 companies out there, we're really the only one that has that as a focus.

Taki Sarantakis: Yes.

Nick Frosst: So, yes, it's been very promising to see Canada and the world be like, think about it. It's been very motivating to think about, Hey do we have autonomy? Can AI give us that at every spatial scale?

Taki Sarantakis: Yes. Now, Amy, you're at one of the world's great AI institutes.

Bailey Kacsmar: Bailey.

Taki Sarantakis: Oh, sorry, Bailey, you're at one of the world's great AI institutes, called AMMI. Do you guys talk about sovereignty? Do you talk at all about, This is important for Canadians, or this is important for Americans, or Europeans, or on and on and on? Or do you talk about mostly the capabilities, and you're kind of agnostic about the country of usage, or the country of deployment?

Bailey Kacsmar: Well, for one thing AMMI is one of the three pan-Canadian AI institutes, and each one of them, including AMMI, covers tons of things. So, there's people working both on what does it mean to have AI for Canada, how do we take advantage of the expertise and infrastructure that we have here in Canada and use it to the best benefit of Canadians, but then there's also the whole gamut of people working on just advancing AI in general into completely new spheres.

In terms of sovereignty, I won't speak for the entire organization. I'm one of, say, I think we have 50 fellows right now, and that's not including all of our research scientists and the rest of the team. But we're talking about this in large part because there's been this realization via the mass deployment of all these LLMs that we're interacting with these technical systems ran by organizations completely outside of our control in our country and our jurisdiction.

Except that was already the case. Everyone is using all these systems, whether you're using Google, Microsoft, Apple, or otherwise, none of those companies are based in Canada. They have bases in Canada because they have to, but they also have ones in the States and throughout the whole world. This isn't actually a new problem, it's just now it's a different technology that has the focus in terms of whether or not it's being held under Canadian control, and which laws apply and all the complications of cross-jurisdictional boundaries.

And so, in my world of research, one of the things that we focus on is how can you ensure that those unauthorized actors can't access things? Is it possible to still be working within the infrastructure that is our status quo without having data be in some form that someone you don't want authorized or some actor can't force a company to give access to the data?

And so, that's an active and ongoing world of research in terms of, can we have computing over encrypted data for these particular services? Can we have meaningful privacy protection of training data or query data for AI systems? And to what extent is that not possible? And we don't have answers to all of those questions yet. So, if we're using AI or any technology in a domain where we need those answers, sometimes the answer is just, you have to wait.

You have to decide when and where you should apply it to best take advantage of the technologies, but without using them in places they're not ready for it yet.

Taki Sarantakis: Now, Abdi, I want to take something that Nick said, and I want to kind of extrapolate from it, and then ask you a question. So, Nick basically said there are only 4 countries in the world that can do this. There's about 200 countries in the world, depending on the day that you count. So, is that saying that if you fast forward a little bit longer, 196 countries in the globe are not going to be sovereign, or they're going to lose out because they don't have these sovereign capabilities? Talk to us a little bit about what it means to be sovereign in the AI space and not be sovereign in the AI space.

Abdi Aidid: Yes, sitting here today, it means that if there's a domestic technology industry in a country that's not one of those four and it wants to make use of these super high-powered sophisticated tools, then it's going to depend on an off-the-shelf solution in that way. And it's also going to then be making bets that the laws of that host country, that that company is from, are sufficiently consonant with their own values.

And so, part of the play for sovereignty is to say that if Cohere is our Canadian AI solution, then how do we invest and support Cohere, but also make sure it's accountable to the Canadian public, because Cohere will answer to Canadian institutions and not every single newfangled AI regulatory apparatus everywhere in the world.

And so, as a matter of fact, it's a massive advantage for Canada. And if the short form of your question is, is it true that there's 196 countries in the world that are perilously disadvantaged right now, the answer is yes.

Taki Sarantakis: Any last words on sovereignty before we move on to our next topic?

Abdi Aidid: There's the America angle that we haven't talked about.

Taki Sarantakis: Talk to us a little bit about that. Apparently, there's this guy. He's new, and he says stuff about Greenland and other places.

Abdi Aidid: Yes, he's new, but we've known him for our whole lives. I think it's important to recognize Canada's position in the world.

In the European Union, they're, in some ways, modelling data privacy regulation for the world, even though the GDPR doesn't mention privacy. It's a data protection regulation. And they have the AI Act now, which in many ways benefits from their data protection regulatory institutions.

In the US, you have no federal statutory privacy law. You have HIPAA, which is a health privacy law, you have the Fourth Amendment prohibition on unreasonable search and seizure, and you have some common law torts of intrusion upon seclusion, but you don't have a general privacy law in the United States.

Taki Sarantakis: It was actually debated at one point and they said no, no.

Abdi Aidid: There's active measures to stop it. There's a bunch of different bills right now that are percolating through the House and Senate that will never get passed. In fact, the president issued an executive order to effectively restrict AI regulation happening at the state level. So, people ask all the time, when is the US going to have AI regulation? And I say, around the day after never because they don't have privacy regulation, they don't have meaningful privacy law.

Taki Sarantakis: But also, they don't have an interest in it because they want to keep scaling these companies.

Abdi Aidid: Yes, and part of the challenge for Canada is that a lot of Canadian companies – Canada is what, 40 million people, it's roughly the same population as California and one of its small neighbouring states, so if you're a Canadian company, then the smart money is on developing for the US market.

And so, one of the challenges that we have is, do you have the kind of gold standard aspirational set of privacy and data regulations, or do you have ones that still facilitate Canadian business without creating an effective double tax? Because Canadian companies will have to comply locally and internationally where they do business.

And so, there's an awkward geopolitical posture that gets compounded by the fact that the US has something called the CLOUD Act. So, the CLOUD Act in the...

Taki Sarantakis: Talk to us a little bit about the CLOUD Act, because I think that is one of the things that more Canadians should appreciate. Very few Canadians actually understand the provisions of the CLOUD Act to the extent that it affects their life.

Abdi Aidid: Yes, so super simply, without getting into the legal contours of it, data that's hosted in the US or routed through the US can be accessed by US federal law enforcement through a process. And so, there's a special interest if you take privacy to be a value, a national value, a natural justice outgrowth for having data sovereignty so that you can avoid that intrusion, which might feel increasingly arbitrary.

Taki Sarantakis: So, does that mean, just to kind of bring it down to a certain level, if your health data is stored at Google or Microsoft or something, the US government, after a process, has a statutory right to that data?

Abdi Aidid: Bailey's laughing because health data is the exception. That's the one that you can't send, that Canadian health information custodians cannot host Canadian health data outside of the country.

Taki Sarantakis: That's actually good. I didn't know that.

Abdi Aidid: Which tells you how, when we treat data, when we think of what's important...

Taki Sarantakis: When we decide something consciously, we decide something consciously as opposed to making it...

Abdi Aidid: But I don't want to over— this is getting turned into a law lecture, I'm so sorry.

Taki Sarantakis: No, but it's something that people need to understand because we hear a lot of people say, Well I don't care that it's AWS, I don't care that it's Microsoft, I don't care that it's Google. You might not care, but there are consequences to not caring.

Nick Frosst: It might be the case that most citizens are not thinking about this very much, but I can tell you that businesses around the world are increasingly thinking about this.

Abdi Aidid: They don't want to comply with arbitrary subpoenas either. That's a lot of time.

Nick Frosst: So, there's all the largest firms outside of the US are looking for ways to decouple, are looking for ways to get technology providers from people outside of America.

Taki Sarantakis: Exactly. Now, our next conversation. We talked about how Canadians have kind of low trust in some of these things. And we have done a lot. We have disproportionately punched above our weight in the creation of AI. We are punching disproportionately below our weight, not only in capturing the economic benefits of AI and related things, but even something as simple as adopting AI.

We have, and I think we've already mentioned it, but I think we'll talk about it more here, we have, I think it's the lowest rate of AI adoption according to Statistics Canada of any OECD country. And when you think about the fact that a lot of this stuff was invented here. A lot of this stuff was paid for indirectly by Canadian taxpayers. A lot of the advances were made in Toronto, in Alberta, in Montreal through various entities and universities. You're one of the few that have managed to scale in Canada with these technologies.

Talk to us a little bit about the global picture, and then maybe talk to us a little bit about how you managed to be an outlier.

Nick Frosst: Yes, it's true that the Canadian ecosystem, its contribution to research has been – I mean, the technology was invented here. The technology was invented and then through stubbornness and grit was turned into something real. Geoff Hinton was working on this for 20 years with people telling him it was a terrible idea.

Taki Sarantakis: Exactly. Why are you doing neural networks? Give me a break.

Nick Frosst: Yes. And yet, Cohere is, I think, one of the largest startups in Canadian history. I think we're certainly up there. And we're not looking around seeing a bunch of other people doing it. And I hope one day we are. I hope one day there's a whole bunch of Canadian homegrown startups building cool and useful technology.

I think the reason for that is fairly simple. I think it's just flywheels of funding. People got it in their heads that if you wanted to make a startup or you wanted to fund a startup, you had to go to Silicon Valley. And that made both of those things explode, and they feed off of each other. And a willingness from the public and the private sector in America to adopt American tech as soon as possible.

Those two things together, you have the American government going in and investing and purchasing American technology right from the beginning. And you have huge numbers of investors and huge numbers of people who want to found companies, and programmers who want to work at them. That just started a cycle that created what Silicon Valley is today.

Now, I think there's an opportunity in Canada to do something similar, but I also don't look at Silicon Valley and say, Hey, the version of life, the vision of life they're exporting is the vision of life I want. I think most people – when I was an undergrad, I was a techno optimist. I was really excited about the future of the technology. I'd say I've been pretty let down for the past 15 years. I think most people hate their phones. Lots of technology that was exported from that flywheel is not something people are thrilled about anymore.

Taki Sarantakis: It's addictive. It's causing our kids to have anxiety. So, on and on.

Nick Frosst: So, I still love science. I still love technology. I love language models. I think they're a beautiful technology and can be so useful at doing so many things that I don't want to do. And I think it's a good thing to have those flywheels for economic prosperity, for sovereignty of the nation, for sovereignty of the individual.

So, we need those flywheels. We need to get more investment. We need to get more builders. But I don't think we should be looking at Silicon Valley and saying, Yes, exactly that. Let's do that here. We have to do our own version.

Taki Sarantakis: Is it fair to say, and I'm going to go to the others in a moment, but is it fair to say you have to have a certain stubbornness in your industry not to go to the United States? That it's just so much, I'll say, easier to get capital, easier to get customers, easier to get what you think is a more reasonable or accurate valuation for your company or your assets. Do you have to consciously say, No, I'm going to stay here despite this, despite that, despite this third thing?

Nick Frosst: Yes, I would say our lives would have been easier.

Taki Sarantakis: Your professional lives.

Nick Frosst: Professional lives would have been easier had we gone to the States years ago. It would have been easier to get funding. It would have been easier to get our first few customers. And I say we're one of the largest startups in Canadian history. We're one of 10 companies that can build foundation models. We are, by no means, a large player when you look at the scope of things.

But when I look at what that could have looked like, yes, it would have been easier, but it wouldn't have been better. It wouldn't have been what we wanted to do. It would have been a very different experience, and we would have ended up being a very different company. And I don't love the direction that the companies down there are going in.

Taki Sarantakis: Right. Now, Abdi, in addition to all the other things you do, I don't know when you sleep, by the way, you also spent a long time at something called Blue Jay Legal, which is one of the top AI and law companies in the world, specifically AI and tax. You and your partners, you also did this in Canada. Talk to us a little bit about the same kind of stuff that Nick was talking about, like easier, harder, why stay, why not go.

Abdi Aidid: Yes, so in my prior life, I was the VP of a company called Blue Jay, which initially used machine learning and later language models to predict legal outcomes. So, I wear two hats in these conversations. I'm a lawyer and a law professor, and in that world, I have commitments to such quaint ideas as the rule of law and justice and privacy.

As an educator, I'm training my students to be out in the world practicing law but also hoping that they are entering professions that can be lucrative for them, they can bring their whole selves to.

The whole other side of things, I'm a legal technologist. I still build models that predict legal outcomes. And the way I reconcile that is by saying that I'm not a techno evangelist. I'm just somebody who's convinced that they can be helpful in limited constrained context, especially for things like access to justice, which is a big focus of my research.

What we did in Canada was kind of own a space, and so it's my view that a lot of legal outcomes can be predicted. It's my view that there isn't as much magic to what we do as we think, in terms of the forecasting, that there's a data-rich environment that's being underused, and probably most people's needs are going to be met with a combination of a human and technological solution.

We focus on tax law. The reason we focus on tax law was because number one, you have a codified area of law. There's not a ton of rules –

Taki Sarantakis: You have rules. You have a lot of rules.

Abdi Aidid: – it lends itself well to prediction. You have case law that interprets the rules, and so you have an additional data layer of context, and it's less fraught than other areas of law.

So, one of the big challenges right now, in the legal world, is that people are building legal tools for things like criminal justice and criminal adjudication, which has, let's call it, all kinds of unhygienic things lurking in the historical data. And so, there's good reason to hit the brakes and say, we don't want to project that stuff forward.

Tax, you have a little bit more of a self-contained universe of mostly sanitary legal outcomes. And so, we picked well, in terms of the –

Taki Sarantakis: Why is that company in Canada? Why didn't it leave? Why didn't it go to Boston? Why didn't it go to Silicon Valley? And what would have happened had you left? Would you be bigger faster? Would you have had more customers? Would you have had more capital?

Abdi Aidid: Yes, the truth of the matter is, in Canada, there's two challenges. There's lots of challenges being anywhere, but there's two challenges that are unique to the Canadian entrepreneur. One is the low liquidity investment markets. There isn't a ton of investment –

Taki Sarantakis: Access to capital.

Abdi Aidid: Access to capital is one problem, but the other is that you don't have as much labour mobility in Canada, if this makes some sense. In the US you have the culture of the in-and-outer. You have a person that goes from government to tech to the private sector to whatever it may be, and to people that are being very creative about what roles they can play. And so, it took a while for startups to be an attractive option for otherwise risk-averse, good, well-educated Canadians.

For a while, there was the additional tax that Canadian companies were paying, which is giving up a lot more equity than the modal US company would in order to attract and retain top talent. There were all these ways in which it might seem suboptimal. We had an important reason for needing to be in Canada, or wanting to remain in Canada, which is that we were kind of inventing this important application, which meant being close to the research.

So, the founders are professors. They're people who, in a lot of ways, benefited from the dynamism of places like the University of Toronto, where I work now. And that was critical. We were a company because we were trying to do this newfangled thing in this unexplored area that the epistemic community really mattered and loomed large. If we were trying to build a discrete retail application where all that sophisticated thinking is happening in California, then it wouldn't make sense to be there. There was a local advantage too. It wasn't all disadvantage.

Taki Sarantakis: Good. Bailey, any thoughts on this, in terms of anything at AMMI that speaks to this? Do you guys spin off companies that go to the US, or do you spin off ideas that don't get commercialized here?

Bailey Kacsmar: So, we have like a shorthand that split – there's AMMI Academic and AMMI Not-for-Profit. I'm more AMMI Academic.

Taki Sarantakis: So, we invited the wrong person is what you're saying.

Bailey Kacsmar: No. I mean that there's a lot of work that AMMI does with companies that want to connect with Canadian research expertise. They might say put someone in contact with me if they want some privacy expertise, or if they want specific RL expertise, they might go to say Martha White or Michael Bowling.

So, we have these connections, and AMMI works directly with a lot of companies in Canada, and they go through the AMMI system. But that is run by tons and tons of people, I'm not responsible for running it. So, in terms of the corporate side of things, I'm not particularly involved, but I think there's something to acknowledge.

But just in general, if you're talking like Canadian versus the US, even in academia, US tends to be more money, but there also tends to be other side effects associated with that maybe make it less appealing to work in those spaces, or just a straight up different system. My first thought when you were talking was, could you even take a legal-based AI system to another jurisdiction because the laws would be completely different? Still well-structured, but you'd still have to reformulate it for a different domain.

And so, I think one of the things that that highlights nicely is I like to say we can do really cool boutique AI and technology advancements in Canada, because we have a lot of these really cool specialized domains throughout the country and high-quality researchers working with academia and industry, having them work together can do advancements that don't require this crazy monetary investment that isn't as available in Canada as is the norm in the US.

Taki Sarantakis: Now, I want to shift, staying on the commercial side, I want to talk a little bit about the P word, and the P word is procurement. One of the things that we hear constantly from people in this space is that governments in Canada don't do procurement as well as they should. And they don't use procurement as a policy instrument to further what we talked about a few moments ago, which is sovereignty.

Yesterday – I think it was yesterday – the Prime Minister made an announcement on our new defence investment initiative, and a big feature of that: Buy Canadian. And we haven't really thought that way as a country, I want to say in a long time, but it's been maybe 100 years since we kind of went, Buy Canadian. And we have some protected industries, telecom and some transportation sectors, but this is the first time in a long, long time we have overtly said we are buying Canadian. Talk to us a little bit about the importance of buying Canadian, and then maybe we'll circle around to something else the Prime Minister announced that involved you and your company a while ago.

Nick Frosst: I want to start by saying I'm not an expert in running a government, not an expert in nation building. In figuring it out. These policy decisions are complicated and fraught.

Taki Sarantakis: You don't have enough data, I think. It's not data infused. It's a data problem.

Nick Frosst: I don't have data, yes. So, I'm certainly not an expert in this. I do think from first principles it's a good idea. When I see what has happened to technology in other places in the world, in particular America, I think it's a good idea. It seems to build resilience. It seems to build autonomy by having the government, which is the largest organization in any country, invest and purchase from other organizations within that country. It seems like a good way of spreading this around and building up something that becomes independent at all spatial scales again.

So, I think it's extremely motivating. And to hear that, and to feel a bit of a change in the country that is reflected in the policy decisions, that re-characterizes Canada not as just a little country up there, with being a bank account or being a provider or purchaser from other countries around the world, but instead a powerhouse. Instead, a country that can build and should build and should export its ideas and its technology and its skills and create something bigger, that's extremely motivating.

I think it's the right call and I'm excited to see it implemented. I'm curious, I don't know the history, but I don't know how we got to a place where we didn't do that.

Taki Sarantakis: Well, generally speaking, I mean oversimplification, but we have always kind of thought of ourselves as a middle power, and middle powers follow the rules and rules-based trade agreements and rules-based – the WTO says this, NAFTA says that.

But I think we lost the plot on that because if you go back, it wasn't just the current incumbent in the White House. President Biden, the very first thing he did when he became President of the United States was he ripped up the presidential permit on Keystone XL. Probably the hallmark of President Biden's administration was Buy America. President Biden declared baby food a national security issue at one point, and you could only buy American baby food.

So, this is something that I think other countries – we're kind of late to this party, so to speak. Bailey, any thoughts on Buy Canadian on procurement?

Bailey Kacsmar: The only thing I was thinking as you were talking about that is, and this is perhaps a strange direction to take it, but I grew up in Canada as well, and being, I don't know, 10 years old, grade 5, we're learning about Canadian history and nationalism, and... what?

Abdi Aidid: Is it Avro Arrow? Is that coming?

Bailey Kacsmar: No, I don't know what that is.

Nick Frosst: It's a great story. We'll tell you about it.

Bailey Kacsmar: Sorry.

Taki Sarantakis: People are groaning. We'll talk to you offstage.

Bailey Kacsmar: Sorry.

Taki Sarantakis: That's fine. But you were growing up in Canada?

Bailey Kacsmar: Yes. And so, when the teacher was trying to get us to define what it means to be Canadian and what is Canadian nationalism, all we were able to come up with was we're not British and we're not American. And so, the fastest way to amplify Canadian nationalism is to try to counter one of those two points.

Taki Sarantakis: Right, to say what we're not as opposed to what we are.

Bailey Kacsmar: So, if you have an instance, like many years ago when we had the big push for Canadian beef because of the different testing for cattle, and then you have certain countries making claims about which part of the country Canada is, it revives Canada's nationalism because we're not American, thus we push more Canadian.

And so, to me, it's just a very natural reaction that Canadians kind of cluster back together in opposition of being told, No, you are this thing that is kind of the core tenet of what Canada is not. We don't agree on a lot else.

Taki Sarantakis: Now, Abdi, when I was going to university, we were learning that things like this were crazy. It was like, you don't favour domestic markets. You don't pick winners. You let the market decide. You expose things to competition, and people get better and better and better, and you buy that. And fast forward to yesterday, we're now buying Canadian, at least in certain sectors, but at least now we're talking about, should we buy Canadian? Should these trains be Canadian? Should this algorithm be Canadian? Should this airplane be Canadian? Talk to us a little bit about that.

Abdi Aidid: Yes, if you have a background level of stability and rules that you can trust and agreements that will generally guarantee that there aren't any surprises, then it makes sense to say, We're not going to build military jets here because we can buy cheaper ones, which is better for Canadian welfare if we find the cheapest international producer that's up to our standard. That makes sense until –

Taki Sarantakis: Until it doesn't.

Abdi Aidid: Until it doesn't. And so, part of it is the impetus, the risk calculus for Canada has really changed. So, I understand that. Can I just think about procurement?

Taki Sarantakis: Absolutely, that was what we're talking about.

Abdi Aidid: I'm just looking for— there might be like a soapbox under the desk that I can put on here. I was having a conversation earlier today that I was relaying to Nick in the green room, which is, people are talking about ICE a lot. I'm not going to go on an ICE diatribe here. In the US, though.

Taki Sarantakis: But you're welcome to.

Abdi Aidid: Immigration and Customs Enforcement. And I was watching a debate between some people about whether ICE should exist. There was this question, and its funding was being withheld for a time being, and there were two people in the conversation.

One was saying, well, if you defund ICE, then how could you possibly have immigration enforcement? "You mean you want to suspend immigration enforcement?" And the other person was like, "No, but until ICE gets their act together, et cetera."

And I was looking at this, and I was saying this is totally emblematic of a challenge that we have, a kind of like cognitive and cultural challenge, which is that we can't imagine a world where things aren't the exact way that they are right now. ICE started in 2003. It's younger than "Hit Me Baby One More Time." That song was on the radio and there was no ICE. And so, however you feel about ICE as an apparatus, there was a world before it. And I'm coming to procurement. I'm not relating procurement to ICE, but a little bit.

Nick Frosst: No, you can hang out on the soapbox.

Abdi Aidid: Yes, Yes, I'm going to stay here for a moment. I'm comfortable here for a moment. The way we approach procurement right now is not made for the current world of technology. We have basically taken hardware procurement and used it in a software world, and now we're using it in an artificial intelligent world.

Those are different things. When you're buying hardware, you're imagining a technology that's fully baked and done. If it's broken, you send it back, and they can't send an over-the-air update to improve the CPU or whatever it is. That's not what we're talking about in the context of software, or any of the kind of digital technology.

As a matter of fact, if you have an enterprise-level engagement with AI company of some kind, believe it or not, there's iteration, there's a dialogical process. They're actually desperate to hear from you. Our challenge at Blue Jay for a long time was that people weren't telling us enough about what they wanted. They weren't passing along enough information to us so that we could make appropriate fixes because the thing was not fully screwed in all the time in the way that your CPU is if you're buying enterprise hardware.

That should empower public servants to think through, okay, if it is the case that when we purchase technology that there's back and forth, that the company wants to hear from us because their development cycle involves some iteration, that actually they're interested in this enterprise-level engagement because they want to explore the possibility of purpose-built tools. There could be some bespoke elements, but no matter what, they actually want to hear from us. That gives you an opportunity to pass along information in a way that you were never really able to intervene prior.

So, that's one thing. The procurement processes that we use in all governments, this one no exception, are a little unduly restrictive because they don't contemplate that as an important part of the process. That's one.

Taki Sarantakis: You are so polite. A little, what did you say? A little unduly restrictive.

Abdi Aidid: I forgot what I said. I don't want to repeat it. I forgot what I said.

Taki Sarantakis: A little unduly restrictive.

Abdi Aidid: I can't remember. There's no recording either!

Taki Sarantakis: Do you work in the government?

Nick Frosst: A little unduly restrictive?

Abdi Aidid: This year.

Taki Sarantakis: Oh yes, this year. You're a visiting scholar?

Abdi Aidid: I've adopted this speak, yeah.

And then relatedly on procurement, there's never a – I'm thinking about remember Fast and the Furious? There's like 10. Yes, there's like 10 of them.

By the way, I remember I watched like 1 and 2 and then I watched 10. And in 10 at some point, Ludacris is getting launched into outer space. So, I was like, I thought this was a show about street racing. A movie about street racing.

Anyway, they had this thing –

Taki Sarantakis: Is this related to your band? [crosstalk]

Abdi Aidid: Oh, okay. They had a thing called the –

Taki Sarantakis: I don't know where you're going yet.

Abdi Aidid: Neither do I. Let's see where I'm going with this.

Taki Sarantakis: Let's see where it lands.

Abdi Aidid: They had a thing called the NOS button. Remember that? Where they're racing and then they hit the button, and they just get catapulted forward.

Taki Sarantakis: Is this like what old people like me would call the turbo button?

Abdi Aidid: Yes, turbo button basically. So, we desperately need that in the world of procurement for two reasons. Number one, because there is the potential, believe it or not, of a lockout of the public sector from competitiveness when it comes to service delivery.

Why? Because almost all embrace of AI is like a day too late, no matter what. It's even a little bit late in the private sector, but the fact of more nimbleness in the procurement process has meant earlier exploration. And so, there's a personal risk to public servants that don't get early enough AI exposure that they under-index on transferable skills if they want to go do something else at some point in their life. So, that should appeal to the self-interest.

But the other piece of it is that by the time it gets mature, and all the components of your agreement are signed, and the thing is ready, there's highly likely to be a top-down mandate that you use it everywhere, all the time. And so, you've missed out because of the glacial pace of procurement on what would have been the exploration, experimentation moment when you could meaningfully provide the feedback. And so, you got to hit the turbo button on procurement.

And I'm not sure exactly how that – I mean, I have some ideas about if you want to hear about them later, about how that can happen, but it has to. It has to.

Taki Sarantakis: Yes, I agree. Now I want to circle back to you. I hinted about this before we started the procurement discussion. Yesterday, the Prime Minister announced, as we said, the Buy Canadian. A little while ago, I forget how long ago, he signed something with you, or he announced something with you. What did he sign or announce?

Nick Frosst: Yes, Cohere. Yes, we signed an MOU with the government on building AI and using it within the government. And that was something – we were enormously proud of that MOU, and we take that duty with a lot of weight. We're very excited and honoured to be able to collaborate on making AI useful for the government. That letter is a letter.

Taki Sarantakis: I was just going to ask, we have a lawyer here, what's an MOU mean?

Abdi Aidid: An MOU is effectively an expression of mutual interest, and it specifies some terms, and they're binding insofar as they've crystallized what you said to each other, but the MOU has to be superseded by some ultimate agreement, like a purchase agreement or something like that, which we haven't signed yet.

Taki Sarantakis: Now, Nick is very polite and very everything. I'm less polite and I'm less all of that stuff. I'll tell a little story. The day after the MOU was announced, I called somebody and I said, "I want the Canada School of Public Service to be the first organization to use Cohere." And I was told, "You can't." And then I said, "But the Prime Minister yesterday, blah, blah, blah." It's like, "You can't." And I'm like, "So we announced something, da, da, da." "You can't." Click. So, this is kind of one of the things, I think – because a lot of bureaucrats are listening to this, a lot of officials – I think one of the things we don't understand as officials, we don't get this in our bones as much as we should, which is time is very different outside of officialdom land.

The time from an MOU to a procurement vehicle or something, we might think, Oh, if we're really quick, that's 2 years, or, Oh my God, we worked really hard, we did it in under 3 years. That's a completely different sense of time outside of governments. Talk to us a little bit about time when you were an entrepreneur, and talk to us a little bit about time, because you're kind of a current live entrepreneur.

Abdi Aidid: Yes, so the government procurement cycles were long, but the sales cycles in the private sector can be long. You can want to have an enterprise deal with a private sector client, and it might take them a long time to agree to buy, but once they do, then it was about papering it over, and it was formalities.

Taki Sarantakis: Right. So, you date a long time, but when you get engaged, you get married quick.

Abdi Aidid: You get married pretty quickly, and it's kind of inverted in the public sector, where actually they tell you right away, "We really like you."

Taki Sarantakis: We really like you. Yes, we really like you.

Abdi Aidid: Actually, I'll say yes if you propose right now.

Taki Sarantakis: I love you. I've never seen anything as nice, as lovely as you are.

Abdi Aidid: Thank you. Thank you. All right, I'm ready.

Taki Sarantakis: What do we do now? Let's talk about it for a couple years.

Abdi Aidid: There's a venue that I like. It's available in 3 years. It's the only place I want to get married.

Taki Sarantakis: Okay, but I really want to –

Nick Frosst: We have to end this skit soon or [crosstalk and laughter]

Taki Sarantakis: No, but we won't, because one of us is private sector, one of us is public sector so we will date for a long, long time.

Abdi Aidid: I think the key is the hard part is supposed to be over.

Taki Sarantakis: Yes, because we like each other.

Abdi Aidid: Like we decided we like Cohere. We decided we want a sovereign Canadian AI solution. We have one of the 10 companies. They're purporting to actually build things that are special for us –

Taki Sarantakis: So now it's just papering the transaction.

Abdi Aidid: It's supposed to be –

Taki Sarantakis: It's going to City Hall and getting the piece of paper.

Abdi Aidid: It's supposed to be.

Taki Sarantakis: So, are you at City Hall? Have you got a piece of paper?

Nick Frosst: I'm not at City Hall now.

Taki Sarantakis: You're not at City Hall now.

Nick Frosst: No. I mean, look, we –

Taki Sarantakis: But talk to us about time. Let's go back to the seriousness.

Nick Frosst: Yes. Yes.

Taki Sarantakis: What does a year mean to you? What does 6 months mean to you?

Nick Frosst: We've been a company for 6 years. This technology has been in the public sphere, in the mind of people for, I don't know, 3, 2, like not long, not long at all. I talk a lot about AI and about language models as a new industrial revolution. And there's been previous industrial revolutions, previous technological revolutions. I think about like the advent of the personal computer, the printing press, the steam engine, we've gone through a lot of these things that fundamentally changed how we do work. I think this is another one of those things. Each of those things was actually a pretty long period of time. We narrativize it as though it's like, Yes the Industrial Revolution happened and suddenly it was night and day.

Taki Sarantakis: Electricity everywhere.

Nick Frosst: It takes a long time. So, we're still in early days of this technology and there's a long way to go.

That being said, it's moving so quickly elsewhere. And we work with some companies, like we work with RBC as a customer of ours. We've deployed North, which is our agentic platform, like the thing that employees use within RBC to automate their boring tasks. We've deployed that very quickly to a huge number of their employees, and they're a 150-year-old company. And they move incredibly quickly for that scale. We've deployed North to companies that have rolled it out to like 10,000 employees immediately after we deployed it in a safe and secure way.

So, on the one hand, we're early in the technology, and there's a long way to go. On the other hand, the rest of the world is moving really quickly, and we are falling behind. So, I don't know what is normal for government, and I'm not an expert in government, and I have never worked there, so I don't know what the standard time scale is. But I do know that we're at a disadvantage right now, and that it would benefit from being faster.

Taki Sarantakis: Now, Abdi, back to time. I've heard you say something very, very interesting, and I think it's important for people to understand this because it dovetails nicely with what Nick just said in terms of the world moving really quickly. You say that when we look at a given AI application today, it's like looking at a star. Talk to us about— unpack that for us.

Abdi Aidid: Yes, you think you're looking at the present, you're looking years in the past. Today's AI is yesterday's R&D. I don't know what Nick has going on right now that he's planning to deploy in 2 years. I'm sure it's going to be jarring. And the key thing, I think, is that we also have to be able to anticipate.

So, a technology that Cohere comes to us and says, hey, we're ready to roll this out in your organization, is technology that they spent the last couple of years working on. And they might have something new for us that's incrementally better or considerably better in 2 years. But in those 2 years when the new thing is available, we'll be maybe 65% of the way through our procurement cycle on the first thing.

And so, we are already sort of like epistemologically in a weird place vis-à-vis AI because it's all subterranean. It's not like drugs that need FDA approval or something where we know verifiably what's in the hopper and what's coming. We don't know. The development is subterranean until it's ready. And so, in order for us to better contend with even the ethical challenges, even the legal and policy challenges, we have to be –

Taki Sarantakis: We have to get going. We have to start interacting.

Abdi Aidid: We have to start running the race a little bit alongside technology because we're being outpaced in ways that I think are deleterious for the big picture concerns that we have.

Taki Sarantakis: Yes, and it's interesting because if you buy today's technology, but you don't deploy it for 3 years, it's already out of date. And I'm not talking about like Cohere or anything like that. It is, this area is moving so quickly, so rapidly that at a certain point, I think what we're saying is we're chronically, in the public sector, not just in the Government of Canada, but in the entire public sector, we risk being left behind structurally, if that makes sense.

Bailey Kacsmar: Well, I mean, if we're talking about – I always get stuck in these conversations of our technology is moving so quickly. And all these other things fall behind. But it kind of creates a false race. The technologies, even as you said earlier, this isn't a CPU. This stuff can be updated. It's only an issue if you treat like version 1.2 is significantly different than 1.3.

But if you formulate it as a piece of software that's supposed to provide certain functionalities and can have these additional ones, if you give bounds about what is and is not acceptable behaviour for the software, you don't need to be playing this race game.

The internal workings of, say, a particular LLM is not what is impactful to, say, people using it in the public sector. They care about what comes into it and what goes out of it and the restrictions on that. Is it something that is okay to provide internal information to? Yes or no? Is it something that outputs things that are reliable, or because it's a probabilistic technology, is it something that we shouldn't be using in these settings that require verification?

All of these things are something we can formulate instead of getting stuck in this loop of, oh, it's changing so fast. You don't actually care about that. We do, because we have to make it and deploy it, but that doesn't have to affect the deployment and use of it. You get updates to your phone all the time. And most of the time, most of you have it set as automatic. That's good. You don't want to miss a security update. But you don't notice. Your phone updates, you move on with your day. The software, as long as it's not completely changing its functionality, should be doing the same.

Taki Sarantakis: Oh, you clearly don't work in the government.

Bailey Kacsmar: Oh, I don't, but that doesn't mean that the way it is has to be the way it works. You don't have to push yourself into this corner where we're always having this race. You can define the requirements for the software without saying it has to be version 1.2.

Taki Sarantakis: We buy the AI and that's good for 10 years. But anyways, that's another –

Abdi Aidid: Thinking about procurement more flexibly than not over-specifying in the procurement process, anticipating the possibility of things like updates. The big challenge right now is the cycles have been so long that now wholesale new things have been added, like an agentic capability. And so, we're still talking about the language model, and then now there's an agentic feature. That's big.

But most things are not going to be big in that way. And so, if you design your procurement to anticipate some change, I think Bailey's point is super well taken.

Taki Sarantakis: Yes. Now we're down to our last few minutes, and so I want to close by – I want to give each of you the opportunity to give the people in the room and the people online – I want you to give them some wisdom. And they're officials, they work on these things. The fact that they're listening to you today, taking an hour and a half out of their job, probably means that they're more interested than the average public servant in these things.

So, I think you don't have to tell them about the importance of this, but tell them something that you think that, as officials in the Government of Canada, they need to know in this space, or they need to know a little better? Who wants to start us off? Bailey, do you want to start us off?

Bailey Kacsmar: Sure. Normally the middle's safer than this. All right. So, if I was to go with anything in this space, when we're talking about data and AI, we've had a lot of conversation today about using it and keeping up with it.

But if I could add one point that we haven't touched on, or at least not explicitly stated, is that it should always be purposeful. We shouldn't be saying just adopt it wholesale. We shouldn't just be saying find a way to use AI in your space, in your domain, no matter what, because it's not necessarily a right deployment.

There's so many useful ways we can use AI, whether it's an LLM or something a little more old school or conventional that can really benefit a system, but only if you understand the domain you're working in. And as someone working in that domain, you should understand it and see what would be useful here, what type of automation. Because largely when we're talking about any form of AI, it's ultimately a form of automation. What task it does and how it does it is varied, but ultimately that's what it is.

So, what purposeful way can you use it rather than just kind of trying to scattergun and hope we just throw AI at everything and see what sticks. Because that's when we get into these messy situations or we end up with trust issues where people are like, oh, well, you told us to use it, and we used it here and this not great thing happened. And then you have to build back.

As soon as you lose trust, whether it's a technology or a policy, it's so much harder to get back to using it effectively. And given the potential of how we can use these things, you don't want to be ending up in that situation where you have to rebuild the ground floor.

Taki Sarantakis: I love that, Bailey. Abdi?

Abdi Aidid: That's brilliant. You should listen to Bailey is my advice. But really, I really love this idea of think about effective, safe, helpful uses of AI. And by the way, that also goes to frivolous uses of AI. Sometimes I use GPT to write little funny limericks for my daughter, and she asked me one time about whether I could show her an animal juggling, so I made it do an elephant juggling. That's kind of a frivolous use of AI, and there's some environmental consequences that we haven't discussed that I think people should also keep in mind. So, maybe throttle your frivolous uses.

Invest in your own credibility is some important advice I would give you, which means also being a credible objector. I've made this point ad nauseam. Ask yourself what things in your life are you absolutely committed to doing the analog way, and you'd probably find that it's a shorter list than you think. I think about this in the context of my frustration with critiques of AI that are premised on what the technology can and can't do today. You have to hold open the possibility it'll be able to do it tomorrow, and the question is, do you still object?

So, I have a list. For example, and some of you heard me say this before, I'm against the use of AI in criminal sentencing, bar none, never. You could build the optimally deterrent algorithm that is perfectly debiased and doesn't project racist histories forward, and I would still be against it. Why? I don't think we should optimize sentencing. I think it should be hard. I think that someone should go to sleep at night thinking about what they're about to do the next day if they have to sentence somebody. I think it should weigh on their conscience about whether or not it's appropriate to deprive someone of their liberty. I think it's a communitarian thing that we should all struggle through.

So, the question is, how many of those things do we have? There's nothing you could tell me about the technology's improvement that'll make me want to use it in that context. But I can't use that example where I think there would be a deployment that's too socially fraught to object to using a summarization tool in the office. You lose credibility when you do that.

So, the question for you is, what are your no-go zones? Jealously guard those and consider the possibility of lower-stakes absorption of AI in your life. I would say that that's how you don't eliminate yourself from the debate, because right now there's an important conversation to have about the future of work and what it looks like. We're all participating in the chance to remake what professions look like. And a lot of people are taking themselves out of the conversation because they're against the whole enterprise. They're not disambiguating between appropriate and inappropriate uses.

So, I would encourage you to think through, what are your no-go zones, and focus on those.

Taki Sarantakis: Nick, close us off.

Nick Frosst: I loved both of these. Both of these are great. Yes. We opened this talking about how there's a 12% adoption rate and now we're the lowest in the G7, or something. So, even though you're here, and even though you're listening to this, and you've chosen to take time out of your day to listen to this, I assume that puts you in a group of people that are more interested in using it.

I do think, in addition to what you've both said, it is good to say, Yes, you should think about what you don't want to use it on, and you should also think about what you do want to use it on. And statistically, not very many of us are doing that in Canada, and more people are doing that elsewhere. I think that's kind of generic, that goes without saying. I think the thing I'd like to add to that is I've recently been trying to encourage people to think about this technology as not beyond them. It is not that complicated.

And I know that sounds crazy. A lot of people get their arms up in the air about that. But learning how to program a little bit has never been easier. It's never been easier in your life to learn a little bit of Python. And I hear people talk about learning how to program, or learning a programming language, completely differently than they talk about learning Spanish.

I've never heard anybody say, "Spanish? I couldn't possibly, oh my God, I could never learn" or like any other language. It's hard, it's really hard to learn another language, but you all know what needs to get done to do it. And if you wanted to learn a little bit, you could. And that would help you. That would serve you a lot. Even just learning a tiny bit of a new language to a country you're going to, or even in Canada, in our case, with such a multicultural country, learning any language is going to be helpful here. And so, people do it.

I think we should think about programming in the same way. I don't think you should think that programming is beyond you or something you couldn't have done. It really has never been easier to learn a little bit of programming, and LLMs are a great tool for learning how to program.

After that, understanding how a large language model works, how like today's AI works, is also not that complicated. Creating them, wildly complicated, super complicated, and I don't encourage you to try. But understanding the basic mechanisms of it, definitely not beyond you. You can definitely do it, and that will help you when figuring out what you want to use it for, what you don't want to use it for, where you want to put it in your life.

So, I encourage you to do both of those things.

Taki Sarantakis: So, I want to ask the audience to join me in thanking Abdi, Bailey, and Nick for spending a wonderful hour and a half with us. And I know we're all leaving a little bit more knowledgeable and maybe even a little bit more hopeful in this area than we were when we sat down. Thank you.

[01:29:04 CSPS animated logo. Text on screen: canada.ca/school-ecole.]

[01:29:10 The Government of Canada wordmark appears.]