Business Data Lab "Canada's Economy, Explained"
EP 10
SPEAKERS
Avi Goldfarb, Marwa Abdou,
Avi Goldfarb 00:00
So I'm very excited about the technology, and I'm very excited about Canada with this technology, we do have amazing talent in this country around AI, coming out of multiple universities, still have many of the leading scholars for the technology. And while you might say, Who cares about the scholars, the scholars train the students who then build the companies. Okay? And that has been what's happened for the US companies and for some of the Canadian companies that have been thriving. And so I think there is an incredible opportunity now that corporate Canada seems to be waking up, that the government seems to be waking up to what this technology can do. And I think it could, with the right investments and the right encouragement and rhetoric, lead to better health care for Canadians, better education for Canadians, better government services for Canadians, better financial services for Canadians, better retail for Canadians, et cetera, all across the board.
Marwa Abdou 01:01
Welcome to Canada's Economy Explained. I'm your host, Marla Abdou. As we continue to unpack the forces shaping Canada's future and what they mean for businesses, workers and policy makers, we're turning to one of the most talked about and most misunderstood drivers of global economic change technology, and specifically artificial intelligence. And the timing for this conversation couldn't be more critical. Just last week, g7 leaders released a major joint declaration on AI, recognizing it as a driving force in global competitiveness and calling for stronger international guardrails to manage its risks and unlock its benefits. The world isn't waiting from global AI alliances to national AI strategies. We're seeing a race not just adopt AI but to shape its direction. But we're not talking about sci fi robots or viral click bait. We're talking about the real economic disruption that AI is driving, and the opportunity for Canada to lead, if we get it right. Because this isn't just about algorithms, it's about power, and as AI transforms how economies operate, the epicenter of global influence is shifting away from legacy g7 players and toward emerging tech powerhouses like India and China. The global south is no longer watching from the sideline. It's investing. It's scaling, and in some sectors, it's leapfrogging. AI is fast becoming a new arena for economic strategy, and in a world that's grappling with deepening divides, humanitarian crises and rising instability, AI is becoming a new arena, not just for competitiveness, but also a fault line for inequality. So where does that leave Canada? How do we go from potential to power? What does it take to embed AI into how we work, how we govern, how we compete globally, and why, despite our early leadership in AI research has Canada lagged in adoption, in scaling and in productivity gains to help us make sense of all of this, I went to sit down with one of the world's most influential thinkers on the economics of artificial intelligence. Avi Goldfarb, he is the Rotman chair in artificial intelligence and healthcare at the University of Toronto's Rotman School of Management. AVI is also chief data scientist at the Creative Destruction Lab, which he's co founded alongside Ajay Agrawal and Joshua ganz just to give you a few headline stats since its founding in 2012 CDL has enabled over 3500 tech startups to launch. It's generated approximately $30 billion in cumulative equity value. It's expanded across 13 global locations, blending academic insight with real world outcomes. This Prius leadership, Agrawal, setting the mission. GaNS, embedding rigorous economic framing and Goldfarb, delivering data driven execution has really made CDL a powerhouse in scaling deep tech innovation. Now, if that wasn't enough, this impressive team has also published two landmark Harvard business review books, prediction, machines, the simple economics of artificial intelligence in 2018 and power and prediction, the disruptive economics of artificial intelligence in 2022 Avis and his co authors, work has fundamentally changed how business and policy leaders understand AI, not as something mystical or magical, but as an economic force, a technology that dramatically lowers the cost of prediction, and in doing so, it reshapes how decisions are made across sectors and societies. And that reframing matters, especially for those we've long feared would lose the most, because when you strip away the hype. Right?
AI is ultimately about control over decision making in a world where choices about hiring, about credit, about policing, healthcare and public services are increasingly automated, AI can reinforce existing inequalities, or it can help us redesign systems to challenge them. Avis work forces us to ask not just where AI is going, but who it's serving. That's why his voice matters more than ever, especially for Canada as countries around the world race to capture AI's economic and geopolitical advantages. Avi brings a calm and a clear lens that cuts through the noise. He reminds us that AI doesn't just drive change on its own. Institutions. Do people do and the decisions they make matter just as much, if not even more, than the technology itself, as we heard in our last episode, and we'll hear again in this one, there are layers to even the most familiar ideas. Part way through our conversation, Avi brings up Bill C 27 a proposed law aimed at regulating AI and strengthening privacy protections. It matters because it sets the guardrails for how AI can be used in decisions that shape our everyday lives, like whether we get hired or qualify for a loan. These are the building blocks of power in the age of AI, shaping who has control, who benefits and who gets left behind. We began the conversation with a big idea that stuck with me since I first read prediction machines in 2018 that AI lowers the cost of prediction. It's not about replacing people, but about improving how we make decisions. I started by asking Avi, why was that framing so important to him and his co authors back then, and how well does it hold up today in this new era for Canada and the world? Let's dive right in.
Avi Goldfarb 06:59
We started with our understanding of technological change. And so we're economists. We study technology. And there's a long tradition in economics where when you think about technological change, it's something fundamental has gotten more effective, more efficient. So there's economists who studied computing and computers and said that made arithmetic cheaper, and there's others who study the internet and said that was search and communication and copying. And so we were trying to think through what, what did this new generation of machine learning tools that we're calling AI, what were they? And under the hood, it was computational statistics, it's prediction. And so we thought through and read the literature and in terms of what people were actually doing with machine learning with these AI tools, it was pretty clear that many of the applications were prediction, like whether someone's gonna pay back a loan, and even those that didn't feel like prediction, whether it's medical diagnosis or writing or coding under the hood, prediction was what was happening that has not changed with Gen AI. And so one of the challenges in using AI tools in business or or elsewhere is to recognize that under the hood, it's prediction technology, even if your intuition isn't that. Oh no, writing is a statistical process, because engineering wise, it turns out now, writing is a statistical process.
Marwa Abdou 08:20
You know, as you mentioned, you've always drawn this sharp distinction between prediction and judgment. Prediction is it input into decision making, as you put it, but with large language models, llms now generating everything from business strategies to marketing copy to software code, we're seeing that boundary becoming increasingly porous, and we're witnessing this blurring of the lines between prediction and decision making. Do you think that this moment calls for a sort of redefinition of prediction itself, one that accounts for, perhaps, outputs that increasingly mimic human judgment, that mimic human creativity, and even just in terms of strategic reasoning?
Avi Goldfarb 09:14
Bluntly, no, but I'll tell you why. So what the Gen AI models are doing is they're changing who provides judgment. So in decision making, in most organizations, it's individual managers or individual workers who are earning predictions, maybe from an AI, maybe not with their own judgment to make a decision. When you start automating and when you start using an LLM or another Gen AI model, what you're doing is saying that judgment is no longer happening in my organization. The judgment is happening at the vendor, whether it's by open AI or anthropic or here or somebody else. And so there's still absolutely human judgment in those models. There's humans who are deciding what good output looks like. And in fact, a lot of the effort in designing good. Good llms is about the team at OpenAI or anthropic or cohere or wherever else, judging what good output looks like. It's not that there's no human anymore, but the human is different, and they're probably not in your organization anymore, and that is a real first order management challenge.
Marwa Abdou 10:18
There was a time not so long ago, when the dominant narrative around AI was essentially, smart robots are coming to take our jobs. The message was, brace yourself. Replacement is inevitable. But that story has since evolved. We've come to see that the reality is far more nuanced. Uptake has been uneven, maybe even slower than expected in some areas, deeply disruptive in others. Looking back now, do you think that there were blind spots in how we imagine that AI would transition, whether it is among policymakers, business leaders or even economists? Did we underestimate the institutional changes, the risks, or even the timeline that real transformation would require?
Avi Goldfarb 11:09
There's lots that we got right, and there's lots that we didn't. The economists who were thinking about it understood that this distinction between automation and augmentation is a false dichotomy, because almost always one person's automation is another's augmentation. If you automated, for example, the diagnosis process, that's going to automate a lot of things that physicians do, but it's probably going to augment your nurse and your pharmacist. And so this challenge between recognizing that automation doesn't mean human replacement, and even when it does, it might mean replacing one human with another human or team of humans. And so don't get me wrong, this is not a statement about disruption, because that could still be incredibly disruptive, but it's the idea of the end of work and the irrelevance of humans over time. Most economists who are thinking about this, including, you know, including me, didn't, didn't really, you know, go to that extreme. What I think we under appreciated, or at least I under appreciated, was the speed of technological change. So even since November 2022, when chatgpt was released, the pace of change has been extraordinary, so at some point that is likely to slow down. But so far, we're two and a half years in, and it's still going strong. And that's been incredible. And it's been incredible for sort of thinking through what the models are gonna be able to do and and how, you know, our work lives can be changed and our productivity improved for the better. And depending on who the hour is and our work lives, there's gonna be better and worse. There's gonna be winners and losers, for sure, the thing that is still an open question is how fast the transformation comes in most organizations.
So like we look at economic history, it's likely to be slow. So just because the technology exists and it's fantastic doesn't mean that most organizations gonna be able to do it. So the 1990s we had the.com bubble, and everyone was all excited about the internet. That bubble burst, but many of the predictions were right. They were just a little early. The Internet did transform how we lived and how we worked. But, you know, the many of the companies that got the biggest boost from the internet didn't exist in 1995 or is 96 or 97 and Google was later. Meta was, Google was came out of beta in 99 meta was in the early, 2000s and Apple didn't really get the benefit until the late, you know, somewhere between 2007 and 2010 so it the predictions were, were right, but the transformation was slow. Many people in industry, on the tech side, especially, say this time is different and it's going to be fast. I'm skeptical because of what we see in the past. I don't have a there's no it's not definitive. There's no formal proof that says technological change and adoption is always slow, and maybe this time will be faster, and we'll see.
Marwa Abdou 14:08
In Power and Prediction published in 2022 you describe AI as being in the in between times that transitional period between the invention of a transformative technology and it's widespread adoption. Canada was early out of the gate in terms of funding fundamental AI research, but we're lagging in terms of adoption. You talk about this diffusion problem. So what we mean by that is that in the context of AI, it's that we are referring to the lag between the development of a cutting edge technology and their widespread adoption. Can you unpack what that diffusion gap looks like in practice, and what's holding us back?
Avi Goldfarb 14:53
Yeah, so with technology after technology, it's clear that it just takes time. And the reason it takes time. Is we often have to reinvent the way work happens, the way organizations operate, in order to take advantage of the technology. So in power prediction, we emphasize a discussion of what happened in factories of electrification. So factories in the 1880s were powered by steam and water. Factories in the 1920s were powered by electricity. But it took 40 years for that transformation to happen, and the reason it took so long is because the factories the 1880s weren't designed for electricity. They were designed for steam and water. Electricity wasn't that much cheaper, it was just different. It allowed you to put the machines where you wanted. That led to a different kind of factory. And so right now, it kind of feels like we're in the 1890s it's clear that this is a great technology, but we haven't figured out what the organization of the future looks like. What does it mean if we have machines that can write for us, that anytime we need to fill in a form at work, the machine can do it for us and do it faster, and in some cases do it better. What does it mean that we have better demand estimation, better predictions of inventory and all this for how the organization functions? And since we're still figuring it out, it's hard to say to one organization, this is what you should do, and this is how it's really attempting for especially the non tech companies to just wait. Let's let someone else experiment, make mistakes, and then we'll be and then we'll follow. It's an old Canadian strategy, figure out what the Americans do, and then fast follow and apply it. Here. There's real risk in in software doing that generally, and an AI in particular, because, because software scales so easily, once someone else figures it out, they might be able to come in and take your customers.
Marwa Abdou 16:46
We're seeing some interesting provincial moves. Quebec has doubled down on AI research hubs and commercialization through Mila. Alberta has been exploring AI and digital health and diagnostics. Ontario has been investing in modernizing public service delivery with AI enabled platforms. Are these efforts right on track? And What lessons do you think we can take from them to scale more nationally, especially if Canada is intent on leading in terms of responsible, inclusive AI.
Avi Goldfarb 17:22
I'm really excited about what we're seeing in Ontario and Quebec and Alberta. It's been great to see provinces and also the federal government investing in AI's potential. The same time, I do see a push to regulate, and much of the regulation is to add bureaucracy to anybody who wants to use AI for anything consequential, whether it's healthcare or education or financial services, anything that actually would matter in people's lives. And I worry that the investments we're making to make AI happen in public services and healthcare and elsewhere are going to be exactly made illegal, or at least so complicated that it's not worth the bother by the additional layers of bureaucracy and so so far, I haven't seen many of those bills passed, but I know that there's a real push because there's genuine fear about AI. Yes, yes, there's still humans making decisions, but it's different humans, and those humans may not be in Canada. Ultimately, we might want to think through there's a push to regulation, but in that process, I do worry that we're going to make the very investments that could transform public services and our economy much more difficult to actually implement. Make a difference.
Marwa Abdou 18:44
Who do you think is doing it well? Who can we learn from?
Avi Goldfarb 18:47
Americans have done AI regulation very well, in my view. Okay, I know that that is not a universal view. The European academics think the opposite of my point of view on this, but my take is they are enforcing the law, as in, if somebody is using AI for something that is against the law, then they get in trouble for that. They are enforcing competition. They're taking on Google and others to make sure that the tech market in general and the AI market in particular, remain competitive, and they're not creating a layer of bureaucracy, of approvals for using AI for all sorts of things where it could really make a difference, like I'm most excited about AI and healthcare. Healthcare has been healthcare. Productivity has grown very slowly, healthcare has become much more expensive, and it's got a little better, but not enough better, perhaps, and better diagnosis tools with AI, better operations, operational structures, better staffing. AI could really make healthcare better. I. But healthcare is really consequential, and so I worry that our risk aversion in healthcare is going to make us accept a mediocre reality that we have now, for fear of maybe that mediocre reality being a little more mediocre relative to the potential for being much better.
Marwa Abdou 20:18
I want to pick up on that sectoral focus so most of the AI conversation today revolves around, you know, productivity, automation, innovation. Looking ahead, do you think that we're going to see AI reshaping the way industries work? And, you know, even pulling back in terms of the broader picture, even reshaping global power dynamics. Could we be entering a world where countries are no longer competing on just GDP or military strengths, but on training data, on compute infrastructure, on rule setting authority in terms of AI,
Avi Goldfarb 21:01
Okay, I'll ask the answer the second part of that question. A lot of those things, in my view, are about GDP or military superiority or both. And so the dimensions of that competition are changing. But I think the fundamental aspects of it, it being like growing the economy and ensuring sort of secure supply chains haven't gone away as the world gets more dangerous, the secure supply chain thing is clearly a bigger deal with more emphasis than it was 1020, years ago. But that doesn't have much to do with AI. There's a political scientist named Paul Shari who has a recent book called Four battlegrounds, where he emphasizes these, you know, the different dimensions of competition. And AI, those dimensions are, in many ways, good old fashioned industrial policy, which is, you got to make sure you control the chips. You got to make sure you have compute. You got to make sure you have good judgment, and people who know how to do things right, like that's, that's something we've we've understood for a long time. AI is the latest and greatest technology in that long Ultra in that old tradition. And on the econ side, we're seeing something very similar, which is, yes, data is an important part of competition today, and that you need good data in order to train good models. But that's a resource, and we know we have good economic models of how to think about resources now we have a different one that has some slightly different econ characteristics from like oil, but, but it's still a resource and it's valuable, and it's going to be an element of economic strength.
Marwa Abdou 22:48
Going forward, we're seeing a growing body of research from MIT McKinsey, the OECD, highlighting how AI is driving sectoral shifts. Retail and financial services are already well into their transformations, while sectors like construction, logistics and even healthcare, as you've noted, are just on the cusp. Where do you see the greatest risks for disruption are as AI continues to evolve? Are there specific sectors or types of jobs that should be especially on alert and on the flip side, where do you see the biggest opportunities? What kinds of new roles or functions might emerge out of this technology? More broadly, how can we and how should we be thinking about the ripple effect around re skilling, around wage polarization and economic inclusion, especially in a world where the cost of prediction is dropping but the value of judgment and adaptability is rising.
Avi Goldfarb 23:50
Great set of questions. So let's start with what's right for disruption. Now the three sectors that have had very slow productivity growth over the last 50 years are the public sector, health care and education in all three cases, a lot of what you do is filling fill in missing information as prediction. And so prediction tools have the power to transform all three the same time. There are real vested interests fighting change, and there is real risk in doing things wrong. And so while those are probably, in many ways, the sectors with the most potential for AI to transform what what we do there, it's also going to be really hard, and it might be slow. So like there's this strange dichotomy between the sectors that I'm most excited about AI in, I can also imagine best interests and regulation and other things making that transformation quite slow. On the other side of things, you have, for example, computer programming, coding and. And AI makes coding much, much more efficient, okay, and it empowers, in many cases, mediocre coders to be much better. And so how does that play out in a workforce sense? Well, right now we have, like graduates the University of Waterloo getting job opportunities and the University of Toronto, I should say, getting job opportunities south of the border for, you know, $200,000 US or more.
Okay, that's amazing. There's a sense that at least over the next few years, there's gonna be fewer entry level jobs like that. Okay, so that's disruption, but there's also a sense that there might be many times as many coding jobs at 75,000 a year, okay, coming out of college, there's disruption on the one hand, and there's opportunity in the other. And wages can fall, and, you know, employment can rise, depending on the details of how the technology plays out. We could see this. We could see wages rise and employment rise, you know, or one increase in the other decrease. There's something Jay Josh and I talked about, an old paper in it came out in science a couple of years ago, and then recently, as an economist at MIT named David otter, who's started to measure this carefully on thinking through what does expertise mean? How does technology change expertise and recognizing this, like the wage employment, the differences between wages and employment, and you can have great news on employment and bad news on wages, like we did in taxis. So a lot more taxi drivers today, if you include Uber and Lyft, but if you were a taxi driver before that, your way just went down. But we have millions more people with that job. And for everybody who wasn't a taxi driver before, who now has that job, things are probably better. And you know the computer science, the programmers grad, those grads, it's not quite taxi driving, but the top might have their wages go down, but it might create opportunities for lots and lots of others.
Marwa Abdou 27:07
I want to shift gears to ethical considerations and regulatory frameworks so there's a growing body of research into the environmental and social footprint of AI systems from your vantage point, what are the most pressing ethical or safety concerns that are tied to the next wave of AI that maybe we're we're not thinking about or we're not necessarily giving appropriate weight to.
Avi Goldfarb 27:37
So, the most important thing is to recognize that a machine isn't making decisions. Even when things are automated, there is a human responsible. There's human judgment embedded in every output from an AI. And as long as the law makes it clear that there is a human responsible, that the companies aren't allowed to say, Oh, the AI did. It wasn't my fault, then I think we'll make great progress toward responsible and ethical AI and to Bill C 27 the Canadian AI act. That idea was central to the act, not sure if you know, it hasn't been passed yet. It's been around for a while, but when I read that act, I was really encouraged, because that central point was what everything else was built on. There's, like, there's some nitty gritty around the edges that I didn't love, like, it wasn't clear that if their definition of what a of what AI was might include calculating an average the way you might have done in seventh grade. But maybe reasonable people would read the law and say that's not how it will be enforced. But generally, the first and most important ethical consideration is to recognize that machines aren't making decisions.
Marwa Abdou 28:51
The new prime minister has signaled an ambitious agenda, from creating a dedicated AI ministry and appointing an AI minister to mandating AI deployment across government, as well as advancing digital trade and critical minerals infrastructure under the one Canadian Economy Act. How well aligned Do you think these policies and priorities are with what Canada needs in this moment in time, and how well do they place some of those guardrails in place that you've mentioned.
Avi Goldfarb 29:22
I think it's too early to tell. So it's encouraging that there is a minister for AI, but I don't yet know what that means, so there wasn't a separate letter for him. And so we don't know exactly what how this is going to play out, but it's a good first step, and then we'll see what, what the AI policies are, which we we just don't know. We haven't seen them yet.
Marwa Abdou 29:45
What would give you more confidence that we're perhaps moving in the right direction? Is it the passage of legislation like Bill C 27 is it clearer institutional frameworks? What kind of signal. Policy, market or otherwise, would tell you that Canada is getting serious about turning its AI potential into real economic impact.
Avi Goldfarb 30:09
It would be nice to have a law that does say that there is a human responsible for every AI action. So that's a passing bill, but for the most part, it's a good early first stage would be rhetoric around the idea that this could massively improve our productivity, as opposed to being something we have to fear. And I will say the early things that the Prime Minister has said are, you know, AI is a way out of our productivity trap, which is different than what we are hearing from the previous government, especially in the last few years, that AI is something we have to fear. And so I find that encouraging. And I'd like to see you know, that rhetoric brought into, you know, brought into practice to say like this, this is something that, after years of very slow productivity growth, may actually allow us to get out of this, this rut, and improve the economy, not just in healthcare, education and public services, but across the board.
Marwa Abdou 31:05
Avi, boardroom to classroom, policy making to startups, you've helped shape how many of us think about AI, not just as a technology, but as a force that's fundamentally reshaping how we work, how we decide and how we build. I'd love to turn inward for a moment. I want to ask what first drew you into the study of AI and its economic implications? Was there a question, a challenge that perhaps made you feel it was urgent or exciting to you at the time, and looking back years later, after all the momentum, the breakthroughs, the hype, does that original motivation still drive you, or has your perspective evolved as the conversation around AI has grown even more complex?
Avi Goldfarb 31:59
So there were two things that originally motivated me. So the specific story is Jay, Josh and I, we run this organization called the creative destruction lab, and the creative destruction Lab is a program for science based startups. We started it in Toronto in 2012 or Jay started in Toronto in 2012 and we're now in in lots of universities around the world, but in our first few years in the program, which was a general program for science based startups, we had a lot of these AI companies coming out of Toronto computer science, in many ways, because at the time, Toronto computer science was the world leader in AI and deep learning. It is now a world leader, but it's no longer necessarily the world leader. So we saw through our startups what this technology might mean. That was part one that was exciting. Part Two That was exciting is I actually had an intuitive sense of what the technology could do. I spent a lot of time doing statistics, and I taught. I used to teach stats, and this leveraged those tools. Okay, so that was piece two, and piece three. That was exciting is for Canada. This was a place where Canada could lead. We were leading in research, and if we were early enough, maybe we could have Canadian industry lead as well. The first two continue.
It's clearly a transformative technology. I'm excited about the startups I see. I'm excited about the big companies I see. As the technology's gotten a little more the computer science and the technology's gotten a little more complicated, actually, my statistical intuitions a little bit less useful, but I still can get my head around what's going on and and think about the economic consequences. And in many ways, those economic consequences are coming to play a little faster than I would have thought. So that's very exciting. Commercially, Canada hasn't led. We're probably doing better than we are in other deep tech areas for a variety of reasons, but we're not for the most part. Given the lead we had, we are no longer far ahead in AI and a lot of our leading AI researchers who went into industry left the country. A lot of the researchers who still did research stayed, and that's wonderful. We still have a great research environment, and we're training lots of undergrads and grad students to do exciting things, and we continue to Toronto at the forefront, Montreal, Alberta, Waterloo and elsewhere, but Canadian industry is only starting to wake up now. Over the years, Ajay and I in particular, have been, you know, in lots of Canadian companies, meeting them, trying to convince them that AI was something to invest in, and I heard over and over again, especially before chatgpt came out. Well, I believe you that this is an exciting technology, like they believed us, but we don't know what to do with it, and so we're going to wait for someone else to figure it out in our industry, whether it's financial services or insurance. Or your healthcare, or whatever else, we'll wait someone else is going to figure it out, and then we'll be a fast follower. And I'm worried that by the time we get up to speed on being a fast follower, a company that's not a Canadian company will be so far ahead that they'll end up dominating the Canadian
Marwa Abdou 35:18
market as well. Avi, I'd love to give you the floor. Are there any final thoughts that you'd like to leave our listeners with, whether it's a key takeaway, something that we didn't touch on, or just what you think is most important to keep in mind as we navigate this next chapter of AI and economic change?
Avi Goldfarb 35:36
So yes, because I just realized I did not want to end on that pessimistic note. Okay, I'm like, so I'm very excited about the technology, and I'm very excited about Canada with this technology. We do have amazing talent in this country around AI coming out of multiple universities, still have many of the leading scholars for the technology. And while you might say, Who cares about the scholars, the scholars train the students who then build the companies, okay? And that has been what's happened for the US companies, and for some of the, you know, the Canadian companies that have been thriving. And so I think there is an incredible opportunity now that corporate Canada seems to be waking up, that the government seems to be waking up to what this technology can do, and I think it could, with the right investments and the right encouragement and rhetoric, lead to better healthcare for Canadians, better education for Canadians, better government services for Canadians, better financial services for Canadians, better retail for Canadians, et cetera, all across the board.
Outro 36:47
Thanks for listening to Canada's Economy, Explained. Follow us for updates and visit Businessdatalab.ca for free tools and insights, and be sure to join us next time.