Transcript
Transcript: Reflections by Abdi Aidid
[00:00:00 A series of images of people walking along busy urban streets; a Canadian flag flying on the side of a building; an aerial view of Parliament Hill and downtown Ottawa; the interior of a library; a view of Earth from space. Text on screen: Leadership; Policy; Governance; Innovation.]
[00:00:16 Text on screen: Review and Reflection; Produced by the Canada School of Public Service.]
Introduction: Public servants, thought leaders and experts from across Canada are reflecting on the ideas shaping public service: leadership, policy, governance, innovation and beyond. This is the Review and Reflection series, produced by the Canada School of Public Service.
[00:00:26 Abdi Aidid appears full screen.]
Taki Sarantakis: Welcome to the Canada School of Public Service. My name is Taki Sarantakis. I have the great honour of being the president here at this institution, and today we have yet another great treat for you as part of our series of chronicling interesting Canadians that have thoughtful things to say.
We are here with somebody you might not have heard of, but I think you'll hear from him in the future. He is a professor of law at the University of Toronto. He is a former visiting scholar at Yale University. He is involved with one of Canada's great little AI companies, Blue J Legal, which we'll talk about as we go forward. He's also a Canada Research Chair on AI and ethics, and I believe he is also writing a book on AI and ethics. So, Abdi Aidid, sir, welcome.
[00:01:34 Text on screen: Abdi Aidid, Assistant Professor, University of Toronto Faculty of Law; Canada Research Chair, Artificial Intelligence and Access to Justice.]
Abdi Aidid: Thank you. Thanks for having me.
Taki Sarantakis: So, we're going to talk about all of those things over the course of the next little while, and I love talking with people like you because a few reasons why.
Number one, you've done a lot of the things that everybody else is doing or trying to do. You did it years ago, which I always find is super cool. And number two, you are a very thoughtful academic, but you've also done the thing. And sometimes academics can't do the thing – and I say this as somebody who desperately wanted to grow up to be an academic – and sometimes the people that do the thing can't explain the thing in thoughtful ways, but you do both. So, let's start. Tell us a little about yourself. Where were you born?
[00:02:25 Overlaid views of downtown Ottawa, and the waterfront of downtown Toronto.]
Abdi Aidid: Sure, I was born in Ottawa, actually, right here. I grew up here until I was about 7, and then we moved to Toronto where I spent the rest of my years. But I lived first— Baseline?
Taki Sarantakis: Baseline. Oh yes.
Abdi Aidid: Uplands?
Taki Sarantakis: Uplands. Baseline.
Abdi Aidid: And then I went to Charles H. Hulse Elementary School when I lived in an area called Cedarwood. And then I went to Robert Bateman because we moved to South Keys area.
Taki Sarantakis: Wow, so you are a local and a Toronto boy.
Abdi Aidid: Yes, that's right. Ottawa does feel like home, even though every time I'm back here it feels dramatically larger than what I remember.
Taki Sarantakis: Yes, I'm not going to ask you to tell us which city you like more because that would be unfair. But I have a similar thing. I spent almost exactly half my life born and raised in Toronto and the other half of my life here. So, it's interesting. What did your parents do?
Abdi Aidid: So, my parents. My father's a postal worker, remains a postal worker. He's worked at Canada Post for my entire life.
Taki Sarantakis: Is he still here?
Abdi Aidid: He's in Toronto, working, getting close to retirement age. My mother, who has passed, spent some years as a public health nutritionist. And so, City of Ottawa, and the reason we moved to Toronto is that she got a job with the City of Toronto.
Taki Sarantakis: Yes.
Abdi Aidid: Now both of them, I should say, are refugees from Somalia. And so, they came, requalified for their jobs, and I was very lucky to have them because they were both really hard-working, thoughtful people and very bookish, which is the reason, I guess, for me becoming who I am now.
Taki Sarantakis: Speaking of bookish, you went into law. Tell us a little bit about that.
Abdi Aidid: So, I went into law for the same reason I would say most people do, which is this risk aversion. I thought it would be a path to some stability. Actually, my mom wanted me to do it, which I think is probably the best reason to do something.
But the thing that I learned really quickly, because I didn't know many lawyers, if any, was that it was this remarkably dexterous education where you could affect policy. There's a whole bunch of things that implicate the law, whether or not they're actual legal practice. And so, I felt like even if I wasn't going to be a conventional lawyer, there's other things that I could do with the law education.
And so, I was excited about it, and then I became a conventional lawyer. So, after graduating, I spent some years as a litigator, first in New York, and then in Toronto.
Taki Sarantakis: Now, at some point you started to get interested in this thing called AI. And not only did you start to get interested in this thing called AI way before ChatGPT and generative AI, you started to get interested in AI and the law. Start us a little bit along that path of discussion.
[00:05:17 Overlaid image of the book "The Legal Singularity: How Artificial Intelligence Can Make Law Radically Better", by Abdi Aidid and Benjamin Alarie.]
Abdi Aidid: Yes, so sometimes I disappoint people when I tell them about how I come by my interest in AI. So, I think there are some people who like AI because they like gadgets. That was never really me. There's people who, they'll talk to you about AI, in the next breath they'll talk to you about crypto and things like that. That's never been me. I'm actually just somebody who was convinced that it could be useful in resolving some issues that I felt like I was confronting. For example, I felt this profound sense of guilt that, as a lawyer, I had these skills that I could use in service of people, but they could probably never hope to afford – at least the ones who needed the most – were never really able to hope to afford my services. In fact, here's an interesting fact for you: in my years as a lawyer, I've never been able to afford myself. So, if I wanted to retain myself, as a lawyer, I'd be out of my own price range.
Taki Sarantakis: So, does that mean you're a really good lawyer, or does that mean you don't get paid enough?
Abdi Aidid: It depends who you ask. But I would say, as a matter of fact, it just means that it's out of reach. Legal services are out of reach even for people who have stable incomes.
And so, I thought that AI was one of the things that we could deploy to resolve this concern. And at the time, there wasn't a lot of work being done on it.
[00:06:37 Overlaid images of articles:
"Industry Watch, Law and Word Order: NLP in Legal Tech", by Robert Dale, 2018, Cambridge University Press;
'Natural Language Processing and Machine Learning for Law and Policy Texts”, John Nay, 2018, NYU;
"Legal Natural Language Processing from 2015-2022: A Comprehensive Systematic Mapping of Advances and Applications", Quevedo Caballero et al, IEEE Access.]
Abdi Aidid: There were some maybe academic research projects that would suggest that you could use things like natural language processing to read cases, but no one was being really inventive. And so, even though I was working as a lawyer and enjoying that actually full-time, I had my eye on maybe working on some more creative projects.
Taki Sarantakis: And what year, roughly?
Abdi Aidid: It's about 2016.
Taki Sarantakis: 2016. So, we're here in 2026, so this is about a decade ago, long, long, long before AI became mainstream. Now talk to us a little bit, were you a professor then, too?
Abdi Aidid: No, I started teaching first as an adjunct professor, which if you're not familiar with the academic nomenclature, it means that you're a practitioner who's teaching part-time. And so, I was teaching legal research and writing.
Which, by the way, was probably the most instructive for me, in terms of showing the need for AI, because we were teaching students to conduct research by entering search queries into databases.
[00:07:41 Overlaid image of a screen shot of research being done on LexisNexis.
Text on screen: LexisNexis is a leading legal research database used to teach case law search strategies.]
Taki Sarantakis: LexisNexis.
Abdi Aidid: Yes, we were training them to use all kinds of different databases. And the kind of students that would be rewarded as good researchers were the ones who would come up with the best search terms. And you were just like; there has to be a better way to do this.
And it wasn't that long ago that we were teaching people to do research by navigating the stacks of brick-and-mortar libraries.
Taki Sarantakis: Exactly.
Abdi Aidid: And so, I thought that that was a place that we really needed some change because information wants to be free, as they say. And it felt like you had to effectively solve a riddle to find a case that you know exists, so I thought that was one place where we could deploy some AI.
And so, when I was teaching students, I realized they didn't have the same compunction about using new technologies that we did. The legal profession is pretty conservative, institutionally. You could imagine a courtroom 100 years ago; it probably looks a lot like it does today. And the students who, I think, just wanted to cut to the chase were ambitious in a way that inspired me, and they could also immediately diagnose some of the flaws in the way that we were doing things. And so, I thought I want to spend more of my time figuring it out.
Taki Sarantakis: So, it's really hard to explain to people – Google's roughly 30 years old, the commercial internet is roughly 30 years old. It's really hard to explain to people who didn't live it – the high transaction costs associated with finding information. You had to physically get your butt to a library, you had to go stacks, you had to look through card catalogues. And so, I think that started informing you to the, "Wow we've got all these tools, but we're not pointing them in the legal area as much as we could be."
Now talk to us a little bit about when you started going from, "I've got this intellectual itch" but when did you actually start doing something with AI? How did that come about?
[00:09:51 Overlaid image of an academic project, "How Artificial Intelligence Will Affect the Practice of Law", by Anthony Niblett, Ben Alarie, and Albert Yoon.]
Abdi Aidid: Yes, so actually there was an academic project that was born out of the University of Toronto. My colleagues, Anthony Niblett, Ben Alarie, Albert Yoon, realized really quickly that the same data science techniques that people were using in other fields – insurance, where they were using data science tools to do things like set premiums or do claims adjustment – those same tools could work effectively in the context of the law, and we could actually predict legal outcomes.
Which, by the way, was, I could argue, the long-held aspiration of the law. Basically everything that people do in law is prediction. If somebody comes to you and they say, "I have this legal problem," they don't want you to recite the multi-part legal test. What they really want is for you to tell them what's going to happen. What's a court going to do? How is a court going to react? Am I going to attract scrutiny from a regulator?
And so, they were early in trying to leverage these techniques against the body of case law to try to predict legal outcomes. And they brought me in to help build out the company. And so, I joined as the director of legal research, and I was a relatively early employee.
[00:11:00 Overlaid image of Blue J Legal's homepage.]
Abdi Aidid: And by the time that I transitioned to full-time academia, Blue J was this massive thriving company, which I take no credit for, but it was an exciting time to problem solve and to ride the wave up in the world of legal AI.
Taki Sarantakis: Just like it's hard for people who grew up with Google and the internet, we're now at the point where there's some people that the totality of their AI experience has been generative AI. But surely that wasn't the case when you guys were starting. Talk to us a little bit about what was AI before AI became ubiquitous?
Abdi Aidid: Yes, so I think this is a really good question. I would encourage people to think about AI conceptually. I will say, people like me, academics like me, we've done the public a great disservice because we've overinvested in technical explanations of what AI is. But I think people can understand it conceptually.
So, let's talk about wave 1 of AI and law, which was really machine learning and algorithms. What does that mean? It's about taking a large volume of data that computing power is required to fully synthesize, and it's about identifying insights from that data. So, imagine a credit card company looking at your purchasing history, your income, your habits, and trying to determine what the appropriate credit limit for you is.
Taki Sarantakis: And that's something we've been doing for a long time.
Abdi Aidid: For a long time. Making a prediction from existing data. The thing about machine learning is, if you think about like 12th grade math, it's like finding a line of best fit in a scatter plot, is the way I like to think about it.
Now, the thing about machine learning is it's limited to what exists in the original dataset. So, I like to think about it like if you give a sculptor a block of ice, they can make a beautiful swan, but it'll never be larger than the block of ice. That's machine learning. It's about pulling insights from existing data.
Taki Sarantakis: It's contained, in a way.
Abdi Aidid: It's contained, and it's also vulnerable to all the limitations that exist in the original data set. So, when we talk about things like bias and discrimination, if there's, let's call it, unhygienic stuff lurking in the original data, there's a risk of projecting that forward in the ultimate prediction.
So, generative AI is a different thing, because what it does is it takes all of that ability that we have in machine learning and says, "Let's learn from it, let's take this data and train models on it, let's learn everything there is to know about syntax and diction and where a nose is on a face, et cetera, and let's create net new information." So, it could actually make the block of ice bigger, comparatively. So, generative AI, generating, machine learning about identifying and surfacing insights from existing data.
We used to only do really machine learning. So, when you think about what you might call algorithmic decision making, that's exactly that. That's, "Are you eligible for this public benefit or not?" Well, let's look at profiles of people who've been eligible in the past. Let's reconcile it against your particular profile and let's see where you fit.
Taki Sarantakis: Right. And the rules.
Abdi Aidid: And the rules, and the law. And you inform it with a bunch of different parameters. Generative AI says, "What's all the data that we have historically on applications for this benefit? Let's learn what a good application looks like and let's maybe also draft applications. Or let's draft decision orders to supplement our decision-making."
[00:14:25 Overlaid image of a computer screen showing the introduction page of ChatGPT. Text on screen: New Chat, ChatGPT, Examples.]
Abdi Aidid: And so, the possibilities, when ChatGPT came out in late 2022, became endless because not only could we do the work of surfacing the insights, we can generate work products. We could actually say, "Predict a case outcome, but also draft reasons."
And so, in the world of law, this was a seismic moment, more so actually than the shift from text, brick-and-mortar, physical bound volumes to digital databases was. All that was, was us platforming information we already had and making it retrievable. Now we're taking that information and treating it like data and performing functions on top of it, and we can do all kinds of things.
So, that's exciting. And fundamentally, right now we're gearing up for another shift, which is to agentic AI, where we take everything we've learned from machine learning, all our generative capabilities that we have with generative AI, and we're investing that in these AI-enabled personas which can engage in these multi-step transactions without supervision at any one constituent step. And so, that's amazing.
Taki Sarantakis: Yes. So, if I'm understanding you, I think what you've said is most of us understand AI as generative AI, which is the second generation after machine learning AI. And now we're about to enter the third generation with some seriousness, some scale, some scope. And the third generation is different from the second generation because?
Abdi Aidid: Well, think about machine learning as trying to replicate our cognitive architecture in a way that we can use computing power to do things that we can't, like synthesize large volumes of data. So, that's machines thinking like us.
Taki Sarantakis: Like us.
Abdi Aidid: Like us. Generative AI is machines looking and sounding like us, and agentic AI is machines behaving like us. That is a remarkable trajectory.
Now, I don't want to describe them as distinct phases because we're doing it all right now. Credit card companies are using machine learning to determine what your credit limit should be, and they might be using generative AI to issue you a letter saying they're extending your limit, but they also might use an agentic tool to be the call centre responder, for example. And so, it all flows into itself, but we've learned a lot along the way.
And so, the question for us is, how do we coexist with these tools? What are the right rules that we should use to constrain them or enable them?
Taki Sarantakis: Now you're starting to get us into a really interesting area. This is an area that you're an expert in, so I'd like to hear from you because now what you're starting to talk about is technology and society, and technology and legislation, and technology and regulation.
And so, now it seems, to a lot of people, that the world is shifting dramatically, that all of the solid footing that has been known in law, in medicine, in accounting, things that took – how many years did you spend at law school?
Abdi Aidid: 3, but I also did graduate law school.
Taki Sarantakis: Yes, graduate law school. Things that took years of very intensive things –
Abdi Aidid: Yes.
Taki Sarantakis: Little kids appear to be able to get those results now. Start talking to us a little bit about what that means. It's not just our technological capacity. It seems to me it's also, how are we going to deal with this technological capacity?
Abdi Aidid: Totally. Here's, I think, an important thing we all have to accept, which is you can't un-ring the bell on AI. The technology is here. And companies, technologists are ambitious and they're going to want to do more. And we have a limited capacity to restrain it. And so, the question for us is, how do we absorb it in ways that are safe and responsible? And law plays a big role in that because it helps us determine what are things that are non-starters for us. What kinds of obligations should we impose on the public? What kinds of obligations should we impose on developers of AI technology?
And so, for me, this is the core question of our moment, is how do we absorb these tools into our day-to-day, given that they're here to stay and given that they have this pretty remarkable upside. How do we absorb that while also limiting the downside? Because the downsides are indeed real.
And so, for me, I think about regulation a lot. Regulation is a complicated thing because it automatically comes across to people as you saying, "I want less of this." As a matter of fact, regulation can be a set of enabling tools, too. They can be tools that are calibrated to give you maximum predictability in a marketplace. Banking regulation has not stopped us from having bank accounts. It has just made it such that when we have bank accounts, we all understand what we can expect.
And so, regulation is a big ask right now, too, because there's so much uncertainty in the world of AI. Think about it. If we made a set of AI rules today, we'd be making bets on what kinds of technologies are likely to endure. And we don't necessarily know today.
Taki Sarantakis: Exactly. It would be like, to me, when I speak about this, I always say that the analogy is the automobile was invented, created, and you can't regulate/legislate every possibility of the automobile in 1915 because it's just beginning. Society interacts with the technology, makes changes, goes forward.
Now, you said something that I think a lot of people have trouble understanding, and it strikes me as odd that they have trouble understanding, which is the following: the legal use of AI in any society will ultimately be decided by legislatures and courts. Some combination of those two things will decide, define how we use AI lawfully in a society. Yet you hear all these people having all this angst about, "Is AI ethical? Is AI good for society? Is AI bad for society?" Are those questions too big?
Abdi Aidid: I think those questions presuppose a couple of things that aren't true. One is that we can meaningfully stop AI, which I think is the horse is out of the barn on that one. And so, I'm not that interested in a debate about is AI good or not. I recognize it is what it is, and I'm trying to create structures for us to use good AI well.
And the other thing it presupposes, these conversations, is that people are going to continue to care about AI as such in the future. I think about this a lot. I don't actually know what technical process underlies most of the things that I do in my life.
I've recently learned that when you depress the brake pedal, it doesn't actually clamp the wheel. There's some software that intervenes along the way. But I'm fundamentally incurious about that because I trust it, that it's going to work most of the time, and if it doesn't work, I know who to call. I know to call the mechanic. I don't know how my laptop works necessarily, but I know that I can go to the Genius Bar and get some help when it doesn't work.
And so, I think that there's a possibility that in the near term, maybe the medium term, that AI becomes invisible as a thing to us. And right now I suspect that part of why we're so invested in looking under the hood is because we don't trust it. We think it's a newfangled thing that scares us. Well, if we build a sufficient ecosystem of constraint around it, a genius bar, a mechanic that you can go to, some rules that are predictable, then I think you'll see that curiosity about the technology wane a little bit.
Then we can have the real important conversation about, is this a good deployment? Is this a fair outcome? Is this the right application? Right now, the question of AI or not, I think, is a little too basic for the stakes.
Taki Sarantakis: Yes, if you think about, you go back to previous big general-purpose technologies, you think about debates around the introduction of railways, or electricity, or things like that. You could have the philosophical debate, but I think you really nailed it, which is the moment you stop thinking about these things as philosophical debates and you're, "I no longer worry about the quality of the water that's coming out of my tap, or I no longer worry about the voltage of what happens when I activate my electricity" then the magic starts.
Abdi Aidid: I think so. But part of the reason that I think I can get frustrated with the normative debate about is AI good or not is because it's actually the wrong philosophical debate to be having right now, because there are rich ones that we're not having, which we could have.
For example, think about it. We spent generations investing so much of our self-worth in our ability to perform tasks. Well, what does your sense of purpose look like in a world where you have shared competency with a piece of technology? That's a question about who you are in the world, what your value is, that we should be contending with and thinking about, and that I think is a richer conversation to have than the one about is AI good or bad.
Taki Sarantakis: Yes. I have a little theory on that. I'd love your thoughts. I think this technology is the first one that is directly – I don't want to say challenging but maybe challenging – challenging white-collar workers. Doctors, lawyers, accountants, journalists. My parents were blue-collar workers. And when we were having this debate in the '70s and '80s about globalization and the like, a lot of the white-collar workers were like, "Oh, this is progress. It's good for everybody. You'll get new jobs, etc." I think AI is now challenging a type of social strata that is maybe feeling a little threatened for the first time.
Abdi Aidid: I think that's exactly right, but before I get into that, I have to also say even within that social strata, the effects are not evenly distributed.
So, a lot of the AI disruption that you're seeing right now is at the junior level of that white-collar social strata, which for me, as a Black man, that is of particular concern. Why? Because all of the work that we've been doing around economic mobility in the communities: resume workshops; interview prep; scholarships for kids was just to get people to the entry point, and now that's the first that's going to be automated out? And so, you're like, "Wow, is that a ceiling on the economic mobility of vulnerable communities after a generation of hard work?"
And if you look at the other areas where AI is disrupting the labour force, it's in what we used to pejoratively call unskilled labour, the gig economy, so to speak, as well. And that's a place where, say, people of colour are overrepresented as well. So, I'm looking at it and I'm like, "Wow, even within the disruption, you're reproducing the same background inequalities that exist in the analogue economy, let's call it."
So, that's one thing I'll say, which is to say that there are members of the social strata of white-collar professionals who have decision-making authority, let's think director level or better, that they can insulate themselves pretty effectively for a time being.
Taki Sarantakis: For a while.
Abdi Aidid: For a while. And also because a higher proportion of those professions have regulatory bodies, they can also do the monopolistic, cartel-istic thing and resist it.
Look at law, for example. Law has fought vociferously against –
Taki Sarantakis: It's self-regulating.
Abdi Aidid: It's not just self-regulating, it's self-regulating with a monopoly on the provision of legal services.
Taki Sarantakis: Right. I, as an uneducated, as a non-bar person, cannot call myself a lawyer.
Abdi Aidid: You can't call yourself a lawyer. So, imagine this, you have a monopoly on the provision of legal services, and you self-regulate, which to me is like a perfect petri dish for all kinds of anti-competitive germs to grow. And so, I think there's going to be that dying gasp for a while from some of the professions that's going to buy them some time. And the question is, in that time period, do they evolve or die?
And I think there's a chance to evolve. Why? Because it turns out the public actually wants more from its white-collar professionals. They want better services, they want more services, they want it to be more affordable, more economical, and they want things like discoveries. They want us to make meaningful progress.
And so, again, this might be a function of inequality, but the upper reaches of the white-collar professions are also the ones that are poised to gain the most because they can supercharge what already is baseline expertise.
Taki Sarantakis: Exactly. Now, you talked about a very important issue that I want to stay with for a few moments which is, maybe to combine two parts of the conversation so far, that the debate about AI good, AI bad isn't that interesting, but access to AI, AI and equity, AI and affordability, that seems to me that's a debate we should be having.
Abdi Aidid: Oh, it's a massive debate we should be having. The debate around AI's propensity to generate harm is a good one.
Here's how I think about it fundamentally. We are more capable than we've ever been to do good in the world, and we are also more capable than we've ever been to do harm in the world at scale. And so, that's a solemn responsibility, and I think it depends on us having the right debate.
So, an example of a debate I'd like to see is, are there things that we never want to use AI for? Specific things. So, I give the example often of criminal sentencing. So, I'm against using AI in criminal sentencing. But I'm not against it because I think the technology can never do it. I hold open the possibility that there will be a world where we can develop an optimally deterrent algorithm that can be fully de-biased, that can blunt any racial or systemic effects, and could sentence somebody –
Taki Sarantakis: To 4.798362 days in jail.
Abdi Aidid: I still don't want it. If we perfect it, I still don't want it. Why? Because I just don't think criminal sentencing should be optimized. I think my view of it, as a social responsibility, is that it should be the thing that keeps us up at night. Judges should struggle to decide to put someone in prison. We should all feel, as a community, the effects of that.
The question for me, and that I want to ask everybody, is how many things in our social and public life do we feel that way about, where no matter how well the technology can do, we still don't want it because we have some independent reason, a value that we think supersedes efficiency, optimization, speed, affordability. I'm interested in identifying those and then jealously guarding them.
Taki Sarantakis: Yes, like for example, one of the things that quickly comes to mind is maybe voting. Like, we might say, it's a really rational thing to sit down and say A, B, C, D, E, F, G, therefore you vote for candidate B. But the citizen hasn't grappled with that. The citizen hasn't engaged.
Abdi Aidid: Yes, that undermines something that we think is fundamental to representative democracy, which is deliberation. The idea of optimizing voting and presupposing people's choices. So, again, the thing I want to drill down on is, "What are the kinds of things that we think are virtuous because they're analogue and depend on being analogue for their virtue?" I want to identify those and invest in protecting those.
The problem is, some people have a notion that they don't like sentencing algorithms or notion they don't like algorithmic voting, and they'll use that to resist low-stakes applications of AI in the workplace, like summarizing documents. Then you're not a credible objector anymore at that point. And the thing that I want us to do is start disambiguating types of AI, uses of AI. So, we're not talking about this indeterminate morass, and we're talking about particular utility.
Taki Sarantakis: Yes, the specificity. And it's like, AI is helpful and a societal good here. AI is either not helpful or not a good societal good there.
Now, I think what scares a lot of people is in terms of how we legislate, regulate, litigate, which is how we decide virtually all of our contentious issues. I guess another fourth way would be war. But given that we have these mechanisms for dealing with contentious societal issues, at some point we will land on some of these. It'll take a long time. Maybe it'll take 30 years. I mean, we're still litigating the internet and we're in year 30-something of the commercial internet.
Abdi Aidid: I'm glad you said that because I think the internet is an interesting analogue for what we're talking about here. Why? So, a lot of people will point to the fact of some AI companies failing, or they'll point to the fact that there's actually been regression in some AI models, as an example of how this might all be a bubble. And I say this; the internet had a bust. There was a boom and a bust.
Taki Sarantakis: Pets.com.
[00:33:12 Overlaid image of the Pets.com sock puppet mascot.
Text on screen: Pets.com was launched in 1998. The online pet supply company went public and shut down in just 268 days. Its sock puppet mascot became iconic through major ad campaigns.]
Abdi Aidid: Pets.com. You had examples of contradictions in the marketplace, and it took time to learn what the public wanted. But even though there was a bust, would you deny that the internet suffuses every aspect of modern life?
And so, the thing I want people to do is take a longer view of what's going on. If we're trending towards something like a singularity, it doesn't mean that it's going to be a linear trajectory. There might be a messy interregnum, choppy waters in the meantime, where we work out the contradictions or we decide what we like and don't like. That takes time.
And so, don't count on progress every day. But have a view, a slightly longer view than a single economic quarter, I think. And that's an important thing to keep in mind.
Taki Sarantakis: Now, one of the things that you've taught me that I think about all the time is you've said to me, and to others that I've heard you speak to, that when you're looking at an AI application today, it's like looking at a star. It's something that it may appear real to you, but it's like this dead thing in the universe.
Abdi Aidid: That's right.
Taki Sarantakis: The real stuff in AI is the stuff that's in the labs that's just about to go into production tomorrow.
If this world is moving so quickly – every day there's a new amazing Claude this, a new amazing ChatGPT that, a new amazing Gemini, and they're all leapfrogging each other in this infinite race of additional functionality – how can legislation, regulation, legislatures, politicians, the legal community, the judicial community, do we have a hope in hell of managing this process, this interregnum?
Abdi Aidid: Yes. It's a good question. So, there's a couple of different models. Think about the way that we deal with technological innovation in the context of prescription drugs.
We say, "Hey, for the privilege of being able to sell prescription drugs, you have to tell us what you're doing before you get it to the marketplace." It's more of a product safety model.
Taki Sarantakis: But also we have to approve it, too.
Abdi Aidid: We have to approve it too. So, we put all these ex-ante obligations on the developers, and we make them do active disclosure. That's one model. You might argue that that model, as technologists often do, would stifle innovation in the context of artificial intelligence applications.
So, then the question is, "Well, what can we do to get some information so we can anticipate, as a public?" And so, things like some disclosure requirements; things like maybe recognizing that there's certain applications that are particularly high risk where we might require an approval process like they're contemplating in Europe. So, there's ways, I think, of surfacing some of the subterranean AI development. We have to be clever about it.
And here's, I think, the more important insight, which is we have to think smartly about rule design. Okay, what do I mean by this? So my students will, if they're watching this, will at some point yawn as I describe this, but I often talk about there being two kinds of laws. We have rules and we have standards.
Rules are like, you can't drive more than 50 kilometres an hour on a public road. The idea being that it's a binary determination. An officer stops you and you can't say, "Officer, I was driving safely, and I wasn't near a school zone and there were no kids around and by the way it's 3 AM." If you're driving 51, you broke the law. It's up to the officer now to determine what to do. Compare that to a rule that would say, "Proceed reasonably safely on the public road." That's a standard. It's more open-ended. Contextual factors matter more.
Now, when would you want a rule? You'd want a rule in a circumstance where you have perfect knowledge about what constitutes safety. So, when we're driving on the public road, you know that 55 is too high for pedestrian safety. You know that the flow of traffic needs to be unimpeded. And so, you can sort of set the limit. But think about when we use standards. When we say things like merging on a highway, we say merge when safe to do so. Why? Because we can't anticipate every possible circumstance.
So, if we made an AI law tomorrow, a set of rules for AI, would we choose rules or would we choose standards? We might choose standards because we say we can't anticipate every scenario, contextual factors matter. If we decided to go with a specific set of rules, then we'd be risking being under-inclusive. What do I mean by that? We would say, "Don't build this machine learning algorithm." Well, what if the ecosystem outpaces that specific command?
And so, the question for us is, can we draft our legislation, our regulation, our commands in such a way that anticipates uncertainty, and it's flexible enough to adapt over time? And for me, we sometimes get it wrong because we want to overprescribe, but what happens is it gets outpaced. Those rules don't have a long shelf life. Other times we want to be too amorphous, too open-ended, but those don't give enough commands to the marketplace.
And so, for me, it's about being intelligent about rule design as well and not just deciding whether or not to regulate.
Taki Sarantakis: Yes, so it's basically, if I'm hearing you right, it's a what instrument to deploy in what circumstance.
Abdi Aidid: I think so. And being super intentional about the way rules can incentivize good or bad behaviour, desirable or not desirable behaviour.
Taki Sarantakis: And also, I think another way of thinking about this is we're going to be regulating, legislating, and litigating in this area a lot as things interact with reality, with new models, with new capabilities, with new on and on and on. And I think a lot of people are really in the world of, "Oh, we can't do anything until we get it absolutely perfect." That we don't want to touch this right now because it's still growing.
But touching it right now doesn't necessarily mean this is what it's going to be forever. It's give and take, in terms of, coming back to the car, we never said, "All automobiles for the rest of time will be made out of wood, and will have steering that wood on the steering wheel."
Abdi Aidid: No.
Taki Sarantakis: We adjust as we go forward.
[00:40:02 Overlaid images of early models of cars.]
Abdi Aidid: And also, we take a long view. It took some time, after the car was invented, for it to occur to everybody that we should mandate seatbelts, for example. And so, you can also imagine a world where we decide on what the general safety applications can be.
And the car one's a good one for me because it really shows you something about the human psyche in these contexts. I was reading recently about how there was a movement when cars were first designed and developed to actually try to make them appear more esthetically like horses.
Taki Sarantakis: Because that's what we knew.
Abdi Aidid: The idea is that's what we knew. And so, people are okay with the idea of a conveyance, they just don't want it to be different than the one that they're familiar with. And there's another interesting analogue that I want you to think about, which is elevators.
So, when we first started doing research into elevators, developing them, the presumption was that they were going to be primarily for private residences. Today, probably less than a percent of elevators are for private residences. Most people have stairs in their home. Maybe there's elevators in high-rises and in primarily commercial settings. And so, one thing is that we thought we knew what the tech was for. It turns out that we let the world tell us where they wanted it.
But the other thing that was interesting was we had elevator operators for many more years than we had to, probably like a couple generations longer than we had to.
Taki Sarantakis: This is like movies of New York.
Abdi Aidid: It's like movies of New York, exactly. We had them for a long time. Why? Because for a while people insisted on a human touch. Same reason we might have parking attendants today. Even though the technology has been capable of taking your ticket and dispensing your change for 50, 60 years.
I think the point here is that don't also presume displacement automatically, all the time, immediately. Recognize that we, as a public, are supplying the technological marketplace with information about how we want to use these things, and that information is being considered.
Taki Sarantakis: Now let's talk a little bit, maybe I don't want to call it nefarious, but maybe the non-legal uses of AI, because it seems to me that if you look at some of the things that had high potential for destructive capacity in the past, things like high-scale armaments, tanks, the nuclear bomb, these were things that largely were limited to either state actors or ridiculously well-resourced private sector actors.
That's not the case with AI today. AI today, relative to its capability, is ubiquitous, is available. I think that's another thing that scares people, too. That you're giving a tremendously powerful tool to you don't know who.
Abdi Aidid: Yes, I'm of two minds on this one. So, one is that I'm not fully bought into the idea that the average wrongdoer has the ability to access tools that can do things like commit, make, facilitate destruction and harm, really simply. For the most part, people have access to off-the-shelf tools that are still resisted by cybersecurity systems and remain detectable to law enforcement, for example. But the reason I say I'm of two minds is because I don't think it's long before you start to see an economical, maybe free tool that can do things like commit significant fraud at scale.
So, I say that in that gap, in the period between when we think it has already happened and when it actually will happen, I say we should rise to the challenge and think about ways that we can have more robust safety. I think for a long time we've been asleep at the wheel about the possibilities of things like elder abuse – this is one I think about a lot, where people that get scammed out of money via email phishing or via this relationship cultivation that happens digitally – we should turn a more serious eye to that, and maybe AI is going to give us the occasion to.
And so, I wonder also about it creating space for us to address issues that were longstanding because they become more acute now.
Taki Sarantakis: It's interesting because up until very, very recently, if you were a company and you were a victim of a cybersecurity breach, you wouldn't tell anybody. It would be like this shameful thing, and it just kept perpetuating attacks. And so, naming the thing is a first step towards dealing with the thing.
Abdi Aidid: This is also true of things like bias and discrimination. So, one of the things that AI does is that it shines a light on existing social challenges. I give the example often of, imagine that we decided to set all of our wages algorithmically, and what would happen is that everyone would just upload their LinkedIn page to some wage system, and it would tell you how much you get paid at your next job. Like I said about machine learning, it would be trained on historical data about wages, about your resume, your background, etc.
Taki Sarantakis: It would lock in today.
Abdi Aidid: Yes, right. So, one of the things it would do is it would prevent you maybe from making more, so it would calcify the possibilities. But the other thing it would probably do is, absent any other constraints, it would project the gender wage gap forward. Why? Because it'd be trained on historical wage data, and we know that there's disparities in historical wage data. So, there's two ways to look at this. One way is to say, let's unplug the tech, which I might support in that context. Fine, unplug the tech. In fact, I would say socially we'd be tripping over ourselves to be the first one to unplug it if a machine said, the man gets 100 cents, and the woman gets 79 cents.
But think about it. The gender wage gap persists. It exists socially. It's diffuse in organizations and relationships, and we haven't overcome it. And so, we could look at it as a technological problem, or we could say we should pay women more. Don't let the tech take you off the hook from doing the right thing socially.
Taki Sarantakis: It's not creating the imbalance or the injustice. It's reflecting the existing imbalance or injustice.
Abdi Aidid: It's holding up the predictive mirror, but it's also doing another thing, which is creating a new occasion to address the problem. Why? Because people have a hard time accepting that they might have these biases and these challenges themselves. When it becomes disembodied and we say the tech is doing it, then people are suddenly like, "Well, we can't have structural racism anymore."
Well, this example with judges and criminal sentencing. Judges are very skeptical of sentencing algorithms, but the algorithms are trained on historical data on sentencing. So, when they have racially disparate outcomes, that's because the judges were handing down racially inequitable sentences. If they weren't going to hear us about things like unconscious bias or structural racism before, maybe they'll hear it now.
Taki Sarantakis: Exactly. I always say it's easier to un-bias an algorithm than it is to un-bias a human, because a lot of the things that you see are from datasets are the cumulative decisions of humans.
Now, let's close off this way. In addition to being a law professor, in addition to being a scholar, in addition to being an AI entrepreneur, an expert in AI ethics, and on and on and on and on, probably your most important job is you're also a parent. And let's talk a little bit about AI and the future of our children. And what kind of – I don't want to say world that we see AI ushering in – but what are some of the things that you think about vis-à-vis AI and being a parent?
Abdi Aidid: Okay, I'll start with the opportunities and the things that excite me about it.
So, one thing that excites me is our children being in a world with frictionless access to information.
[00:48:36 Overlaid image of a diverse group of children seated at a table, making crafts with an early childhood educator.]
Abdi Aidid: And that can be everything from information that they need for their education, for their careers, but also if information is swirling around and the costs of accessing it are lower, then I imagine that we're going to have greater occasion for things like scientific breakthroughs; we're going to have greater opportunity for things like reducing the incidence of disease; we're going to have, I think, significant social, health, economic upside associated with the free flow of information.
I think about the possibilities for them and their creativity if their labour is no longer tied to their capacity for information retrieval, but where they're rewarded for being able to synthesize and analyze and adjudicate and add value to the information as opposed to just reflect and amplify it. So, I'm excited about what they can be and what that world looks like. I'm excited about the possibilities of them living in a world where they can know more about one another, which I think AI enables. And so, there's a lot that I'm excited about.
I'm also excited about ways we can de-bias the world through technology. So, my children are my children, so they happen to be what you would call today, "minorities", which I don't necessarily believe will be the case in the future. But a specific thing that we have, for example, is some racial inequities in healthcare. The example of it being hard to detect melanoma and other skin cancers in people of colour.
[00:50:09 Overlaid image of a doctor examining a medical X-ray image.]
Abdi Aidid: Well, one of the things that AI is doing, for example, is because it can synthesize larger volumes of data, it's telling doctors to look beyond their limited experience, and it's actually able to point them to other indicators of disease. And so, I like the idea of them living in a world where those things become less of a risk to them. Somebody's blind spots become less of a risk for their flourishing. That's exciting to me.
There's a couple of big concerns that I have that I think we need to solve. One is the environmental concern. So, as it stands today, AI is quite resource intensive.
[00:50:36 Overlaid image of an article from MIT News, titled "Explained: Generative AI's environmental impact." An area of the article is highlighted, and reads:
"…a generative AI training cluster might consume seven or eight times more energy than a typical computing workload."]
Taki Sarantakis: I think already about 10% of the US grid, 8 to 12% is AI.
Abdi Aidid: It's a massive energy consumer in ways that I think, and I continue to believe, are the result of failures to invest in green and renewable energy prior. What's happening right now is exposing fault lines in the way that we conserve energy today. And so, I want to see meaningful environmental policy writ large where AI can slot into.
And then the other thing that I have some concern about is, are we going to find ways to – let me change this – I'm not concerned that we're not going to find ways. I'm concerned that we're not going to all find ways to have meaning and love and care in our lives in person.
So, what do I mean by this? There's a bit of a concern that children are interacting with artificial intelligence tools and depending on those tools as a replacement for the old school in-person relationship. And so, what I would love to see is us harnessing AI for all its flourishing capabilities, but also maybe using some of the time that we can maybe get back for one another to actually spend time among each other and cultivate personal relationships in that way.
Taki Sarantakis: Instead of falling in love with bots that know everything about us and tell us what we want to know because they know everything.
Abdi Aidid: And regulation has a role to play here. Why? Because the stickiness of these tools is entirely a design choice. Taking an LLM and anthropomorphizing it and making it sound like an earnest person you can trust is a design choice.
Taki Sarantakis: Yes, it's like making the car look like a horse.
Abdi Aidid: It's like making the car look like a horse. Maybe we can do other things that can make it less of an attractive nuisance for children. And so, I have a lot of hope for the future. I'm actually a relentless optimist when it comes to what we can achieve with AI. We have to focus on those two things.
There's like a runner-up concern that I have, an honourable mention, which is that I think we need to improve our media literacy, massively. I understand the allure and the concern around deepfakes. But people are also falling for stuff that's not convincing. And I think that's a civic and media literacy concern. Of course, we're all going to be tricked by technology that's better than the best CGI movie effects. That could happen to any one of us. That could happen to an AI expert. It probably will continue to improve.
But we also, I think, are falling victim to misinformation where there's an educational intervention that can happen. And my concern here is that we might limit the AI possibilities because we're concerned about things like misinformation, when those are problems that are longstanding and those are eminently resolvable through becoming better readers, for example.
Taki Sarantakis: Professor Abdi Aidid, thank you for giving us a little glimpse of your sparkling personality and more importantly, your sparkling intellect. I am heartened knowing that people like you are going to have a big say in how people like me think about things like this going forward.
I'm just heartened by the fact that you write about this, you think about this, you teach this, and I think we're going to need more people like you in other disciplines, like AI and medicine, AI and architecture, AI in construction, AI and roadways, and on and on and on. Because I think you're exactly right. I don't think the question is AI good, AI bad. I think it's, "How are we going to grapple with AI in this specific domain, in this specific context, so that we get the best societal outcomes?"
Abdi Aidid: And, and how do we subordinate it to our interests? How do we make it do for us rather than us have to kowtow to it? That's the thing that I'm interested in. I'm interested in the appropriate division of labour and the right relationship to technology, and I think we need to dominate it and use it for our benefit.
Taki Sarantakis: Thank you so much for spending this time with us.
Abdi Aidid: Thank you.
[00:55:25 The CSPS animated logo appears onscreen. Text on screen: canada.ca/school.]
[00:55:30 The Government of Canada wordmark appears.]
