🤝 Meet Ajay Agrawal. He is an economist and professor at the University of Toronto's Rotman School of Management, where he founded the science-focused startup incubator Creative Destruction Lab and serves as the Geoffrey Taber Chair in Entrepreneurship and Innovation. An expert in the economics of science, Ajay answered all sorts of questions about AI and labour.
You were relatively early to the AI wave. What first made it an area of interest for you to study?
I've been studying economics of science and technology for about 25 years. My PhD dissertation was on the economics of commercializing science. We do very well in science, but we do poorly in commercializing it, and so given that that was the topic of my research, I ended up getting involved in that.
I decided that I would try to, rather than have a big national impact, do something small in my own backyard, and so I founded Creative Destruction Lab. And the mission of Creative Destruction Lab is to enhance the commercialization of science for the betterment of humankind.
In 2012, a grad student came to the brand new Creative Destruction Lab, and wanted to use this technique of deep learning to predict which molecules would most effectively bind with which proteins to enhance how we do drug discovery. In 2013, another grad student
was using the same technique, but to predict fraud in financial transactions. Another came, using the same technique to predict, using 30 second audio clips on the phone to predict early detection of Alzheimer's, and then the third one was using video surveillance data to predict elevated risk in home surveillance.
And so at that point, I was like, wait a minute, this same core technology is being used in all these different applications. And so, in 2014, I had seen enough of these examples, I started to study it, and study it in the same way I've been doing other areas of economics of science. And then in 2015, we decided to dedicate an entire stream of Creative Destruction Lab to the commercialization of machine learning, and it's hard to believe since it's everywhere today, but in 2015 that was the first program in the world dedicated to commercializing machine learning.
Why has Toronto, and the University of Toronto in particular, historically been such a hotspot for AI breakthroughs?
University of Toronto is a great research university. Canada has many great research universities. The University of Toronto is one of the biggest, but just having our research system enabled people like Geoff Hinton [one of the so-called Godfathers of AI] to do what he was doing and to attract students who would come and junior professors, so that Toronto had a disproportionate number of people working on this because they were crowded around Geoff. And so that's really what gave us our head start.
What are some recent success stories from Creative Destruction Lab?
We see things early, earlier than when the venture capital companies see things. They see things when they're starting to have some kind of commercial traction. We see them before that. So I would highlight those early AI projects from Toronto — like the first company I was mentioning in 2012 that was called Atomwise. And then another one that followed soon after that was called Winterlight. They were the ones doing early detection of Alzheimer's, and then another one was called Deep Genomics. And so they kind of catalyzed the field.
And we've had similar success in, for example, quantum. We had the first quantum computing stream in the world, focused on commercializing quantum, in like 2017. And then we just had the first one of our alums go public, which is Xanadu. And so that's, you know, that's a notable success. But I think that the successes are less sort of about the companies specifically and more about the system. The ecosystem has brought together the researchers, the investors, the entrepreneurs, and the scientists, everybody — it's like alchemy.
You've written that AI in itself is not a goal, it's a tool to achieve a goal. What do you think is the biggest mistake organizations make when implementing AI into the workplace?
They don't set a sufficiently ambitious goal. The vast majority, what they do is they take a process the way it is today, and then they figure out what they can do with AI to do it a little more efficiently — like automate some steps to cut some costs — as opposed to asking, “what is something much more ambitious that we could do?”
And so when Prime Minister Carney asked the AI minister [Evan] Solomon, to put together a strategy, and then he put together a task force, there were a bunch of people on that task force, and I was one of those people. My recommendation for Canada was to set moonshots. I had one in healthcare, one in defence, one in homelessness, one in natural resources, and one in education, and each one was to set a wildly ambitious goal that you could not do without machine intelligence. And machine intelligence alone wouldn’t solve the problem, but it was a key unlock that along with some other parts of the solution would drive it.
Can you give a specific example of a moonshot?
Many Canadians feel proud of our healthcare system because it's, in theory, universally accessible. Except that many people have to wait so long for treatment that, if they have the means, some leave Canada and go somewhere else for their treatment. And some end up suffering great consequences because of how long they have to wait before they actually get help. So the moonshot in healthcare was to reduce all the waiting times across the board by 90%. And that's the kind of goal where there's no way you could do that without machine intelligence. But you can't just point AI at it — it requires a number of organizational design changes in addition to the implementation of AI. AI is like an optimizer, and so it can make many things much more efficient, but it can also redesign how we operate things because we have such a powerful prediction tool.
So how do you get the Canadian government to buy into a moonshot?
Great question. I mean, I proposed it to them and they've just recently released their strategy, and so the proof will be in the pudding. We'll see what actually gets done.
Another interesting concept that you've talked about is the “genius supply shock.” Can you explain what that means and why it's relevant for a modern workforce?
Imagine that some rich person came along and said, “hey, I am going to donate to The Peak and give you 12 Nobel Prize winners for the next year. They're going to just come in, sit in your office, find computers, and they'll be there from nine-to-five and do whatever you need them to do.” My guess is at least for the first few months, nobody at The Peak would really know what to do with them. And that's what I mean by supply shock, which is all of a sudden, all these companies and organizations get access to this incredible intelligence. And we're not set up to use it. Because we're not used to having it. And so we don't even know what to ask it to do.
Right now people are using it to do things like, “help me compose an email,” and “help me figure out what recipes I should cook for a party I'm hosting on Saturday night.” And so, it will take a while for companies and organizations and governments to figure out how to leverage this incredible power. And the reason we don't know what to do with this is because we've optimized all our systems for our current level of cognition. We've got a handful of people and this is what we do with them. So the supply shock is going to lead to a total restructuring of almost every part of the economy, and that's why it will take time for everything to adjust to get used to it.
How do you see this playing out?
I was at a cottage maybe a month ago with a small group of people, and Chris Hadfield was there. We were playing some music — everybody had guitars and things — and then, we took a break because there was going to be a SpaceX launch. So everybody put down their instruments, walked into the TV room, and we watched this launch, and Chris was narrating it, like he was sitting there and explained to everybody what's going on — why is Starship so important. And what he said was, he said, “Starship is going to reduce the cost of taking assets to low Earth orbit by about 100 times.” And then to kind of make the point so that people in the room understood what that meant, he said, “imagine if all of a sudden you woke up one day and instead of costing 400 bucks to fly from Toronto to Mexico, it costs four bucks. How would things change if all of a sudden it costs four bucks to fly to Mexico?”
And so in economics, we call this general equilibrium. So the first effect is if it was four bucks to fly to Mexico, that a whole bunch of people would start flying to Mexico, like just for the weekend or something because it's so cheap. Now that a whole bunch more people are flying to Mexico, there's a lot more demand for hotel rooms, and because there's demand for hotel rooms, the prices of hotel rooms go up. So even though it's now way cheaper to fly to Mexico, it's way more expensive to stay there.
My point is, you get all these ripple effects. That's what's going to happen in the economy with AI. AI is making prediction cheap. And as it becomes cheaper, it has all these second and third and fourth order effects in the economy, and everything has to readjust, just like with the construction workers in Mexico. Prices adjust in order to put everything back in equilibrium. And we'll all figure out what the kind of new world will look like and start adapting to that.
Is another issue that — even when we identify what to do with all these geniuses — workers will have an incentive to sabotage them in an attempt to protect themselves?
There will be some of that, for sure. But I think that'll be a short-term thing. In other words, eventually people will get used to it and they will figure out that there are very compelling things that they can do. Like, there used to be rooms full of people that every time someone made a phone call, somebody would be in a switchboard and they would pull out a cable and they would put it in another slot, that was their job, and I don't think anybody says, “oh, I really wish I could go back and have that job.”
A lot of people, particularly young people, are very pessimistic about AI. What would you say to them to dissuade AI doomerism?
In my view, we have done a very poor job of communicating to young people the incredible opportunity that lies before us. The difference between what we could do with AI versus what we're actually doing is a massive gap. That means the returns to imagination are huge. There are so many things that we could imagine, and at the moment we are just scratching the surface.
So there are all the grown-ups who are running companies and working in companies, and they're taking AI to make their workflows — like their taxes, or accounting, or whatever — a little more efficient and increasing their profitability. That is small potatoes. Meanwhile, Elon is trying to make us a multiplanetary species, and also putting chips into our brains, and already there are 25 people on planet Earth that have a chip in their brain and are able to interact telepathically with their computer. Just think about what they want to do on their computer and move their cursor around through thought. In other words, there's all these things happening.
So, in my view, for young people, the opportunity is so incredible that nobody even knows what to do with it yet. It’s like we're building the railroad and opening up the West, and the frontier is limitless — like it's literally limitless. And probably because we haven't had that kind of limitless frontier for quite some time, people don't know what to do with it. I bet young people will have way more creative ideas of what to do than people who are already well-established in their careers. So, I'm very optimistic, and I think it will just take a few years for examples of the types of new things that people start coming up with, and then the floodgates will open.
This interview has been lightly edited for length and clarity.
