Groundbreaking ideas and research for engaged leaders
Rotman Insights Hub | University of Toronto - Rotman School of Management Groundbreaking ideas and research for engaged leaders
Rotman Insights Hub | University of Toronto - Rotman School of Management

The Simple Economics of Artificial Intelligence

Read time:

Ajay Agrawal

Here is a transcript of the video:

Most of the things we build with AI’s will be tools. They will be very specific prediction machines that will focus on a very specific prediction problem.

And we’ll use those tools like we use any tools in business in order to make some process more efficient. So it will enhance the productivity of whatever we normally do towards our ultimate goal of executing against the strategy of the organization. So whatever your organization, whether it’s a for-profit or not for profit; a business, a hospital, a school, whatever it is, there will be a strategy that will use AI tools to make it more efficient in executing against that strategy.

But occasionally, a prediction tool so fundamentally changes the underlying economics of the business, that instead of using it to execute against the strategy, it changes the strategy itself. And that’s what I’m going to talk about next.

A lot of this is straightforward blocking. It is, if you think about the application of AI part science and part art. The science part are the things you can write down, explain to people, and put in…roughly think of it as putting it into a manual and people can follow step by step through the manual, and execute.

But there’s part of the business application of AI that isn’t a science. It’s an art. And that’s what I want to talk about here, is the art part.

Avi talked about the first…I would say there are three categories of art. The first is what Avi [Goldfarb]. When Avi was describing dropping the cost of prediction he said, remember your Economics 101 with downward sloping demand curves. As something becomes cheaper we use more of it. And he gave a bunch of examples, and he talked about using prediction for familiar things like in insurance, predicting various risks in terms of demand forecasting, supply chain management, things we traditionally use it for.

But then he talked about using prediction for problems that weren’t traditionally prediction problems, converting problems into prediction problems in order to take advantage of the new cheap prediction.

And the example he gave was driving. Driving was, even as recently as six years ago, people working on autonomous vehicles said that was an intractable problem. We wouldn’t have a driverless car on a city street in our lifetime. Until it was transformed into a prediction problem and effectively became, as he said, predicting what a good human driver would do.

And then we took translation and converted that into a prediction problem. That used to be a rules problem; we used the linguistic rules for translating from one language to another. Until we turn it into a prediction problem.

And now machines can predict, in most cases, at the same level as reasonable translators, and they are getting up to professional translator levels in some categories.

And that is an art form. It is an art form of taking problems and finding out…and thinking through which of these problems can we convert into a prediction problem?


This video was filmed as part of the Big Ideas speaker series on April 16, 2018.

Ajay Agrawal is the Geoffrey Taber chair in entrepreneurship and innovation and professor of entrepreneurship at the University of Toronto's Rotman School of Management where he conducts research on the economics of artificial intelligence, science policy, entrepreneurial finance, and geography of innovation. Professor Agrawal is a research associate at the National Bureau of Economic Research in Cambridge, MA, co-founder of The Next 36 and NextAI, and founder of the Creative Destruction Lab.