As an expert in AI, Joshua Gans spends a lot of time separating hype from reality. Currently a Professor at the University of Toronto’s Rotman School of Management, he teaches MBA students networking and digital marketing strategy—including how companies can use technology to compete through innovation. We sat down with Joshua to discuss the new book Prediction Machines: The Simple Economics of Artificial Intelligence, that he coauthored with Ajay Agrawal and Avi Goldfarb. There’s a lot of talk in today’s market around the possibilities of AI. But are there concrete examples of AI’s benefits today in today’s business world?
“…new businesses that start their data collection from scratch today may end up creating the best AI data.”
There’s a lot of talk in today’s market around the possibilities of AI. But are there concrete examples of AI’s benefits today in today’s business world?
I’ll admit it: AI is seeing a lot of hype right now. In the book, we take a different approach to what AI developments over the last 10 years have been all about. We’re not talking about general intelligence—replacing humans and all their cognitive abilities—but just one facet that hasn’t been exploited previously, which is our ability to predict.
Normally, we think about prediction in the context of forecasting. With weather, we gather historical data about wind and precipitation and other factors, and we produce a prediction about what the weather’s going to be tomorrow, and next week. But prediction isn’t always about the future. Computer vision is one example: when you give a machine an image and ask what it sees, what it’s really coming back with is a prediction. The computer is asking “What would a person think is in this picture?”
Prediction is all about making better decisions. With weather predictions, you can determine what you should wear. When you have a prediction regarding what’s in an image—take an MRI, for example—you can then make a decision on the right course of treatment. Seen in that regard, AI is kind of boring. It’s just a better statistical technique. But it is such a huge advance, that prediction itself is going to become better, faster, and cheaper. That’s going to open up a whole lot of uses for prediction that weren’t there previously.
So how does AI move from hype to real value for businesses?
One of the concerns in the back of our minds as we wrote this book was recalling what happened with the computer revolution and the internet revolution. With those, there was a lot of hype, and a lot of companies spent millions of dollars on things that weren’t really thought out. We don’t want to repeat that mistake. What we instead encourage services to do is to say, “If prediction is going to help better decision making, let’s take workflows from our organization and break them down into all the decisions that have to be made in order to go from an input to an output, and in that process, identify where the sources of uncertainty are.” It’s there that you’ll start to understand where AI might be useful in reducing that uncertainty and making better decisions.
This is a process that happened previously with computers. People broke down the workflows and tasks, and worked out where computers were going to be useful. It led to this movement, 20-25 years ago, called reengineering. What we’re suggesting is that there’s an opportunity to do that again.
In the book, you write that everyone has a lightbulb moment with AI—a moment where it clicks for them. Will every industry need to see their own lightbulb to embrace the technology?
Some people are ahead of the curve, asking, “Can AI help our business?” In other situations, it does need to be more tangible.
For instance, one supermarket chain that we studied used AI to predict their cold storage levels, which if optimized can be a critical cost savings. Food space is expensive and you need to hit the right balance of supply and demand, otherwise you risk spoilage. This chain applied machine learning to look at what was driving the demand for yogurt in Canada. They found that a significant driver of whether there was more yogurt or less than expected in the store at the end of the day was the weather. Even a reduction in a few degrees in an otherwise cold Canada changed buying behavior around yogurt. And that was something that was completely surprising. They started seeing gains—5% here, 5% there—and that all starts to add up. Those are the sorts of times in which people say, “Oh, this could really matter for us.”
AI is fueled by data. There are companies that have vast amounts of data to work with, and others that are behind in that area. Will AI give an unfair advantage to data-rich companies?
That’s really quite difficult to say. You need data for AI, but the simplest way to scare everybody is to say, “They have data and you don’t.” I have no doubt that this is why companies like Google, Facebook, and Amazon have such great leadership in AI at the moment, because they’ve already been thinking about data, and they’ve been collecting it the right way. The standard business that’s been collecting data, without thinking about precisely how it’s going be used, is not necessarily in the same position. For AI, you need the right data—structured in the right way, measuring the right things, clean. It may well be that the new businesses that start their data collection from scratch today may end up creating the best AI data.
Where should AI capabilities sit in an organization?
That’s the tough question many organizations are facing. I see AI as part of an analytics core function right now, because there are still many amount elements requiring data scientists to understand the data. But long-term, that should change. First, there’s the choice of outsourcing verses building capabilities in-house, both with their own benefits and risks. It will also depend on which pieces of the organization the AI function is meant to impact. General-purpose capabilities make sense in a centralized function, but more departmentally specific functions might drive the ownership into those respective groups. Take HR, for example: they’re trying to predict if a new hire will be productive or if someone should be promoted. Today, there’s a lot of data collected that could help these predictions, but it’s all sitting in files, not being used.
What steps can business leaders take to get their business ready to take advantage of AI?
Beware of history. Beware of people bringing technological gifts. AI is ultimately distinct. Having a good understanding of the technology and what it can bring you is going to serve you well in understanding whether you’re being sold something real and what its potential is. In other words, it’s very important that you have people close to you in the organization that can tell you whether the potential benefits are real or not, from a data science perspective and an operations perspective.
That being said, there are enormous benefits to being open to experimentation. If you have a large organization, letting teams go and find an application of AI—not as a core operating piece of the business, but as an experiment on the side—can have great benefits. You need to manage your risk but take advantage of the opportunity that AI presents.