The challenge now is to identify situations in which prediction will be valuable.
How Cheap Prediction Creates Value
Prediction is the process of filling in missing information. It takes information you have — often called ‘data’ — and uses it to generate information that you don’t have. Much discussion about AI emphasizes the variety of prediction techniques using increasingly obscure names and labels: classification, decision trees, neural networks, topological data analysis, reinforcement learning, and so on. These techniques are important for the technologists interested in implementing AI for a particular prediction problem.
We will spare you the details when we emphasize that each of these methods is actually about prediction — using information you have to generate information you don’t have. The challenge is to identify situations in which prediction will be valuable, and then figure out how to benefit as much as possible from that prediction. As indicated, when arithmetic became cheap, we started using it on problems that weren’t traditional arithmetic problems. Whereas we once created photography with chemistry, we transitioned to an arithmetic-based solution: digital cameras. A digital image is actually just a string of zeros and ones that can be reassembled into a viewable image.
The same goes for prediction, which is used for a wide variety of traditional tasks, from inventory management to demand forecasting. And because it is becoming cheaper, it is being used for non-traditional prediction problems. Kathryn Howe, of Integrate.ai, calls this ability to see a problem and re-frame it as a prediction problem ‘AI insight’, and today, engineers all over the world are acquiring it.
For example, we are transforming transportation into a prediction problem. Autonomous vehicles have existed in controlled environments for over two decades. They were limited, however, to places with detailed floorplans such as factories and warehouses, so that engineers could design robots to maneuver with basic ‘if-then’ logical intelligence. For example, ‘If a person walks in front of the vehicle, stop’; ‘if the shelf is empty, move to the next one’.
However, autonomous vehicles could not function outside of a highly predictable, controlled environment until engineers re-framed navigation as a prediction problem. Instead of telling the machine what to do in each circumstance, engineers recog- nized they could instead focus on a single prediction problem: What would a human do? Now, companies are investing billions of dollars in training machines to drive autonomously in uncontrolled environments.
Imagine an AI sitting in the car with a human driver. The human drives for millions of miles, receiving data about the environment through its eyes and ears, processing that data with the human brain, and then acting in response to the incoming data. Engineers are basically giving the AI its own eyes and ears by outfitting the car with sensors (e.g., cameras, radar, lasers). So, the AI observes the incoming data as the human drives and simultaneously observes the human’s actions. When particular environmental data comes in, does the human turn right, brake or accelerate? The more the AI observes the human, the better it becomes at predicting the specific action a driver should take.
Critically, when an input such as prediction becomes cheap, it can enhance the value of other things. Economists call these ‘complements’. Just as a drop in the cost of coffee increases the value of sugar and cream, for autonomous vehicles, a drop in the cost of prediction increases the value of sensors to capture data on the vehicle’s surroundings. For example, in 2017, Intel paid more than $15 billion for the Israeli startup Mobileye, primarily for its data-collection technology that allows vehicles to effectively see objects (stop signs, people, etc.) and markings (lanes, roads). When prediction is cheap, there will be more prediction and more complements to prediction, and these two simple economic forces will drive the opportunities that prediction machines create.
From Cheap to Strategy
The single most common question corporate executives ask us is, How will AI affect our business strategy? We use a thought experiment to answer that question. Most people are familiar with shopping at Amazon. As with most online retailers, you visit its website, shop for items, place them in your cart, pay for them, and then Amazon ships them to you. Amazon’s current business model is ‘shopping-then-shipping’.
During the shopping process, Amazon’s AI offers suggestions of items that it predicts you will want to buy. The AI does a reasonable job, but it is far from perfect. In our case, we actually
purchase about one of every 20 items it recommends. However, considering the millions of items on offer, that’s not a bad ratio.
Imagine that the Amazon AI collects more information about us over time and uses that data to improve its predictions — an improvement akin to turning up the volume knob on a speaker dial. At some point, as it ‘turns the knob’, the AI’S prediction accuracy crosses a threshold, changing Amazon’s business model: The prediction aspect becomes sufficiently accurate that it is suddenly more profitable for Amazon to ship you the goods that it predicts you will want, rather than wait for you to order them. Cranking up the prediction dial changes Amazon’s business model from ‘shopping-then-shipping’ to ‘shipping-thenshopping’.
Of course, shoppers would not want to deal with the hassle of returning all the items they don’t want. So, Amazon would invest in infrastructure for product returns — perhaps a fleet of trucks that do pickups once a week, conveniently collecting items that customers don’t want.
If this is a better business model, why hasn’t Amazon done it already? Because if implemented today, the cost of collecting and handling returned items would outweigh the increase in revenue from a greater share of wallet. As indicated, today we would return 95 per cent of the items Amazon ships to us, which would be annoying for us and costly for the company. Its predictive ability isn’t yet good enough for Amazon to adopt this new model.
We can imagine a scenario where Amazon adopts this new strategy even before the prediction accuracy is good enough to make it profitable because the company anticipates that at some point, it will be profitable. By launching sooner, Amazon’s AI would get more data sooner and improve faster. Of course, Amazon realizes that the sooner it starts, the harder it will be for competitors to catch up: Better predictions will attract more shoppers, more shoppers will generate more data to train the AI, more data will lead to better predictions, and so on, creating a virtuous cycle. Adopting too early could be costly, but adopting too late could be fatal.
Our point is not that Amazon will or should do this — although skeptical readers may be surprised to learn that Amazon obtained a U.S. patent for ‘anticipatory shipping’ in 2013. The salient insight is that turning up the prediction dial can have a significant impact on strategy. In this example, it shifts Amazon’s business model from shopping-then-shipping to shipping-thenshopping, generates the incentive to vertically integrate into operating a service for product returns (including a fleet of trucks), and accelerates the timing of investment.
In closing
Leaders across industries would do well to invest in gathering intelligence on how fast and how far the dial on the prediction machines will turn for their particular sector and applications. They should also invest in developing a thesis about the strategic options created from turning the dial. To get started on this ‘science fictioning’ exercise, close your eyes, imagine putting your fingers on the dial of your prediction machine, and — in the immortal words of Spinal Tap — turn it up to eleven. Ajay Agrawal is the Peter Munk Professor of Entrepreneurship and Professor of Strategic Management at the Rotman School of Management and Founder of the Creative Destruction Lab, the seed-stage program for scalable, science-based companies launched at the Rotman School and now running at five business schools across Canada and at NYU’S Stern School of Business. Joshua Gans is the Jeffrey S. Skoll Chair of Technical Innovation and Entrepreneurship and Professor of Strategic Management at the Rotman School. Avi Goldfarb is the Ellison Professor of Marketing at the Rotman School. They are the co-authors of Prediction Machines: The Simple Economics of Artificial Intelligence (Harvard Business Review Press, April 2018), from which this article has been adapted.