This content is locked. Please login or become a member.
Machine Learning
Machine learning is technology that learns from experience. By experience, I mean data. So data isn’t just a bunch of boring ones and zeros. It’s a long list of prior events. It encodes the collective experience of an organization from which it’s possible to learn to derive a predictive model. What’s learned and what you want to get from data is how to predict, because predictions are the most actionable form of insight you can get from data. Prediction is the holy grail for driving large-scale decisions.
So, for example, predict who’s going to buy in order to decide who to contact with marketing. Which transaction is most likely to be fraudulent to decide which transactions to block or audit. Which train wheel is most likely to fail in order to decide which one to inspect. Which healthcare patient should we take another look at before discharging because they’re predicted very likely to be readmitted to the hospital. So on that level of detail, and that’s what differentiates these types of predictive use cases, predictive AI, predictive analytics, same thing. These use cases of machine learning, that’s what differentiates it from forecasting.
Because instead of just predicting, “Hey, is the economy going to go up or down?” One universal overall prediction. Or how many ice cream cones are we going to sell next quarter? Instead, we’re predicting per individual, which individuals are most likely to buy an ice cream cone? And by ordering or prioritizing from most likely to least likely, highest risk or highest opportunity down to lowest, then you draw the line, decide, okay, these are the ones worth treating, incarcerating, setting up on a date, investigating, contacting with marketing, all of those decisions that make up the operations that organizations conduct. A little prediction goes a long way. Predicting better than guessing, generally more than sufficient to deliver a tremendous improvement to the bottom line.
So we have data. We give it to machine learning, which is the underlying technology. It generates models that predict, and those predictions improve all the large-scale operations that we conduct.
The Broader Science of ML
Machine learning as a science really is a bit more general than that. The idea of learning from data to predict is also the underlying technology for generative AI. So in that case, instead of predicting per individual sort of organizational unit, it’s predicting per individual word or token, kind of part of a word. But on that level of detail, “Hey, I’ve written these words so far in the last few paragraphs. What should the next word be?” Predict that. Okay. Then write that word. That’s how you’re applying the same underlying technology, but for a different purpose in order to generate a new content item like written text or images or video, this kind of thing. So generative is also built on machine learning, but when people say machine learning in a business context, they’re usually talking about these predictive use cases.
A Case Study in ML
Predictive AI or enterprise machine learning is so applicable across industries. Let’s take the delivery industry. UPS is one of the biggest three delivery companies in the United States, and they actually streamlined the efficiency of their deliveries by predicting tomorrow’s deliveries. So this is how it works. In the US, there’s about a thousand shipping centers. And what they have to do every night overnight is to decide which packages to assign to which trucks so that tomorrow’s deliveries, when the trucks go out in the morning, they’ll have relatively optimal routes that don’t take too many miles of driving, too much gasoline, too many, too much time of the drivers. And in aggregate, that’s a really hard optimization problem.
Now when they have to start planning and then loading the trucks in the late afternoon or early evening so that it’ll be ready the next morning, they have incomplete information. So what they don’t know is some of the packages that are still coming in later that night. They already have a bunch of packages in hand that they know are meant to go out tomorrow morning for their final deliveries. And they’ll augment that with tentatively presumed predicted deliveries by applying a predictive model for each potential delivery address and saying, “Hey, what are the chances that there’ll be a delivery there tomorrow?” Now they have a more complete picture of all the deliveries needed for tomorrow. Some of those predictions will be wrong, but they’re confident enough that the completeness now actually overweighs some of that uncertainty. That makes such a big difference that in combination with another system that actually prescribes the driving directions, to this day, UPS enjoys savings of three hundred and fifty million dollars a year and hundreds of thousands of metric tons of emissions.