Q: What are you seeing these days in terms of the value of artificial intelligence? Where can it be of particular value in logistics and material handling?
Peck: Artificial intelligence allows us to solve extremely difficult problems in a relatively short amount of time. I came out of the world of academia, where we did these types of things in research mode, and now we're putting them into practice in logistics.
Q: What are some specific functions that take place in the world of logistics where artificial intelligence can be of value?
Peck: Two things that come to mind immediately are order fulfillment – how we pick large numbers of things quickly and accurately – and packaging: how we select cartons in such a way that there will be a minimal amount of air shipped with them.
Q: When it comes to order fulfillment, what can A.I. do better than a human?
Peck: Let me give you an example. Some companies are shipping upwards of 60,000 orders a day. With that kind of volume, you can't simply take out a box, fill it up and do that for all 60,000 orders. You have to do what we call batch picking. Typically you would find a cart with multiple orders on it. But even if you’re handling 10 orders at a time, you're still going to need 6,000 carts worth of work in order to pick those 60,000 orders. That's still not going to make it.
What you need to do is to step back and say, which orders should I put on each cart in such a way that will minimize my walking through the warehouse? If I can put 50 orders on a cart that can all be picked in one area of the warehouse, rather than running all over the whole facility, it would make a lot of sense. But there are other constraints that many customers have. For example, when I get to a picking location, I'd like to have a lot of commonality among orders. There’s also the question of weight. The total cart can't weigh more than a certain amount because it's difficult to push it around. Or a box on the top shelf can't weigh more than a certain amount, because it’s very difficult to get down. You cannot do an exhaustive search of all combinations of the ways we could put these 60,000 orders on carts. It takes too much time.
So you have to use some very clever artificial intelligence techniques. We use what we call biologically inspired algorithms – we breed together solutions to come up with an even better solution. These are the types of things we've taken from the academic world.
Q: Artificial intelligence involves a brute-force approach to information processing – too complex for a human being to even consider. I suppose it takes analytics on the other end to ensure that a coherent result comes out.
Peck: Right. If I look at all combinations, it would take literally years, maybe a century with the world's fastest computer. So you have to figure out clever ways of doing that. You might not get the 100-percent optimal solution, but if you can get 95 percent of optimal in a few minutes, that's really what you want to do.
Q: Where does packing come into it?
Peck: People typically have maybe eight to 12 different box sizes with which to pack orders. With e-commerce, you have a small number of items in a box, and if there’s a lot of empty space or dunnage surrounding it, that costs money. All of the small-parcel carriers now charge on the basis of package dimension as well as weight. As a result, people who are continuing to ship larger boxes than necessarily are paying a whole lot more money for their shipping.
Q: So how do you determine what the right box size should be?
Peck: If I have ten or 12 to select from, I don't have a lot of choices. What we say is, "Why not have 100 box sizes?" With that many, you'd do a whole lot better in terms of shipping less air. But the question then comes up, if I had the luxury of 100 box sizes, what should they be? That's a tough problem.
For one of the major retailers, we looked at the historical data for over a million orders that had been shipped. We knew what was in the orders, the quantity of each order, and the containers they were shipped in. Using artificial intelligence, we could then go ahead and decide which set of 100 or 150 boxes would be best to have on hand.
Q: But it’s going to cost to you have 100 boxes on site. Is that cost offset by the savings from shipping less air?
Peck: You have to have 100, but not six months worth of each type. All you need is a few days of supply. But determining which 100 sizes is a pretty difficult problem. We have what are called cubing algorithms. People take two approaches to that. One is to do what they call liquid fill: what are the cubic inches of this container and the item, and is it going to fit? Typically you start with the smallest item and work your way up. But the problem is, once you've put that first one in, the inside of the box is no longer the same. As a result of that, the liquid fill method really doesn't work well. So others have used what we call 3D placement. It looks at each item relative to the others, to figure out whether or not it’s going to fit.
Q: It’s a puzzle, fitting in all the pieces in three dimensions.
Peck: That’s where artificial intelligence comes in. Even nesting, where each tapered trash can can be placed inside of the other, isn’t the sum of the dimensions of each trash can. Similarly, the inside of trashcans are hollow, so things can be packed in those. The algorithm that you put together has to understand all of these possibilities, in figuring out which box size is best.
Q: Moving from the world of academia, did it challenge any of your theories when you had to apply them to the so-called real world?
Peck: In academia, many times when a problem becomes too difficult to solve, you relax some of the constraints – simplify the problem until you can get a solution that's good enough. That doesn't work in the real world. So yes, problems in the real world are often more difficult than those we find in academia, when it comes down to actually solving them.
