Thanks for joining us Steve! To start off with, we wanted to ask what made you switch from physics and data science to AI and machine learning?
I got into physics because of my interest in mathematics. I think physics, when you get to a deep enough level, really is just mathematics under the hood. And mathematics really is the ultimate source code and operating system for everything in the entire universe. That interest carried over into pursuing an undergraduate degree and eventually a PhD in particle physics. It was great because I got to play with cutting-edge technology. My thesis was on a previously unobserved particle decay process. That was really a fun, formative, intellectual experience. I always just enjoyed problem solving, using math, science, data and the scientific method to approach problems of all kinds.
That’s really what drove all my career choices: problems that are intellectually interesting and fun. So, I suppose I’ve been working as a data scientist, but since before people were calling themselves Data Scientists on LinkedIn. The transition to machine learning and AI is very, very natural. There’s nothing I know like AI. Machine learning gets very much mystified in the popular press and in culture, but it’s just programs; it’s just computers. Computers will do whatever you tell them to do. I’m pretty skeptical of the notion that any sort of AI could really be considered to be self-aware in any sentient, meaningful way.
Can you explain why you believe AI won’t take jobs away from humans?
What I’ve noticed, at least in the gaming industry, is that the stuff we’re looking for AI to do will not. It’s not, “we want to build this AI thing to make decisions that the person will no longer be responsible for.” That’s not what we’re doing at all. An analyst at a casino property spends hours of their time manipulating data, cleansing data, moving data from one place to another, dragging and dropping spreadsheets and so forth. What we’re doing is a way to present them with data so they can absorb it in a holistic way, and make some decisions and recommendations. Those are things a computer can be trained to do very easily. When the person is spending their time hunched over an Excel sheet, it’s not that they don’t have other things they would rather be doing, or that would be more valuable; it’s that they’re stuck doing this boring stuff we just haven’t been clever enough to automate yet.
So, my thinking is that once these things are automated, that will free up the person to do actual, valuable work. In the casino industry, for example, it’s about creating a fun, recreational experience for the visitor and that’s a very human-on-human interaction. The real value of a casino is it’s a fun place where they can go and hang out with their friends, have a couple of drinks and play some games. It’s fun; that’s something you still need people for. You need the machine to do the non-fun stuff.
You’ve previously said, “AI is the co-pilot, not the autopilot.” Can you tell us more here?
The idea is that the AI is doing all the automated Excel-type stuff, all the numerical manipulation. It can make some predictions — some cross correlations that a person could do eventually, but it’ll take them a long time. The person ultimately, especially in the casino setting right now, is always going to have context that the machines will not. When faced with the dilemma of, “why is this certain machine not being played?” Well, is that machine by any chance by a stinky open sewer line? AI probably doesn’t know that, but the person walking the floor will. They can make judgments. That’s what we mean when we say the AI is the co-pilot, because the person can always grab the stick. Take off, landing, when there’s turbulence, when there’s stuff that you have to make decisions about — grab the stick and takeover.
Can you expand on the differences between your traditional offerings and your AI ones? You have a program called ‘Slots,’ but then you also have ‘Slots AI.’ What are the differences and how do they work?
Very broadly, most of what Slots is doing is reporting on numbers. Before you can report the numbers, you have to gather data from all these disparate sources. There are the slot games, table games, hotel, food and beverage, and all these various places that are run by their own different systems, which don’t really talk to each other. The data is in different formats. The player at your property doesn’t know or care about that; they want to walk in the door and have a good time. You can tell your visitor, “Hey, we would have liked to make this experience better for you,” but they don’t want to hear that, right? What we’re fundamentally doing is taking all these disparate sources and different data formats, combining them into this grand, unified framework that should be the same for every property, for every system. Once you have that, and you can give these reports in a very unified way, that’s where the AI can be added on top of that.
Everywhere you are reporting a number, in almost every case, that number can also be a prediction. I’m telling you how well this machine did last week, but with a little bit of machine learning and some clever mathematics, you can now say what will happen next week – or what will happen if this popular machine that you only have two copies of will do with a third copy. Is it going to generate some incremental value, or is it just going to reshuffle players on the floor? Is that going to enhance the player experience or detract from the experience in some way?
It’s really about enhancing the guest experience by taking these reports and findings, and translating them into predictions and recommendations so you can tell the operator something useful. You learn about these humans who are coming to your property to have fun and you can understand what’s really motivating them. What sort of player are they? What sort of things are they going to find more fun or less fun? You can adjust your operations accordingly. It’ll be about what’s happening for them, but also what will happen for you.
How customizable is your software?
We’re always generating new stuff, but the great thing about a machine-learning model is that the customization is built into the product. When we write a predictive model, we train it for your property, on your players and only your players. When we look at your slot machine floor and make recommendations, it’s based on slot machines on your property only. Machine learning,without data sharing, is inherently customizable because the customization is part of the machine. Learning is part of training them outright.
Can you tell us about your Future Player Value product?
We have a fairly sophisticated machine-learning model that looks at the player’s gaming activity, some demographic variables and makes a prediction about what we think this player’s short-term activity in slot and table games is going to be. We update that on a cadence and we push those results to the client instantly. So when the operator pulls up the profile of a given player, it’s right there. Here’s what we think those players are going to do next month in terms of their total activity.
And when operators are building campaigns rather than, or in addition to, segmenting the playerbase based on recent activity, we can now allow these properties to segment the playerbase based on who is going to do what in the next month. It’s about letting the property make decisions about the future – before the future gets here.
What is next for OPTX? And what do you think going to happen to the casino industry at large?
Right now, all the data is very much siloed from property to property. We’re also looking at a program where we would combine data for multiple properties. This is especially relevant for our Slots AI product. You may want to know: how is my property comparing relative to my peer group? Are players having to traverse the casino from one game to another or overlapping and playing various games? Are other games doing well at another property I may not have just yet?
That sort of data sharing provides a big opportunity for additional, better, deeper learnings. We’re moving into that cautiously; if we’re talking about sharing data, we have to make sure we’re going to ensure data controls remain very vigorous. It’s something we only do with explicit permission and with a very well-defined and certain scope; we’re going to use this specific data for this specific purpose. I think having that informed consent of the operator is critical. These operators are trusting us with that data and that’s something we want to treat with the utmost care.
Is there one package or software from OPTX you think people have really benefited from the most over the years?
Probably not the AI stuff...
That’s right. What I hear from properties is they love the ability to mass communicate with players. It’s something that sounds like it should be super easy, but it’s something that always takes a village. It takes a lot of careful coding and careful programming. That’s why we have a great team of developers working on that.
So, whether AI or not, is the focus just on the players?
I really think it is. If you think of the two broad categories of what you can do to the property, it’s messing around with the game content and how you incentivize, reward and enhance the player experience. I think the players’ side is much more important, which is difficult. Slot machines are like computers; they all behave in a very predictable, well-defined way, whereas human beings do not. If you really understand the incentives and motivation for people, you have a good chance of understanding how a population of players will behave. That’s fundamental: the economic premise of understanding behavior.