Video games are big business, and big business attracts criminals, fraudsters, and cheats. Somewhere amongst the millions of law-abiding game players are a large handful of shady characters whose main goal is to extract value for themselves or their organizations at the expense of other players and the game publisher.
Security teams face the daunting task of rooting out these bad guys and halting their fraudulent activities. It’s not always easy.
Unlike most real-world security teams, teams working in online video games deal with millions of players every hour of every day. That’s simply too many to observe with human eyes. Digital fraudsters may also login and out of games, servers, and accounts at will, effectively disappearing and reappearing at times that are convenient for them and their plans. It’s hard to track and find criminals who constantly change location and appearance.
However there are common threads and patterns that can identify the bad guys if you know where to look and how to look. Odd player behavior, known associates, IP addresses, familiar fraudulent play patterns, and many other attributes in aggregate can betray a video game criminal if you can put it all together. But, the current reality of video games is that there are too many players engaging in too many different and subtle behaviors that produce too much data for any team of live humans to comb through.
Human-centric video game security teams may catch a few fraudsters, but they won’t catch most of them; and certainly not all of them. Like many other data-oriented industries that are struggling with the volume and breadth of their own data, video game security is poised to turn to machine learning and related analytic techniques to sift through the terabytes of organic game play to find the rotten players and their ruinous behavior.
There are three roles that machine learning can play in a video game ecosystem. They are:
Machine learning as leverage - A beautiful thing about machine learning is that you can teach it what you know. Anything that you do with data that can be codified into rules, patterns, or tendencies can be fed into an algorithm that learns to do those things you do most often. If you typically search for sequential patterns of player behavior, machine learning can do it for you.
If you often look for outliers in in-game currency balances, machine learning can show them to you before you even ask. Having machines automate your most common and rote tasks is like replicating yourself a million times, freeing up the real you for more intricate or abstract tasks.
Machine learning as detective - Sometimes we recognize cheaters when we see them; sometimes they’re really good and it’s not obvious until later, if we ever realize it at all. Naturally, security teams would go after the obvious cheats and frauds first. They’re the easiest to find and the easiest to stop. But beyond the obvious group is a class of criminals who are operating under a modicum of cover.
It’s not obvious when you observe them whether they are very good players, very creative players, or very fraudulent players. A human observer might have to watch them for a while before making a judgment, whereas machine learning can characterize the early-warning signs and the patterns of behavior that criminals use. Algorithms can detect the subtlest but statistically significant criminal behaviors well before a human could, even assuming that a human was observing the fraudulent player constantly, which is a lot to ask of a resource-constrained security team. Machine learning can bring the most-wanted suspects to the forefront based on multi-faceted digital evidence well before a human would notice - if they ever did.
Machine learning as watchdog - Better than real watchdogs, well-managed computer systems never sleep. All you have to do is teach it what to look out for, and it will alert you when it sees it. Going far beyond simple rule-based systems, machine learning systems can fit a model to suspicious behavior and deliver the primary culprits even if they haven’t tripped any of the obvious wires. Even a simple machine learning model of suspicious behavior can run over a game’s data constantly, day and night, and send messages and warnings as desired.
Machine learning isn’t the solution to all of the problems in video game security, but it can take us a long way towards our goals. We can use it to multiply the efficacy of security teams, let it teach us things we didn’t know about our game ecosystem, and set it lose in the yard to alert us of problematic intruders.
One thing is certain, though: the criminals of the video game industry are operating on the digital level, and the industry’s law enforcement should be digital, too.