Wednesday, March 28th 2018
Valve Uses over 1,700 CPUs to Catch CS:GO Cheaters
Valve was one of the first to run a centralized game distribution + multiplayer service, through Steam, and its in-house alternative to the Punkbuster anti-cheating software, in the form of VAC (Valve Anti-Cheating). Over the years, VAC has evolved to leverage newer technologies. Its latest avatar is VACnet, an AI deep-learning system built up of over 1,700 CPUs, to more intelligently sniff out cheaters. VACnet leverages CS:GO's player-operated replay system. It studies replays of players who have been reported for cheating, and studies their replays for patterns of cheating, such as wallhacks, aimbots, and more.
What makes this different from previous approaches to the problem is that the AI evaluates the behavior of the gamer through their inputs in a way only a human could, before this. It also makes up its own criteria for spotting cheaters, as it learns more about cheating, so creative and new cheaters are quicker to spot. The decision to suspend or ban players ultimate falls in the hands of humans. VACnet sniffs out the most probable cheaters, and reports its findings to human moderators that determine guilt, and hand out suspensions or bans. VACnet has the potential to increase "convictions" of cheaters by close to four fold.
What makes this different from previous approaches to the problem is that the AI evaluates the behavior of the gamer through their inputs in a way only a human could, before this. It also makes up its own criteria for spotting cheaters, as it learns more about cheating, so creative and new cheaters are quicker to spot. The decision to suspend or ban players ultimate falls in the hands of humans. VACnet sniffs out the most probable cheaters, and reports its findings to human moderators that determine guilt, and hand out suspensions or bans. VACnet has the potential to increase "convictions" of cheaters by close to four fold.