CS:GO Fan Busts 14,000 Cheaters Using Homemade AI

CS:GO Fan Busts 14,000 Cheaters Using Homemade AI
Image: Valve
To sign up for our daily newsletter covering the latest news, features and reviews, head HERE. For a running feed of all our stories, follow us on Twitter HERE. Or you can bookmark the Kotaku Australia homepage to visit whenever you need a news fix.

Everyone hates cheaters, and games like Counter-Strike have had a long-standing issue trying to combat the worst of the worst. So one fan decided to take matters into his own and built a deep learning platform, which has successfully nixed over 14,000 cheaters to date.

As reported by The Loadout, one British CS:GO fan and teenager 2Eggs — who received more than $US11,000 for reporting CS:GO vulnerabilities through Valve’s bounty program — explained that they built the HestiaNet deep learning platform “to heal over [CS:GO’s] infestation and to get rid of as many cheaters as possible”.

“To many of us in the community, CS:GO is a home, and Hestia is also the protector of the house,” the creator told The Loadout.

The platform was inspired by a demo from a senior Valve engineer at GDC, where John McDonald explained how Valve was using AI and deep learning methods to combat cheaters.

2Eggs looked at that and decided to give it a shot themselves. The way it works is this: hooking HestiaNet into his main account to review Overwatch cases, the community-sourced anti-cheat program within CS:GO.

From there, HestiaNet reviews the footage and then submits a report to Valve. If the report is accurate, HestiaNet logs that as an accurate find and uses that case as training data, helping it better differentiate between cheaters and regular players.

So far, HestiaNet has reported 14,782 cases with 14,515 successful reports, a hit rate of just over 98 percent. The vast majority of those were for cheats as well, with less than one percent of bans handed out for griefers.

[Thanks, The Loadout!]

Comments

  • This seems like a terrible idea. Valve is already running a machine learning system to detect cheaters, with the overwatch cases providing both human oversight of its decisions and training to have it produce better decisions.

    If you replace the human component with a second AI, you’re compromising those goals. If you got to a point where HestiaNet and systems like it were handling the majority overwatch cases, then you’d essentially have two AIs reinforcing the decisions of the other. This could easily reduce the overall accuracy of the system.

      • Neither. In fact, I’ve never played CS:GO myself.

        If the game is important to you, would you really want someone to compromise the review system like this? Do you want to give real cheaters a reason to dispute their ban? Do you want to risk non cheaters getting banned in error?

        If Valve wanted to set up a cheat detection system with no human oversight, they could easily do so themselves. It’s not the place of a random community member to remove that oversight.

      • If that’s just a joke then fair play. If not, there is a real issue as @Jamesh points out. There is a real risk that game companies will get complacent and just rely on AI to do the work for them. And it’s entirely possible they’ll just ignore people who complain that they were banned falsely because they trust that the AI is “just right”.

        It’d be terrible to be banned for a fluky shot or a run of good luck because by coincidence it lines up with whatever the AI uses as an identifier of cheating. Bear in mind also that there are little exploits in games that aren’t actually cheats. I remember a few locations in CS where the sound of footsteps gave you a really accurate idea where the enemy was. You could “exploit” that and shoot through a wall and regularly get kills as a result. That was absolutely not cheating, but it could look like it to an AI.

        Side note: The bit in the article explaining how the analysis works is badly written. It’s very hard to parse what is being said.

    • False positives? Nah, there’s a far more likely scenario…

      They simply got themselves a fancy magic seer robot that can detect people who are going to cheat in the future.

      No false positives here, no sir.

Show more comments

Log in to comment on this story!