[Reubend]

This is a common misconception. Some players are extremely good at video games, and they look like statistic anomalies / outlier when mapped across the full distribution of players.

Consider, for example, professional gamers. They spend countless hours practicing, and they can easily outcompete casual gamers who don't have the time to refine their skills daily.

Statistical anti cheat is extremely weak in any game where legitimate human players can end up as outliers.

[blueflow]

Its like using APM to identify cheaters in Starcraft 1. Jaedong and Flash will get banned together with the actual cheaters.

[Strom]

Extremely good players have old profiles that they have used for a long time, gradually getting better. Cheaters are either using a new profile, or an old profile with bad stats that then has a sharp uptick.

[madflame991]

I've been playing the same 2 games for many years and I think I got pretty good at them and I've used multiple accounts - under your assumption I look like a cheater.

[Strom]

> under your assumption I look like a cheater

In a proper statistical analysis there are far more variables than what I outlined in my preceding two sentence post. It would be naive to think that I would consider anyone a cheater only based on the account age.

Smurf accounts are also bannable in plenty of games and I certainly support that.

Beyond that, the level of "good" we're talking here goes way beyond dominating in a random match. Cheater stats are usually better than literally the top #1 player in the world.

Take something like Battlefield, where on the public leaderboards the "top players" have a kill-to-death ratio in the thousands. That is so far beyond human possibility, yet they are still not banned because of this aversion to statistics.

[stavros]

I think what's naive is to assume that statistical detection methods haven't been investigated at length by the anti-cheat companies.

When a complete newcomer comes to a field and sees professionals not doing a simple thing, the right question isn't "why don't you just do this, duh", but "I thought this would work, why doesn't it?".

[Strom]

Newcomers definitely make naive assumptions, Chesterton's fence etc.

I'm not a newcomer though, I've worked on both cheats and anti-cheats going back more than two decades. I know how the sausage is made and it's not pretty.

The anti-cheat companies you talk about mostly sell a mass produced product that works very similarly to anti-virus software. Games embed the anti-cheat module and its cheat definitions get updated. Statistical analysis requries both knowledge of the specific game and access to its database. Often also additional game programming to even store the crucial data. A bespoke solution. This can't be mass produced and is expensive, so most games don't have it.

So to bring it back to the newcomer question, I thought this would work, why doesn't it?, the answer is that game companies don't want to spend the money. [1] A classic answer to most annoyances in life, really.

---

[1] An interesting outlier is the online gambling industry, especially online poker. They spend way more money than non-gambling game developers and have much more sophisticated anti-cheat systems, including statistical analysis. It's also fun to see how techniques used to get around online poker anti-cheat detection slowly make their way into mainstream gaming with a delay of about 15 years or so. As a simple example, nobody serious was even running their code on the same system as the game client back in 2005, instead parsing the video signal and simulating HID inputs. [2] Took more than a decade to see popular cheats for regular games go to that length to avoid detection. Not because the cheat developers were less capable, but because the anti-cheats didn't warrant the investment.

[2] Thus taking the battle almost completely to the statistical analysis realm. Are your mouse movements random enough, with good jitter? Does your bot take belivable micro breaks? Does your average performance, including reaction times, degrade at the end of a long session as you get more tired? Et cetera.

[Hikikomori]

Picking out the statistical outliers are not that hard, but will this not have diminishing returns? As soon as the cheaters learns that being too obvious gets you banned they'll change up how they play. Eventually there wont be much difference between the really good players and cheaters, is some false positives okay here?

Many cheaters were already trying to not be obvious, most I've encountered playing various fps games are not the typical spinbot in csgo. Instead they might play with only wallhack, aimtrigger, or even no hack, and only turn on the big hacks halfway through a game if they're not winning or think someone on the other team is hacking as well. In some games they use bots to dunk their stats when not playing.

AI detection is also coming to videogames with anybrain.gg, but seems like these can be countered with AI enhanced cheats no?

As an experienced player with an anti cheat/cheating/security interest it doesn't seem like statistics is the silver bullet you claim it to be, at least as your only detection/protection. It combined with normal protection/detection methods is likely what Riot is doing.

[stavros]

Ah, well, I defer to your industry knowledge, in that case. I'm not very knowledgeable on this sector.