Robocalypse Now: Using Deep Learning to Combat Cheating in Counter-Strike: Global Offensive
ฝัง
- เผยแพร่เมื่อ 17 มิ.ย. 2020
- In this 2018 GDC session, Valve's John McDonald discusses how Valve has utilized Deep Learning to combat cheating in Counter-Strike: Global Offensive.
Join the GDC mailing list: www.gdconf.com/subscribe
Follow GDC on Twitter: / official_gdc
GDC talks cover a range of developmental topics including game design, programming, audio, visual arts, business management, production, online games, and much more. We post a fresh GDC video every day. Subscribe to the channel to stay on top of regular updates, and check out GDC Vault for thousands of more in-depth talks from our archives.
52:58 was an amazing question, this actually happened in 2020 with the development of bots to give bad data and started wrongfully convicting many players resulting in the overwatch system's removal
In the far future. Hackers had to write their cheats to absolutely simulate how pro gamers play and confine their cheat's ability to humanly possible level. Meaning, a bunch of AI bots playing just like humans do. Then, NPC in games starts to act just like humans. Speaking of putting humans out of their jobs.
They are doing this now. Guess not that far into the future.
Ah, I see you've played TF2 in the last couple years.
i think its near impossible to have such an advanced bot that could play like a human, as each player has their own style and skillset so a few bots wont be able to simulate well enough, and as time moves on ai gets better but so do we...gta5 has amazing graphics, but in 2020s you and i will know if a random video being shown as actual news (like a plane crash) is from a game or not, but many people who never heard about gta or older people would struggle to tell them apart
@@tdpro3607 it is already possible tho
those accounts just take advantage of the nav mesh created for bots in single player. it's not that advanced. this was happening in css in 2014. @@SanicStudios
Is there an update talk in 2019 or 2020? it’s a nice topic.
no
59:10 it's over
Yup. They knew the whole time and they've never pushed the issue any further than this.
The game has been totally hijacked by cheaters and that's part of the design.
I'd love to ask if they ever ran widely available/their own cheats and trained the network on them? Because then they'd know when exactly the person/bot was cheating and triggering aimbot right?
U are a genius
Is this a reupload?
I think so, I've watched it before. Maybe it was on the GDC Vault before and is now on the main channel?
it says GDC 18 on the podium, I am going to guess this is from 2018, it also says 2018 GDC session in the description :|
I feel really stupid that it took 30 minutes until I realized...
instead of making an invasive client side anticheat, valve decided to do the hard way instead: train a deep learning model to detect cheaters
The problem with Valve's efforts to avoid a "treadmill problem," is that operating an online multiplayer game is ITSELF a treadmill problem. If they want to continue to make hundreds of millions of dollars like they do, they're eventually going to have to accept that hard, consistent, recurring work is simply a part of software development.
There will never be a magic bullet that solves every current and future problem in cybersecurity. Instead of wasting time searching for the holy grail, they'd be much better off just fixing problems as they occur.
Can confirm there are 50 dots there
I haven't played with a blatant cheater in 2 years
It's beacause you're to stupid to notice them.
The comments on the original video were less toxic and more understanding about the explanations on the anticheat... Almost like if they actually took the time to listen and understand the video instead of just rage commenting because they got a cheater in there game
No mention of it being a reupload, eventhough it certainly is
what a time to be alive
Gaben learning: Gaben final days is to turn his brain into a AI to run valve from the grave.
Dota3 = more_hats(Dota2)
Wishing you a blessed future 🙏🙏🙏
Nothing has changed really even with all those machine learning models valve games still have cheaters in CS GO and recently a year had infamous TF2 botting incident which plagued that game for months so this model clearly didn't work out like intended
It wasn't used for TF2
Doesn't look like they are using it. Atleast not in Dota. In the "cheaters will never be welcome" blogpost they stated they used a honeypot. A frustrated playerbase apparently makes more purchases.
@@UmbraAtrox_ marketing PR move
Their honey post was actually checking whether cheaters access memory outside their camera range which is a decade old camera hack trick meaning they haven't done anything against modern cheats and there are more cheaters than 40k based on dota 2 cheating forum members
That is because it was never turn on yet. CS2 is the first time valve let vacnet not only detect a cheater but do a self ban on it.
This sounds like you're a low trust account mate
"It works pretty well" hmmmmm
Maybe it works well enough given the HUGE, TITANIC amount of cheaters this game has. Maybe we would encounter cheaters in every match if it wasn't because of this.
Comparing to other anti-cheat program, it did a pretty good job
now did they actually fix it because that game used to be unplayable do to cheaters
Is this not a reupload?
That's what I'm wondering.
2018
Nice.
58:36, well, I wonder what OS Sandboxes will bring...
this is so cool
Grrrr, my stomach just growled. For some reason I feel like having a few hamburgers delivered after watching this... it's risky these days, but cravings are cravings. Jokes aside, I have come to despise cheaters in games over many decades and this video goes a long way to assure me as a paying customer, who loves competitive gaming, that Valve goes more than the extra mile to ensure their games stay fair, fun and competitive. Very informative!
everything works as intented
good morning
Dough, pasta roller/ cutter. there, non-meat version of the metaphor
John McDonald IS PURELY AMAZING. LET'S GO AI!
12:00
Interesting how such an impacted company and a well waited game (CS2)
has its last comments only 4 months ago,
when the trailer was released 2 months ago
Just played couple of matches a week ago and I don’t think it works. There was at least 3 cheaters every match.
Yeah
This was recorded before 2019, perhaps things have changed since then
meh cheaters are big noobs who ruin the games for the rest of us
Or perhaps you're low trust factor.
@@clonkex That was my first ever matches, so new players have really low trust factor?
Create a bots only league.
Played like 10 games and in 4 of them I had cheaters. 2 out of 4 games were against completely blatant cheaters.
That says alot about your trust factor :p
@@vishnuprasanth4725 Maybe he isn't prime yet
Trust's working!
Quit cheating
I haven't seen a blatant cheater in years. That's how it feels to be in low trust factor... You did this to yourself so I feel no sympathy...
The problem I see from the initial explanation of deep learning is that it is based on statistics which always has results outside of the trust value. So what happens when instead of y=f(x) the problem to solve is actually y=g(x)?
So e.g. I have adhd with persistence into adulthood, so where the average brain solves y as a function f of x mine always solves y as a function g of x. So what happens e.g. when a behaviour differs from the statistical relevancy in the negative? As e.g. i would usually by the default way of my brain working take longer to get to the point of aiming and then clicking as it gets as much cpu of my brain devoted to it as every other problem and information coming in to my head at the same time, while simultaneously being for the same modus operandi a lot faster in detecting any changes that may give me information on what's going on around me.
So for the bot running on a deep learning solution my active combat could deviate as much as an aimbot but into the other direction of the significant interval while my reaction might deviate into that of information gathering cheating like wall hacks.
Interesting, although it's been two years so maybe things have changed a lot.
Seems like you won't catch players that are cheating by say 10% - maybe if that 10% is outside of other areas?
Also seems like as a cheater I could make my own model I develop from modeling a pro esports player et al, and
operate my cheats under the constraints of the model.
This way I shouldn't go outside of the range and thus not trigger the outliers.
Anyhow I hate FPS cheats as anyone. Sucks getting headshoted from across the map, etc.
So you're suggesting that cheaters could develop (or have developed, as this was in 2018) cheats that defeat VacNet by emulating human-like behavior? I think this is where VacNet comes into play not as the only defense against cheaters, but rather as a piece of a larger system to mitigate cheating. Theoretically a cheater could defeat VacNet, but still be flagged by trust factor, or get hit with a vac ban that detects them in a different manner. Valve could build a dataset of users that are both confirmed cheaters and labeled as innocent by the overwatch system. They could then use these users games to train a model that can detect subtle cheating. I would imagine that even cheats modeled after a specific human player have certain tells that a deep learning system could theoretically pick up on given enough data.
CS2 release around the corner and premium cheat providers are already advertising their new products lol This video is a such a joke.. There are videos out there, cheaters are wrecking havoc in CS2 Premier owning professional teams etc. Absolutely incredible how Valve is oblivious to this problem.
You clearly do not understand the content of this video at all.
This was never about straight up banning people using deep learning, that was stated multiple times VERY clearly.
Your inability to listen isn't their fault
@@jebbyy32 this is an upload from 3 years ago. You're right, this was never about banning anybody. if Valve was serious about solving cheater problem, they'd develop VAC that works - and actually bans cheaters on the spot. Cheaters are using loaders on external drives, infesting the game more than ever, with absolute impunity. Btw, STFU Hogarth
Man they used such brute force for training it. I wonder if they are still doing like this today.
ML has improved a lot in the years since so I'm sure they've improved their processes too
training itself is a "brute force" process
Sucks if you're a legitimate player that ends up facing cheaters all the time because algorithm even though you paid the same amount as everyone else for the same experience (esp as there is no single player)
He clearly explained how unlikely this is
@@jebbyy32 Cheats have evolved to mimic human play, it would be difficult to completely rule out all the edge cases, especially for deep learning.
Overwatch human more human error lol cause of omitted voice chat
This video is soooo old and cheating only got worse. Makes me lose all hope for cs2.
The games played in the lowest trust tier should have steam put their own bots in there (bots that cheat on server level) just to fuck with the cheaters.
Ahhh counterstrike ahhh deep ahhh learning ehhh. Heres ahhh how ahhh to use ahhhh deep ahhh learning ahhh to stop ahhh cheating ahhhh in counterstrike ahhhh.
Nope.. I cant do it.. im so interested in this subject but the speaker is a nightmare :S
I was looking for this comment. Its driving me crazy
Hahahahahahahahahahahh, this guy use ahhhh as a space bar for talking 😂😂
Interesting topic but this analogy is in the way instead of helping. And none was needed as it is a developers conference so a talk about data would have been fine.
Are developers supposed to magically know about deep learning?
@@Linkale_ No but they're supposed to know what a function is and they can understand if you say that Deep Learning is a method to create a function between input and output.
That's all that is needed for this talk.
Isn't GDC also attended by people who have more of a game design background?
I like how Ropz's team is playing cs2 premier and getting ruined by cheaters on their stream. So much for AI deep learning. Deep for sure - in the same cesspool of shit that VAC comes from as well.
Haven't you noticed the cheating pandemic in CS2 only started after wave 3? What if this was on purpose? I believe Valve will release CS2 with a massive ban wave.
@@Leonardo_D1 xd
Crap crap crap.
This definitely give away too much details.
All cheats have to do is to be as subtle as possible, hence jury is not convinced beyond reasonable doubt.
This was kinda the case with how overwatch has always worked though. It's been a thing since overwatches inception
7:57 - How can graph like this even make it to presentations like this. 30% is 3 of 10 not 7 of 10 and 40% is 4 of 10 not 6 of 10. Why are you doing this to yourself man?
good eye. probably just a mistake after a few revisions
@@aug_st I noticed it too because 6 out of 10 was actually worse than 1 out of 2
Wait why don't you use this prejudging technology that forces innocent players to play with cheaters based on "interactions with Steam" to solve crime? Million dollar idea right there..... Wait where have I heard of using stereotyping in crime before... Let me think.
But if it is a stereotype, then those players shouldn't worry anyway, they're not playing against potential cheaters it's just a stereotype.
They're not going to jail, it's just a game
@@Linkale_ "It's just a game" didn't work for racists and trolls when they get banned but its okay for a developer to use that excuse eh? Valve fanboys, what won't you excuse? "Interactions with steam" is so broad a statement it could mean using cheatengine on a singleplayer game to fool around is enough to lower your trust. Would you like to know more?
Valve fanboys: China social credit bad. Trust score good. "...based on how you interact with Steam". You may continue your groveling now.
having a bad trust score means you might get locked out of multiplayer in some video games at most
having a bad social credit score means your life can get fucking ruined beyond repair, to the point you literally can't even leave the country and make your life somewhere else
this is a terrible comparison
@@papermartin879 Your right. At least with China's credit system there is actual proof (as is needed when you prosecute a crime; unpaid bills, outstanding fees, footage of parking like a clown). Valve Trust system is based on no REAL evidence except prejudging you (the only criteria he gave was previous VAC bans on other accounts which in itself raises more questions than answers for even the mighty Google can be tricked into believing a single-user can be many real users). I guess all you have to do is give similiar concepts different names and it's enough to confuse everyone. Pro tip: question more.
@@maxpayne2745 yeah being vac banned is literally oppression
@@maxpayne2745 I dont think chinas social credit system uses a jury system to determine bad actors
Red trust factor "More likely to cheat" so lets put them into lobbies full of cheaters so we push them to cheat or quit. Man these dudes are delusional. Surprised the game hasn't died yet.
you didn't understand a single thing did you... You're more likely to have higher trust factor than low so there are really no reported cases at all of that happening. If you are a non cheater in low trust matches, perhaps for being toxic, then that's your fault and you did something to deserve to be there... Try actually watching the video because he already explained away your entire argument
@@CherryZ So being a little toxic means I deserve to go against cheaters more? Imagine being that fucked in the head.
@@OMGiTzSpAnKy sounds like you're more than just "a little toxic" if you're going up against cheaters 😂. The system works fine for everyone else I know. I'm happy to see toxic players being put into their own lobbies and not ruining my games! Just learn to chill and you'll earn your way back into our lobbies before the next new year!
@@CherryZ Oh it works fine for my couple of friends so therefore it's not broken. Literally retard logic.
@@OMGiTzSpAnKy from the toxic and angry way you're speaking to me looks like you really do belong in low trust factor :D. So no worries, the systems working as intended! Gotta keep kids like you out of our lobbies 😄 but just so you know, whining about it in the comments section wont bring your trust factor up, once you learn to calm down and speak appropriately then you'll start to see your trust score elevate. Until then, enjoy low trust! You earned it!!!