[h0dgep0dge]

I'm curious to know if anyone knows of any research into using machine learning to improve botting. In this age of human jobs being taken by machines, why not have them play our games too?

Very broadly I see two major applications. The first is obviously using machine learning to play the game itself, or supplement hard coded design. It could be useful for complex pattern recognition, among other things.

The second big application I see is using it to audit the "detectability" of bots. The other team is already using machine learning to detect cheating, so why not take a page from their book? If we're able to very accurately differentiate between real players and our own bots, it could allow much better research into anti-ban.

It's also worth noting that these two strategies could be combined into a scheme of what's called "adversarial" machine learning. This would entail training the bot program to create hard to detect programs, while simultaneously training a ban program to differentiate between bots created by the other program and real players.

Let me know what y'all think.

 

[party]

FYI it has already been applied in a couple of ways, though to a lesser extent than you're referencing. A number of bots (private, mostly) have basic machine learning for detecting PKers better, others are able to learn how to pray-flick in OSRS against a number of monsters (Slayer by @Qosmiof2, for example).

 

[parametric]

I'm curious about what kind of heuristics you would even monitor to differentiate between a bot and a player? You have to feed a net something.

Camera Changes Per Minute, Mouse Moves Per Minute, play duration, fatigue?

There is so much data you would need to normalize for a single case, then on top of that you would need a big enough data set from both real players and bots that even with a simple delta rule (If i remember correctly it's the one that can use a smaller data set?) it would seem like a lot.

My machine learning background is limited right now as I have to wait till the spring to have access to the higher level course at my uni.

 

[h0dgep0dge]

Do we not have access to a very large group of runescape players who openly use botting software? We'd need some way for users to easily record data about their session, anonymized of course, and provide metadata about if they were using a bot, but I don't think that's an insurmountable problem.

 

[red]

I'm curious about what kind of heuristics you would even monitor to differentiate between a bot and a player? You have to feed a net something.

Camera Changes Per Minute, Mouse Moves Per Minute, play duration, fatigue?

There is so much data you would need to normalize for a single case, then on top of that you would need a big enough data set from both real players and bots that even with a simple delta rule (If i remember correctly it's the one that can use a smaller data set?) it would seem like a lot.

My machine learning background is limited right now as I have to wait till the spring to have access to the higher level course at my uni.

Quite simple, just analyze user/bot input.
EDIT: not simple in action just the answer was simple.

 

[parametric]

Quite simple, just analyze user/bot input.
EDIT: not simple in action just the answer was simple.

You still need to find heuristics that jagex uses...

 

[red]

You still need to find heuristics that jagex uses...

you can use what the client sends to the server. mouse data, camera data etc.

 

[daniel]

you can use what the client sends to the server. mouse data, camera data etc.

While you're at it hack into the mainframe and make a SQL wallhack coded in VB.NET to DDoS the firewall and get the source code on the private IP address.

 

[staffix]

While you're at it, renew my premium pornhub subscription.

 

[party]

While you're at it hack into the mainframe and make a SQL wallhack coded in VB.NET to DDoS the firewall and get the source code on the private IP address.

While you're at it, renew my premium pornhub subscription.

Let's not derail the conversation, please.

 

[parametric]

you can use what the client sends to the server. mouse data, camera data etc.

Sure but to what extend is that stuff used, I know people who have botted 200m's in multiple skills using a hopping mouse; In rs3 at least. Shit back when I used OSRS processAction in my private bot I had 99's running 24/7 sending the mouse clicks to -1, -1 and because of the nature of processAction my camera never moved.

 

[red]

Sure but to what extend is that stuff used, I know people who have botted 200m's in multiple skills using a hopping mouse; In rs3 at least. Shit back when I used OSRS processAction in my private bot I had 99's running 24/7 sending the mouse clicks to -1, -1 and because of the nature of processAction my camera never moved.

They could have just checked if it was there rather than if it was valid. That's the reason that processAction was patched as soon as they found out, I was using that type of bot too just toward the end.
 

While you're at it hack into the mainframe and make a SQL wallhack coded in VB.NET to DDoS the firewall and get the source code on the private IP address.

Also what have I said that implies that I don't know at least a little about what I am talking about.

 

[aydex]

Sure but to what extend is that stuff used, I know people who have botted 200m's in multiple skills using a hopping mouse; In rs3 at least. Shit back when I used OSRS processAction in my private bot I had 99's running 24/7 sending the mouse clicks to -1, -1 and because of the nature of processAction my camera never moved.

They could have just checked if it was there rather than if it was valid. That's the reason that processAction was patched as soon as they found out, I was using that type of bot too just toward the end.
 

Also what have I said that implies that I don't know at least a little about what I am talking about.

rip inubot

 

[red]

rip inubot

Its still a thing on rs3 lmao

 

[parametric]

It's still doable in osrs but you have to jump through hoops to not be instabanned.