The Ultimate Anti-Ban???

[Taco Bell]

EDIT:

Originally, my plan was to create a way to record my play style, use machine learning to find underlying patterns, and create a script that uses them, resulting in a human-like performance.

However, as pointed out by a couple folks, this is pretty stupid. Mainly because it would require a lot of data, and also because the end result would be a script that may be only useful at doing one task.

Instead now I look to do some real research into creating an AI capable of playing and most importantly learning by itself. 

Ideally, I will get my project to a point where it can run multiple bots which compete with each other to find the best way to bot certain tasks or achieve certain milestones without being banned.

For this project, I will need countless accounts that have completed tutorial island. It seems only right to attempt to create an AI capable of solving tut island.

At the moment, I'm not entirely sure how this will work. But my approach will be to: 

1. setup a bot with the ability to do two things: walk to a square, interact with object/npc
2. have the bot randomly perform actions, and feed the results into a neural network
3. use the RL network to inform further game decisions

If I even get this to work, I expect the bot to look pretty much crazy, trying loads of stupid shit to beat tut island. I'm hoping though, after much time spent on tut island, it can begin to see patterns (maybe something like talking to NPCs is a good first step, or the concept of opening a door to go outside).

My biggest concern at the moment

1. figuring out to setup a neural network that makes sense for this program
2. deciding on a reward which the bot is aiming to achieve at each step.

I think 1. will be solved by much trial and error on my part.

And for 2. maybe getting to the next area, or getting a new prompt can be what the bot aims for.

Well no way better to getting started than trying to solve the first part of tut island, listening to that guy chat loads and getting outside the starting room. 

Edited June 2, 2018 by secretsally

[John Cena]

Good idea, let's see where it takes you.

[Vap0r]

I did something similar with tutorial island but only for mapping a walking path, Seemed to work very well and stopped a lot of bans. I have seen people do something similar for a pking bot. It should probably work but will take quiet some time to develop and even more to work out the possible bugs.

[lebotman]

Need 2 months worth of data.

 

It would work if you knew what you were doing though, gl. 

[Slickk]

Don't bans come from the client you use along with the character behavior? Like you can have the most human-like bot imaginable but if the osbot client is detected, you'd still get banned.

[lebotman]

1 minute ago, Slickk said:

Don't bans come from the client you use along with the character behavior? Like you can have the most human-like bot imaginable but if the osbot client is detected, you'd still get banned.

Nah, all they would know is that its 3rd party. 

[superdope]

This concept is beautifully complex. You are wanting to make a script that is as human as possible which means you want to code the human brain basically. Gonna take a very long time but i would give you all my money

[Impensus]

1 hour ago, superdope said:

This concept is beautifully complex. You are wanting to make a script that is as human as possible which means you want to code the human brain basically. Gonna take a very long time but i would give you all my money

You aren't attempting to code a brain at all. Your attempting to create an algorithm which can recognise and recreate trends in data to replicate human like behaviour. Far from creating a brain.

[packthebowlll]

1 hour ago, Impensus said:

You aren't attempting to code a brain at all. Your attempting to create an algorithm which can recognise and recreate trends in data to replicate human like behaviour. Far from creating a brain.

another client offers this already tho lol

[Alek]

http://lmgtfy.com/?q=runescape+machine+learning

Edit: Here's the TLDR on why I'm being a smartass:

-You didn't mention what machine learning library you would use and why
-You didn't really explain your prior knowledge into it, just kind of loosely used it as a buzz word
-You didn't do any research into prior attempts for botting in video games, yet alone RS. 

 

How do you plan to train it? What specifically are you doing differently than what has already been done before? How will it understand what the objective of the game is? How much are you training it, are you training the mouse movements as well? Are you planning to run this on the main game? If so, do you understand that it will probably take a decade to come up with an intelligent bot that can automate tasks? How do you plan on simulating the training?

Edit 2:
Reading over your post, it sounds like you're talking about a project that already exists. The problem is, is that the machine really isn't "learning". It's copying your gameplay and trying to replicate it using algorithms. I suppose one could argue that its learning, but your bot isn't going to do anything new. You're going to train it to do something, and it will always do it the way you want it.

This is different than saying "go learn to mine", and the bot ends up figuring out that better pickaxes can mine faster, different ores have different rates, different locations have better variety, avoid high traffic areas, etc. Who knows, maybe your bot figures out that pking rune miners is the most efficient way to go mining.

[Impensus]

47 minutes ago, packthebowlll said:

another client offers this already tho lol

Very loosely and far from machine learning 

[Taco Bell]

55 minutes ago, Alek said:

http://lmgtfy.com/?q=runescape+machine+learning

Edit: Here's the TLDR on why I'm being a smartass:

-You didn't mention what machine learning library you would use and why
-You didn't really explain your prior knowledge into it, just kind of loosely used it as a buzz word
-You didn't do any research into prior attempts for botting in video games, yet alone RS. 

 

How do you plan to train it? What specifically are you doing differently than what has already been done before? How will it understand what the objective of the game is? How much are you training it, are you training the mouse movements as well? Are you planning to run this on the main game? If so, do you understand that it will probably take a decade to come up with an intelligent bot that can automate tasks? How do you plan on simulating the training?

Edit 2:
Reading over your post, it sounds like you're talking about a project that already exists. The problem is, is that the machine really isn't "learning". It's copying your gameplay and trying to replicate it using algorithms. I suppose one could argue that its learning, but your bot isn't going to do anything new. You're going to train it to do something, and it will always do it the way you want it.

This is different than saying "go learn to mine", and the bot ends up figuring out that better pickaxes can mine faster, different ores have different rates, different locations have better variety, avoid high traffic areas, etc. Who knows, maybe your bot figures out that pking rune miners is the most efficient way to go mining.

This project is very much an intro into practical machine learning for me. I regularly read white papers on the topic but have yet to do anything that wasn't given in a tutorial. 

I've done some research into people who have botted video games. Mainly the Atari stuff, but my approach is very different. It seemed they use goals, and objectives to determine whether the input was good or bad, and then iterate from that new piece of knowledge. 

Mine will be completely different. Your edit 2 touched explained it pretty well. I'm effectively trying to teach an algorithm to predict what I would have done next if I was there. Indeed, it will not learn nothing new, I'm just curious to see whether creating a bot based from real human input outperforms a standard script.

So my bot won't have artificial intelligence, but I will need to use machine learning to create my sequence prediction algorithm.

My understanding at the moment is that this will use a Recurrent Neural Network as those seem to be good in predicting sequences.

For now, I've setup a hypothesis I want to test in the original post. 

I've attempted to condense the number of inputs to a small amount as possible, which I think could be as simple as: location, my attack status, nearby chickens (their location in relation to mine, and there attack status) and time since last action. I hope this can reduce the amount of data points I need to collect to make a functional clone of my basic actions.

There's a good chance it still fails due to lack of enough data. If that happens, I hope I can at least uncover enough about my play pattern that I could code a better traditional script. And being able to setup my own neural networks would have been worth it.

Maybe one day I can try to make something that actually implements AI. The data set for that put me off, but I'm trying to figure out ways to collect the data on a large scale. I'm hoping this start project will at least give me a fresh understanding of the problem my algorithm is actually trying to solve, and if other data could be used to train my model.

 

 

[john55]

Ridiculous. The sample data you'd need would be substantial. Something only Jagex would have access to.

[Taco Bell]

14 hours ago, john55 said:

Ridiculous. The sample data you'd need would be substantial. Something only Jagex would have access to.

I've been thinking the same thing since I posted this.

I think I need a new plan.

[Taco Bell]

Updated OP with a new plan.

Going to try to create AI that can learn simple game concepts from tut island.