Hi all,
As some of you already know, in real life, I am a Game AI researcher. That means I am an AI research who likes to apply his research on games. I mainly worked on StarCraft in the past, but that is not the subject today.
As a boop fan, it was clear to me that sooner or later, I will try to make an AI playing the game. And I did: it is integrated to a boop Android app I plan to release soon (without using the boop name and the illustrations, since they belong to the editor and are naturally copyrighted - but game mechanisms cannot be copyrighted). I will write another post in that forum when the app is ready.
I am a researcher, not an engineer, so my goal is NOT to make an AI as strong as possible, mastering completely the game (although it is always cool). No, my goal is to create new AI methods, to propose novelty, even if it is less efficient than existing methods. I know that Deep Reinforcement Learning would easily master the game completely, but again, that is not my objective.
I built an AI I believe to be fairly good. I win most of the time against it, but I think it has a decent level. I am currently writing a scientific article I plan to submit before 15 January to an international conference (Congress on Evolutionary Computation 2024), comparing my AI method to other ones of the same type. However, I believe that giving an idea of how good the AI is against humans would be nice, giving a rough estimation of its BGA ELO score.
I contacted BGA, and they allowed me to create an account to let this AI play boop for estimating its ELO, as long as it is clear that players are playing against an AI for scientific purpose. This account is named AI_boop (https://boardgamearena.com/player?id=95213950). Since I can't really make automatic games with that account, I will be behind the screen playing the move selected by the AI (including promotion selections), in real-time games against human players. Maybe you?
I need as many games as possible, against as many different players as possible, from any level. I have a bit less than one month to play these games "manually", so I don't expect to have thousands of games, but reaching 100 games would be very nice.
If you want to play against the AI, feel free to poke me!
FAQ:
As some of you already know, in real life, I am a Game AI researcher. That means I am an AI research who likes to apply his research on games. I mainly worked on StarCraft in the past, but that is not the subject today.
As a boop fan, it was clear to me that sooner or later, I will try to make an AI playing the game. And I did: it is integrated to a boop Android app I plan to release soon (without using the boop name and the illustrations, since they belong to the editor and are naturally copyrighted - but game mechanisms cannot be copyrighted). I will write another post in that forum when the app is ready.
I am a researcher, not an engineer, so my goal is NOT to make an AI as strong as possible, mastering completely the game (although it is always cool). No, my goal is to create new AI methods, to propose novelty, even if it is less efficient than existing methods. I know that Deep Reinforcement Learning would easily master the game completely, but again, that is not my objective.
I built an AI I believe to be fairly good. I win most of the time against it, but I think it has a decent level. I am currently writing a scientific article I plan to submit before 15 January to an international conference (Congress on Evolutionary Computation 2024), comparing my AI method to other ones of the same type. However, I believe that giving an idea of how good the AI is against humans would be nice, giving a rough estimation of its BGA ELO score.
I contacted BGA, and they allowed me to create an account to let this AI play boop for estimating its ELO, as long as it is clear that players are playing against an AI for scientific purpose. This account is named AI_boop (https://boardgamearena.com/player?id=95213950). Since I can't really make automatic games with that account, I will be behind the screen playing the move selected by the AI (including promotion selections), in real-time games against human players. Maybe you?
I need as many games as possible, against as many different players as possible, from any level. I have a bit less than one month to play these games "manually", so I don't expect to have thousands of games, but reaching 100 games would be very nice.
If you want to play against the AI, feel free to poke me!
FAQ:
- What kind of AI methods are under the hood?
I mixed Monte Carlo Tree Search with Combinatorial Optimization (my research field per se). People already tried to use Monte Carlo Tree Search for improving Combinatorial Optimization methods, but to the best of my knowledge, no one tried the other way, i.e., injecting Combinatorial Optimization methods to improve Monte Carlo Tree Search. That's what I did, and it works pretty decently. - Is the AI using Machine Learning in any way?
Nope. Even if we speak A LOT about Machine Learning for the last 10 years, AI is a wide research area and is not limited to Machine Learning. - What kind of games AI_boop will play?
Non-arena, non-tournament, simple games with ELO ranking only. And I will always create games, not joining ones. - Does playing against AI_boop impact my ELO score?
It will, the same way you play against any other (human) opponents on BGA in a competitive game. - How strong is AI_boop?
I don't know, that's what I want to find out! I am winning most of the time against it, but I know this is not very informative. I can say that its style is pretty aggressive. It can play original and interesting moves, but it can also play dumb moves sometimes. :p
I will probably make the AI playing again from tomorrow. Ping me if you want to have a shot!