ASL with AI?

[samtyson]

I have created an psuedo-AI version of my scenario, Saving Ryan's Privates. It has logic to control the actions of the American attackers and the German defenders, and is built for entertainment purposes.

I would love to see how the win/loss/draw numbers stack up over a huge number of iterations. Right now it is looking slightly pro-American, but will time make a difference?

Feel free to check it out, play through some games, and see what you think. If you notice any glaring rules errors, let me know.

SRP Replay

www.texas-asl.com

Just load the page, click the Next Phase button, and follow the action in the results log. The movement phase flows quickly sometimes due to the flow of javascript, but it is understandable after a few plays.

Have fun with it!

 

[Yannick]

That's very cool. What's the "AI" here? Some random or scripted actions?

 

[witchbottles]

@samtyson Glaring rule error:

AI failed to rout the broken german 2-4-7 in the RtPh.

 

[samtyson]

@samtyson Glaring rule error:

AI failed to rout the broken german 2-4-7 in the RtPh.

View attachment 18256

The game ends when the German is broken, so no need to rout.

 

[samtyson]

That's very cool. What's the "AI" here? Some random or scripted actions?

Mostly random (DR based of course) actions, usually weighted for common sense.

 

[Sparafucil3]

Does it base it's reactions on learning or is it always random? -- jim

 

[BigAl737]

Guys like you Sam, who can code something like this have my utmost respect. Why it’s fun to keep pressing the blue button is beyond me but it is. Sort of like playing tic tac toe.

Not an error really but on my iPad, the counters are shifted like you see in the screenshot.

 

[von Marwitz]

Does it base it's reactions on learning or is it always random? -- jim

I asked myself the same thing...

von Marwitz

 

[Actionjick]

I asked myself the same thing...

von Marwitz

Are there AIs out there for gaming that can learn? That would be fascinating. I'm also intrigued with programming an AI with a specific player's style of play. Imagine being able to play against an AI programmed to be Fort, McGrath, Cheney or Pleva!!

 

[von Marwitz]

Are there AIs out there for gaming that can learn? That would be fascinating. I'm also intrigued with programming an AI with a specific player's style of play. Imagine being able to play against an AI programmed to be Fort, McGrath, Cheney or Pleva!!

There sure are. The most known are for Chess and Go. The uncanny thing is that the some of the more recent AIs for these managed to beat the best human players within a matter of less than a month starting off not even knowing the rules at all.

But there are also some for some computergames with more complicated rules.

von Marwitz

 

[Actionjick]

There sure are. The most known are for Chess and Go. The uncanny thing is that the some of the more recent AIs for these managed to beat the best human players within a matter of less than a month starting off not even knowing the rules at all.

But there are also some for some computergames with more complicated rules.

von Marwitz

Thanks!

 

[Gordon]

https://twitter.com/i/web/status/1400959984972689409

And also:

 

[Philippe D.]

There sure are. The most known are for Chess and Go. The uncanny thing is that the some of the more recent AIs for these managed to beat the best human players within a matter of less than a month starting off not even knowing the rules at all.

But there are also some for some computergames with more complicated rules.

All you need to start a self-teaching AI running (at least for the very naive learning algorithms) is a clear-cut implementation of the rules of the game, and lots of computer power. The real trouble with games like ASL is the first part - turning the written rulebook into a set of rules unambiguous enough for a computer implementation seems like a huge task.

Also, I don't know how these algorithms deal with explicit randomness in games (though it's most likely something that's been studied). In Chess or Go, whenever a player makes a decision, the result is perfectly defined in advance (for the next player to make a decision). In ASL, when a decision is made by a player, there may be an arbitrarily long chain of consequences that don't involve any choice by either player (long chains will typically require repeated Sniper attacks, so they are pretty unlikely; but even listing all the possible results of an action would be hard).

 

[von Marwitz]

Also, I don't know how these algorithms deal with explicit randomness in games (though it's most likely something that's been studied).

Probably, the algorithms will deal with 'chances' just as we players do. Or they 'recalculate'/adapt in case of freak events.

Maybe self-learning algorithms have developed to tackle ASL when most of the current players are getting too old to play.

von Marwitz

 

[PabloGS]

Really nice!

The self-learning would be adjusting the probabilities of doing certain actions depending on the win/loss record of previous games.

 

[samtyson]

This was a huge endeavor that I thought would be pretty straight-forward and perhaps simple. This scenario is so small and simple, but there is so much to think about and account for it is mind-boggling. There are 4 unit counters, an HMG, 8 hexes, etc.

 

[Actionjick]

This was a huge endeavor that I thought would be pretty straight-forward and perhaps simple. This scenario is so small and simple, but there is so much to think about and account for it is mind-boggling. There are 4 unit counters, an HMG, 8 hexes, etc.

The number of variables in even a small scenario must be staggering. Good luck with this project and future endeavors.

 

[witchbottles]

The number of variables in even a small scenario must be staggering. Good luck with this project and future endeavors.

just from the dozen or so times I ran the program myself - I can attest that even in such a minute setting, no single game has had the exact same results over the 12 permutations I created in running it.

From battlehardening, to hero creation, to wounding the leader, to watching the HMG go on a rate tear and never expend all its ROF until the DFPh, winning the game in the process in the DFPh for the Germans, to routing, to the CCPh permutations therein, there is a ton of possibilities.

 

[witchbottles]

The next test is to see how Sam decides to balance the scenario, (IMO the "draw" result is the thing I believe is tipping the balance of this scenario).

 

[Tuomo]

Pleva would still win it more than me :-(