#Programming a RISK AI

1 messages · Page 1 of 1 (latest)

opal plinth
#

Hey guys, as a programmer i'd like to try programming some AI for RISK. It must be some kind of graph problem. Is there topics about it that you now ? maybe devs talks ?

vocal rune
#

Lmao, dev talks on how the ai works 😂 We got our first q&a a week ago and the most depth they went into was answering my question on when the dice become the estimation technique and not pure calculatino in BB

#

I know of no resources on this topic when I tried to get into this. I did do some advancments on the topic for endgame 1v1 wd theory but it was not ai, it was simulating the game and running a program with different behaviours to find the best ev

#

That was not an actual ai, deciding what to do though

#

I am very interested in this topic though so if you embark in the journey on this I would love to be by your side to help how I can with game knowledge and my limited python programming skills

opal plinth
#

I know some AI techniques i'd like to try here. On the first thought, I'll try to implement a neural network evolving through genetic techniques (some kind of NEAT algorithm). I'll dig into it in the next days, if I find the motivation to start coding x)

vocal rune
#

Probably would give the best final result but difficult to put into place. If you get to it, lemme know

near topaz
#

Very interesting. I was thinking yesterday about the ai and how to evaluate best play in a situation

tame pumice
tame pumice
midnight haven
#

For a uni project I worked on, I made a collection of references for all of the 40+ publicly available pieces of academic literature on Risk.

Around half were attempts to make a good Risk bot.

I could send you a copy if you like?

opal plinth
vocal rune
midnight haven
real tulip
#

This is not a worthwhile nor wise endeavor. Generalizing AI after teaching it to effectively wage war means the generalized AI will have a huge head start.

opal plinth
# real tulip This is not a worthwhile nor wise endeavor. Generalizing AI after teaching it t...

I don't get your point
You mean the specialized would have a huge head start right ?
Anyway, if we are talking about behavior and strategies, there is out there multiple method to counter specialization, or at least to make the general IA being able to respond.
But the point about having a learning process is not to make the most efficient AI. Game analysis approaches will be better at start anyway, as you say. The point is much more in watching how the learning process makes the IA evolve, and what it considers be the game priority. Some may capture the most possible to maximize income next round, while other may stack up and focus on not getting captured back anywhere ^^

lavish hemlock
#

Ya coding it would be an interesting problem , there are lot of situations like going for territories is optimal for fixed card gameplay whereas holding a couple of guard positions and focusing on card more is better for progressive card gameplay

real tulip
#

my point is that any AI or AI research that is discoverable as a service or code will be assimilated by generalized AI.

near topaz
#

skynet taking over?

opal plinth
worthy ridge
# opal plinth I'd love to ! There must anyway be many techniques, and as its more in my knowle...

I think it would be very hard to "code intelligence into it"
But a self-learning thing like Alpha Go would be very very interesting imo.

They chose Go and Chess (and Poker) because they are more popular.
Those are 1v1, but multiplayer games are way harder to solve.
I also think the next step in research is to move to multiplayer

The graph theory also becomes trivial if you have self-learning. You just give it a list of what's connected and what's a bonus, and it will figure out the rest "automatically"

lavish hemlock
#

There is also AI bot for dota