1

I'm looking for any introductory/accessible reading on AI that can play games which involve social intelligence.

Games like poker, where you might bait someone into overcommiting their hand or threaten them into not betting a lot when they should.

Or warewolf, where there is a group of villagers, and a warewolf. Every night the warewolf "eats" a villager, and in the morning all the villagers get together to kick out a person who they think is the wolf. Wolf wins if he's the last one alive, villagers win if they manage to kick out the wolf before everyone dies.

I know there's some online content on the game warewolf, but how does AI tackle such concepts in general? How are the rules represented, and how does the AI "persuade" and be persuaded?

nbro
  • 39,006
  • 12
  • 98
  • 176
Shiny_and_Chrome
  • 111
  • 2
  • You would probably have to design it. How do humans persuade each other? Chat messages? You probably don't want a whole GPT-2 in your game but maybe you can have some canned messages the AI can say. – user253751 Sep 29 '21 at 12:10
  • thank you for the reply. I have done a bit of reading and I guess what I'm looking for is an entry point into "arumentation theory". How agents can exchange information to make joint decisions that meet their collective goals, or how they decide if they'll believe argument. Is there a good entry point into this? – Shiny_and_Chrome Sep 30 '21 at 03:56
  • I think it's more of a game design question really. Video game AIs often cheat in order to create the desired level of difficulty. Or for easier difficulties, they kind of un-cheat; they cheat to make their position worse. – user253751 Sep 30 '21 at 08:30
  • Are you asking how AI players cooperate with human players, or how AI players cooperate with each other? – user253751 Sep 30 '21 at 08:30
  • In a typical video game, there would just be some system that decides which AI players co-operate, and that's that. If you're aiming to design an advanced game AI that is even able to co-operate with human players, then maybe the AI doesn't need to know whether other players are AI or human. – user253751 Sep 30 '21 at 08:31
  • I remember in Sid Meier's Alpha Centauri, you can make peace agreements with other AI players, basically just by asking. There is some kind of reputation system so they won't agree to peace if you just captured their city for example. AI players rarely initiate peace offers. – user253751 Sep 30 '21 at 08:32
  • I am asking about how AI cooperate and compete with both human and other AI. Specifically how the AI algorithm encodes the concepts of input from other agents as arguments which may or may not be true, and the AI has to make a estimate of the truth value of the statement. – Shiny_and_Chrome Sep 30 '21 at 11:22
  • Probably no different from any other input. A signal from another player that may or may not indicate their intention is no different from a pixel in a picture that may or may not mean it's a cat picture. – user253751 Sep 30 '21 at 13:17
  • Im really sorry for bumping old thread, but searching similar information. Can someone help me? – FillipSmith Apr 01 '22 at 09:19

0 Answers0