Recommend
 
 Thumb up
 Hide
8 Posts

New Angeles» Forums » General

Subject: How feasible would an AI player be? rss

Your Tags: Add tags
Popular Tags: [View All]
Brian Long
United States
Seattle
Washington
flag msg tools
Avatar
mbmbmbmbmb
I really want to play this game, but even getting 2P is hard for me, nevermind 4-6p. I've always wanted to create an AI player for a boardgame (I used do AI coding in the video game industry, so I don't think it's beyond my means), but there are two challenging problems here:

1) How much board state is there? For games with a lot of board state flying around (and that's /most/ games), it's too tedious to keep the AI abreast of the state of the board. And CV might be too hard (for me at least) to automate that.

2) How freeform is the negotiation part of the game? Does it ultimately end up in a concrete, discrete, quantitative proposal that has to be explicitly voted on? e.g. can the question at vote be put concisely to an AI opponent via mobile device interface?

I know an AI player would miss 99% of the nuance of the conversation, but I'm wondering if it's still just mechanically possible (even if not ideal) to involve one in a game to make decisions based on board state and the proposals put to it?
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Dave Schmidt
Canada
Strasbourg
Saskatchewan
flag msg tools
badge
Avatar
mbmbmbmbmb
I haven't played the game but I do think the AI could not replace a player very well. Pretty much everything can be negotiated and I don't think all negotiated things are binding either - murking the water a bit for an AI.

However, it might be an interesting variant to have at least 3 players and then one android player that is somehow indirectly involved in the game. Not sure how this would work exactly though. This would also be interesting given the setting of the game as advanced AIs exist in that universe.

I get the impression from reading/watching reviews of the game that the negotiation is essentially the game.
4 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Cameron McKenzie
United States
Atlanta
Georgia
flag msg tools
badge
Avatar
mbmbmbmbmb
For a lot of euro style games, it isn't too hard to get a decent functional AI going with machine learning algorithms. You could make an algorothm that's able to effectively value and prioritize different actions cards and to vote in its best interest in the absence of any negotiation.

But I think getting an AI to negotiate and bluff is really where the challenge would be. It would also be tough to train the AI to use many assets effectively especially since the way you use your assets is such a big negotiating point.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Brian Long
United States
Seattle
Washington
flag msg tools
Avatar
mbmbmbmbmb
Thanks. Really what I'm asking is how amenable the "inputs" and "outputs" are to feeding into a decision making system. Not whether the system can be "smart enough" to be competitive or not.

Large, non-discrete, or high change rate board state = hard to express inputs (every time someone does something the AI has to be told about it - tedious).

Outputs - decision-making - in board games TEND to be discrete, but games that allow for free-form wheeling and dealing (think, Cosmic Encounter) become impossible to express. Not having played New Angeles, it wasn't clear if the votes produced very discrete/finite outcomes (e.g. enact a card) versus "we are all now bound by this arbitrary thing we verbally agreed upon."

I definitely think New Angeles would not be the best game to pick to have an easy time making an AI for, but it's the first one I've really wanted to play that's motivated me to wonder about it. (Well, that and Rebellion, but I discounted Rebellion due to there being WAYYYY too much board state.)
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Kwijiboe
United States
flag msg tools
badge
Base. GAME.
Avatar
mbmbmbmbmb
If negotiations against Ghandi in Civiliation hasn't taught you anything...
7 
 Thumb up
0.02
 tip
 Hide
  • [+] Dice rolls
Brodie
United States
KC
Missouri
flag msg tools
mb
There are effects like "when a player proposes an offer, you can force them to tell you how they will resolve that offer if they win, and they are then bound to that declaration." Between that and the fairly open nature of wheeling and dealing, I wouldn't want to try to write an AI for New Angeles...
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Brian Long
United States
Seattle
Washington
flag msg tools
Avatar
mbmbmbmbmb
Yeah, sounds infeasible. Oh well!
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Baden Ronie
United Kingdom
flag msg tools
Avatar
mbmb
newobj wrote:
I really want to play this game, but even getting 2P is hard for me, nevermind 4-6p. I've always wanted to create an AI player for a boardgame (I used do AI coding in the video game industry, so I don't think it's beyond my means), but there are two challenging problems here:

1) How much board state is there? For games with a lot of board state flying around (and that's /most/ games), it's too tedious to keep the AI abreast of the state of the board. And CV might be too hard (for me at least) to automate that.

2) How freeform is the negotiation part of the game? Does it ultimately end up in a concrete, discrete, quantitative proposal that has to be explicitly voted on? e.g. can the question at vote be put concisely to an AI opponent via mobile device interface?

I know an AI player would miss 99% of the nuance of the conversation, but I'm wondering if it's still just mechanically possible (even if not ideal) to involve one in a game to make decisions based on board state and the proposals put to it?

There's a reasonable amount of boardstate, but nothing an AI couldn't handle.

Where you'll fall down is the negotiation. Almost everything can be used to create a deal, and those deals don't have to be upheld if anything in the future is promised. 95% of the game is arguing about why players should choose your deal over another. And I wouldn't say you could concisely and effectively put it to an A.I.

For example: There are four players. One (A) makes the main offer and another (B) makes a secondary offer, leaving two players to vote.

Player A tells player C that they'll give them two capital. Player C likes this even more because player A is their rival. Player B offers player D two asset cards and the promise of future support to use at least three action cards to support him. However, Player D has player B as his rival, and thus has to weigh up whether those two asset cards and a vague promise might be worth letting his rival gain a chunk of capital.

And that's not a very complicated example.

1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Front Page | Welcome | Contact | Privacy Policy | Terms of Service | Advertise | Support BGG | Feeds RSS
Geekdo, BoardGameGeek, the Geekdo logo, and the BoardGameGeek logo are trademarks of BoardGameGeek, LLC.