Improving AI Voting on Border Control

Discussion in 'Civ - Ideas & Suggestions' started by Qwan, Feb 24, 2021.

  1. Qwan

    Qwan Warlord

    Joined:
    Mar 5, 2010
    Messages:
    144
    In all my games, for Border Control, all the AIs place one (free) vote for themselves. This is stupid for two reasons:

    1. They should realize everyone is doing this, and so should know that unless they have the least favor, they are wasting their free vote.
    2. I always win this by spending 10 favor for a 2nd vote.

    So I'm suggesting a better algorithm for the AI. This is crude, and I'm sure it can be refined or improved upon, but it's a starting point for something much better than the above.

    Step 1: AI estimates how much favor is worth winning Border Control. How many districts will that AI build in 30 turns? Count how many they have now, divide by the number of turns passed, multiply by 30. (This is probably low.). Assume that each new district would obtain 1 tile via Border Control. (This is probably high for AIs, but the way I play, it's low). Multiply Estimated Districts x 1 tile x cost of buying 1 tile. That's the gold value of Border Control. Divide by the gold value of favor. That's a crude estimate of how much favor winning Border Control is worth to that AI.

    Step 2: AI estimates how much winning it is worth to everyone else. Simple crude method: scale by number of cities, e.g. if another AI has twice as many cities, it's worth twice as much favor to them.

    Step 3: AI sees if it can win. Assume that everyone will vote as much as they can, up to how much favor winning is worth to them. If the AI is not #1, and not close behind, they should know they'll lose.

    Step 4: Any AI that knows it will lose should vote B, not A. That will make winning A necessarily more expensive. They may want to spend a little more than their free vote for B, to help ensure they are not the target of B. (Which, note, makes A even more expensive). ONLY if they suspect they'd be the target of B should they vote A, even if they won't win.

    Step 5: Obviously, the above has to be modified by how much the AI wants to vote on other proposals.

    ----
    The only explanation I can see for why the AI currently votes the way they do on Border Control, is that B would hurt them so much, that they'd rather see someone (else) win A than risk the chance of being the target of B. Even if so, if the AI isn't willing to spend any favor on A, and if they know they won't win if they don't, then they should vote either for a weak (say by score) non-neighbor or for the one with the least favor (who wins the tie-breaker), and hope to get a diplomatic victory point.
     

Share This Page