[freeciv-ai] [baylor@xxxxxxxxxxxxxxx: Re: diplomatic state of the art?]

[Raimar Falke <mHy6X0e4Rmh2=a_&*a+3#@rf.risimo.net>]

To raise the general awareness for comp.ai.games and to contribute
something on-topic.

        Raimar

----- Forwarded message from baylor <baylor@xxxxxxxxxxxxxxx> -----

From: baylor@xxxxxxxxxxxxxxx (baylor)
Newsgroups: comp.ai.games
Subject: Re: diplomatic state of the art?
Date: 28 Jun 2003 13:56:31 -0700

"Brandon J. Van Every III Esq." wrote :
> What's the diplomatic state of the art for game AIs?
> it needs to be relevant to solving the freeform alliance problems

Mind if i ask what specific kinds of problems you're looking at?

Is this something like the making and breaking of alliances in games
like Civilization?

If so, i imagine you'd just need a couple of functions to determine
1)when to make an alliance, 2)when to break an alliance and 3)who to
make an alliance with

Are the rules for this sort of thing complex? i wouldn't think so.
Figure you could make it work with a simple costing function - every
action has a cost and benefit and when the benefit outweighs the cost,
do it

And how do you define cost and benefit? There's probably a couple of
consistent factors:
- If it's a zero-sum game ("there can be only one"), you make
alliances aganst the strongest player and break them when you're
allies are getting strong (the #3 and #6 team up against #1, but if #6
becomes #1, he's king of the hill and a threat to you)

- Only pick fights you can win unless you have a good reason not to.
Did your buddy who #6 become #1 while you're in last place? He'll
probably crush you so stay away from him unless you can ally with #6,
#5 and #4 and all of you combined equal #1. Unless #1 is about to win
the game (=instant death for you, worst possible cost), at which point
you might as well be suicidal and go for it

- When threatened, figure out the least costly option. Suppose the
Phoenicians demand 100gp. Their army, you figure, could wipe out 20 of
your troops which cost 500gp to build. So pay the tribute rather than
fight

- You make alliances or trades with people who have something you
want. You know pottery and alphabets, red team also knows pottery and
alphabets and green team doesn't know them but knows sailing and
astronomy. So you'd ally with green team because there's more to gain
(higher benefit)


Some games will choose making alliances, declaring war, bullying
people and all that based on the personality of the AI. Meaning that
if you define the Translucians as being a peaceful race, they won't
demand tribute or technology from you unless the cost/benefit analysis
works out so that benefit is greater than some personality-specific
threshhold (ex, benefit 125% greater than cost)

All of the above, of course, assumes a cost function that can work on
some pretty simple quantifiable stuff in the short term future (in
AI-speak, it heavily discounts future rewards). But that's how most
humans do it. In the US, my belief is that we always side with the UK
because we view them as parents, so even though France has come to our
aide more often than England (kinda like the greaser Harley-riding
boyfriend), we always pick on the French and backup the Brits (again,
off topic and just my opinion). Completely non-rational, so if humans
can be non-rational ("blood is thicker than water" is supposedly hard
coded, not learned), why not hard code the AI to be the same?

i guess the strategy game developers who work here would know more
about the little gotchas in this field (i've never really understood
how strategy games work), but i personally haven't heard of any
substantial work being done on this topic. Maybe some work in the AI
field on MAS (multi-agent systems) and dynamic group formation and
possibly combinatorial auctions (if you can't make alliances with
everyone in a group don't make alliances with any of them), but i'm
guessing there probably isn't too much there helpful

-baylor

----- End forwarded message -----