An expert system ruleset for the computer AI

[Keybounce]

I have a simple request. (Might not be a simple result, but ...)

For those of you that are really, really good at smithore manipulation, can you provide some rules (a la an expert system) for how you do it, so that an AI can be programmed that knows how to challenge you?

Basically, rather than one person trying to spend a few hundred hours programming an AI for mule from scratch, if we work together to develop a good set of rules for it to follow, we can make a challenging opponent much faster.

So, questions like:
1. How do you evaluate land values
2. How do you manipulate smithore?
3. When do you abandon smithore and go for crystite?
Etc?

[dynadan]

Ok i have given this quite a bit of thought and since no one else has weighed in, I will do my best to give some suggestions on how to fix the current AI.

I think that how the AI currently handles smithore is pretty good.  It always seems to do the right thing for the 1st ore shortage of the game.  The few changes i would make is on round 8 or after i would just set the AI to horde smithore and not to sell for less than $100.  Also, at the end of a round if the AI has more than 30 ore at round 8 or higher it should start trying to switch its ore plots to crystite.  Also, any time the supply of ore in the store is 100 or greater the AI should switch its ore plots to stite.

I think the main reason everyone hates the current AI is how it handles food and energy.  I believe the solution is to give up its food and energy strategy.  I have only witnessed 1 person ever win with a food and energy stategy and it was one of the best players on planet mule (I am looking at you Brad1867)  My solution is simple but should be effective.  Using the same rules as the current AI uses, make the Maximum food plots the AI can have be 2.  And make the maximum amount of energy plots be 3.  This should not only help the AI improve its game, but will prevent it from producing everyone elses food and energy.

Land valuation is a very hard question to answer.  I don't think we have to develop an equation that lets the AI accurately value land.  Instead I think the trick is to come up with general valuations so that the AI doesn't let someone buy land for far below market value (like it does now). 

1.  The AI should always leave itself enough money to buy 1 mule plus $100.  (i know that it will have 2 empty plots if it is successful but i think that's ok)
2.  The maximum the AI should pay for land in round 1-3 should be $800
3.  The maximum the AI should pay for land in 4-5 should be $1750
4.  The maximum the AI should pay for land in 6-8 should be $2250
5.  The maximum the AI should pay for land in 9-12 should be $1500
6.  The maximum the AI should pay for a river plot is $1150

I think with these changes the AI would start playing fairly competitively.  If anyone has a different viewpoint or different valuations, or have any questions about why or how i came up with any of my numbers i will be happy to discuss or debate in this thread.  Again, this won't make the AI play better than the better human players, but should allow the real players to play without being able to manipulate the mistakes the AI makes now, and occasionally give the AI a decent chance to win.

[Null]

it seems like AI should, as probably majority of the winning player do, consider how much money it needs to plot a land before buying the land.

AI Money = $
Next round mule cost +25~100 = &
Land Price = %
Money used durring land auction by AI = (&-$) < %

Players sometimes buy land even if they cant afford to put something on it the next round.  It seems AI should also make similar actions.  Especially if it sees a potentual to put something on it the next few rounds.