25-09-2002 02:55									#1
	Cool Ideas that the AI can't handle Get a bigger avatar today!

Was reading in a lot of different threads here today and saw tons of great ideas that were shelved cos the AI couldn't handle them properly.

Clearly in the case of new units, city builds, and wonders, this is problematic, and something we've been wrestling with in Candle'Bre, but in the case of entirely new concepts (ie re-writing the trade section of the game), it strikes me that one possible solution is to in some way have the entire complex new system bypass the AI?

What I'm thinking specifically is - if it's too complex for the AI, how 'bout adding a Human_player and Computer_player flag to the codes? In this way, we humans can enjoy the benefits of a complex system, filled with lots of strategic choices, and the AI can make use of the existing, simplified system.

Of course, there will be some crossover (as in the case of the econo-system, where human players may wish to buy or sell goods outside the bounds of their own empire, and this could well involve the AI's) - in these cases tho, rather than have the AI's have to make what for them are tough decisions, we could simply abstract it for them....translate those tough decisions (via slic code, I'm assuming) into equavalents in their simpler system, and give them an appropriate gold bonus for making transactions....poof....from the player's perspecitve, it APPEARS that the AI is handling the complex system remarkably well, when the reality is that the AI is largely ignoring the more complex system inpreference for a simplified rules set that it CAN handle.

Just two cents from the new guy....

-=Vel=-

	25-09-2002 14:51									#2
	Support Apolyton, pre-order Civilization IV

The AI is always a big problem but it cant ever be unconsidered. As you said we cant make a feature to be a advantage for the human. I have been thinking in a whole good cost evaluation sytem to make the AI work with it. The major problem with that sytstem is that i will spend more time teaching the AI how to use then actually really implementing this feature wich i have already everything thoughted up in my mind already. So it shouldnt take more than a couple hours to do it once a have the goods chosen and distribute through the terrains.
I totally agree with you. If an AI doesnt support there is no good reason to implement it.

Last edited by Pedrunn on 25-09-2002 at 14:59

	25-09-2002 20:19									#3
	Lose 30 kilos (of popups)

Just curious...

Do you have a specifc AI/Human issue in mind that can be posted here to discuss if SLIC can accomplish the effect.

	25-09-2002 21:56									#4
	Help yourself to an AD-FREE life

Actually yes.
But i will finish the effect now. Shouldnt take long. I have all thought at up.
We can discuss the negotiation and AI after this.

	25-09-2002 22:45									#5
	Put an end to popups!

Hex, thinking in broad terms, something like this would be an example of what I had in mind.

Let's say you (as the human player) know that an AI empire has a source of Jade, and that Jade has been modded such that it provides a +1 happiness bonus to those who import it or have their own supply.

Assuming Pedrunn's goods system is in place (goods have a quantified value and use rate, with the quantity of the good dropping with each use/transaction). Now, this would present a HUGE problem in terms of trying to get the AI to handle the resource well. At either extreme (AI always sells when asked, or AI never sells when asked) detracts from the realism, and the number of factors to determine "sell, y/n?" might require 30-60 seconds of thought and consideration for a human player, but for the AI....tremendously difficult, if not impossible.

This is where the simplified rules set comes in.

Part of the problem is with the scarcity of the resource. The AI has no really good way of valuing the resource, and as such, no good way to "price it." when determining whether to sell, and for how much to sell.

We can get around this problem by simply saying that goods have finite values when in human hands, but when controlled by the AI, they are infinite (with two caveats). First caveat is that the AI gets the happiness benefit for itself *as a property of the good* even if he sells. That greatly simplifies the AI's decisions, cos now he can reap the benefit regardless of what he does.

Put a counter on the resource, equal to the amount that would be there if the resource was finite (and in human hands). When THIS counter reaches zero, the resource will still exist (still show up on the map), but the AI will no longer sell it to human players (prevents humans from exploiting the AI).....so from the human player's perspective, the AI is abiding by the same scarcity rules, but the process is simplified, and the AI is helped by means of getting a tangible benefit regardless of how it behaves.

As to selling or not, that can be pretty formulatic, based on tone of voice, weighted for past dealings, and duration of trade route requested (with finite resources, we'd almost have to have trade routes expire after some set duration, to prevent people from accidentally running themselves out of a given resource).

Also, Pedrunn, any thoughts on having semirenewable resorces? That's something else we've been talking up in Candle'Bre....seems like if you can have a counter counting down, you could have another counting up (ex: hardwoods might start with a value of 60. You use twenty points worth, reducing it to 40, but the resource regrows by 0.25 per turn)....if you don't use them and don't sell them, eventually it'll get back to its predefined maximum.....stuff like that?

-=Vel=-

	26-09-2002 00:19									#6
	Support Apolyton, pre-order Civilization IV

To a great extent, many of the SLICs are aimed at giving the AI help without the cost that the human player pays. A player may look at your example and cry - 'It's not fair, the AI gets all the benefit whether it trades or not, whereas, the human wouldn't get the happiness bonus if it did trade the good.'

The updater code is a similar example because it allows the AI to update units for free, but the human player has to pay for it. On the surface, this seems to be a huge advantage to the AI, but as you noted already, the human player can generate a ton of gold, plus the human player can efficiently use the ability of Rush Buying. Plus the human player (should) always have the advantage of long-term planning, something that no AI has really accomplished. So the AI has no idea/concept regarding the upgrade sequence, and thus cannot plan for it.

The ideal solution would be to have the AI pay for the upgrade until it runs out of gold, then upgrade all of the remaining units for free, but I'm no SLICer - I just benefit from their fine work...and this was how the file was created.

So even though the code is weighted toward the AI, it is a tradeoff that is worthwhile. Without the updater code, the human can usually buy his way to victory and the AI is hampered with a lot of old units - having the code forces the player to carefully manage his money, and fixes the vexing problem of an AI with a lot of obsolete units. A player who carefully manages his money and plans ahead can maintain the upgrade sequence (hence the challenge)...if not he can always edit the upgrade costs to suit his abilities.

The bottom line is that without AI bonuses in any game, the AI would not be able to compete. Players can always find the exploits.

I do think that you will find a lot of the flexability you may need for Candle'Bre via SLIC. It won't be easy, and the SLICers cannot do everything, but make a suggestion and they are always willing to give it a shot.

	26-09-2002 02:21									#7
	Support Apolyton, buy Call to Power 2

quote: Originally posted by Velociryx Assuming Pedrunn's goods system is in place (goods have a quantified value and use rate, with the quantity of the good dropping with each use/transaction). Now, this would present a HUGE problem in terms of trying to get the AI to handle the resource well. At either extreme (AI always sells when asked, or AI never sells when asked) detracts from the realism, and the number of factors to determine "sell, y/n?" might require 30-60 seconds of thought and consideration for a human player, but for the AI....tremendously difficult, if not impossible.

Actual it is a simple Yes or No decission that has here to be made. There are a lot of conditions to accept or to reject. For example is the regard between two players high enough, what are the effects, is this good better then the other good, and so on... Actual it needs just time to be worked out correctly, but I think it should be possible to work it out. I think you can teach a lot the Ai but not everything, because SLIC does not give you the ultimate power.

quote: Originally posted by Velociryx This is where the simplified rules set comes in.

Unfortunatly you can't do everthing or you can't think about everything therefore the rhules will be simplified, but the goal should be as complex as possible.

quote: Originally posted by Velociryx Part of the problem is with the scarcity of the resource. The AI has no really good way of valuing the resource, and as such, no good way to "price it." when determining whether to sell, and for how much to sell.

That's up to the coder to give the AI a good formula to price a good.

-Martin

	26-09-2002 06:55									#8
	Help yourself to an AD-FREE life

At the and is all a matter of writing a good formula with the right and correct use of variable like regard, scarcity, use of the bonus...

Unfortunetly as Hexagonia said we can just make the AI think at the present