AI concept of the '80ies ... why?

[mivayan]

Uh-Nu-Buh said:
In any case, Total Annihilations had a very interesting AI MOD model. It was a text file of units and weights, different for each race, and different for different types of maps, and completely changeable by the player.

It's a long step from a fully customizable AI, but could still mean great improvements for dominions.

Semi-related: is there a way to set a 'difficult' AI when changing from human to AI, for example to set up a single player challenge? Would be potentially bad for multiplay though.

[Wick]

Once upon a time I was playing Wesnoth as the Undead. My leading troops were ghosts with a chill attack. After driving the Imperials back most of the map I noticed he had started recruiting mermen. I thought that was odd since Mermen aren't that good on land and there wasn't much water then I realized they were strongly resistant to cold damage. The AI was paying attention to what I was doing! I felt so validated as a person! ;-) I also thought he should be recruiting mages that could actually hurt my ghosts instead. :-(

[DominionsFan]

I agree with Arralen. In fact imo we will see very good AIs in a couple of years, when most of the players will have a decent PC to run the AI processes fast enough.
Right now, many people have some low end machines still, running complicated AI processes on those would take enormous amount of time. My brother is playing a football manager game [his kids are also loving it, but they are too young to play, so mainy they are only watching their father as he is playing.. ], and in that game, he can choose what leagues to play in. When he runs 60+ leagues for example, even on a dual proc. high end pc, it takes 10-20 mins to process 1 weekend "turn" in the game. I guess the present day comps are still not ready to run these huge processes.

[Folket]

In general there are three kinds of AIs, neural networks, smells and scent, and Brute Froce.

Anyone played Fields of Battle? That AI is quite powerful and selflearning using Neural networks.

Problems with Dominions is that a Game would have to be played til the very end before the Neural Network could be updated. After programming the Neural network, which will be a pain, you will have to play plenty of game before the neural network would give a good AI. If the programming was powerful enough. In general this would probably be the optimal AI for dominions as it would learn from human players. But the amount of cells and how to program them seems to burdensome.

Anyone played Galactic Civilization or any modern chess games? Those games give good AIs given brute force.

Brute force is about calculating as many moves as possible to find an optimal move. Given the amount of possible moves in dominions, the computer would never be able to calculate far enough to separate move from eachother in terms on how good they are.

Smells and Scent is the easiest AI to implement. You just give the AI some rules to follow and let it react on it's surroundings. I think this is the only way to go with Dominions, so all you can do is give better and more precise rules for the AI to follow.

[Folket]

Well, a comment on Brute Force. Brute Force is effective if you have a good algorithm to calculate strengh of positions.

In other words, How strong is my position now? How strong will it be if I recruit one mage?

Perhaps smells and scent should be scent and shouts.

[Uh-Nu-Buh]

Most chess games have a good opening library, position library for mid game, and position library for end game manuevers. They also use brute force, but without the opening library etc. they are garbage....

IOW, without a good starting rulebase (e.g. 10 preapproved good pretenders for each nation, etc.) then Dom III would still have problems.

[Gandalf Parker]

I think that AI switches affect only the starting points for creating the game. The impossible AI gets more points to create its scales and god than you do. So Im not sure that any switches would affect a human that is turned AI. Its basically already in a better position than the AI anyway (smarter creation and better start position than if an AI had gotten that far)

[Evil Dave]

Uh-Nu-Buh said:
IOW, without a good starting rulebase (e.g. 10 preapproved good pretenders for each nation, etc.) then Dom III would still have problems.

I think it's possible to solve the problem of the AI designing good pretenders, or just not-obviously-stupid ones. Certainly, an "openings book" of ten pretenders per nation would do that. It may be enough to write an AI that makes pretenders based on common strategies (blessing, etc) to provide much more of a challenge for SP.

[Arralen]

mivayan said:
Semi-related: is there a way to set a 'difficult' AI when changing from human to AI, for example to set up a single player challenge? Would be potentially bad for multiplay though.

You might try to re-start the server with one of the following command line switches:
--easyai X Nation ai controlled X=nation number (0-79)
--normai X Nation ai controlled X=nation number (0-79)
--diffai X Nation ai controlled X=nation number (0-79)
--mightyai X Nation ai controlled X=nation number (0-79)
--impai X Nation ai controlled X=nation number (0-79)

[Endoperez]

Arralen said:

You might try to re-start the server with one of the following command line switches:
--easyai X Nation ai controlled X=nation number (0-79)
--normai X Nation ai controlled X=nation number (0-79)
--diffai X Nation ai controlled X=nation number (0-79)
--mightyai X Nation ai controlled X=nation number (0-79)
--impai X Nation ai controlled X=nation number (0-79)

I thought those only affected game creation, but I guess it would be worth a try...