New AI

[ryn]

Because I am implementing a Wesnoth AI (to be done before the summer ends ) I want some feedback about what players would enjoy most.

[scott]

Everything! But I think abilities are the most important one. Also it should know how to use healing effectively.

[turin]

It not knowing about shrouded/fogged hexes is the most important to me, because it would really help a lot with certain scenarios I'm designing/have designed. Knowing about more abilities is a close second.

[Pythagoras]

It not knowing about shrouded/fogged hexes is the most important to me, because it would really help a lot with certain scenarios I'm designing/have designed. Knowing about more abilities is a close second.

This would be a serious advance in AI. I can't think of any games with "fog of war"s that has an AI that can get by without "cheating" in this way.

[MadMax]

I actually think the best advance in AI would be to have the game adapt to your style as you play. This adaptation should, however, not carry over between scenarios. That, and there should be a small random factor, so that the AI is not as predictable.

Being able to fare well without knowing what's under fog or shroud would be nice too, and so would knowledge of abilities.

[Elvish_Pillager]

Actually, in my opinion, what we need is an AI that understands that ultimately, the goal is to kill the enemy leader.

The most important bit would be worshipping chance to kill the enemy leader.

[Dave]

I actually think the best advance in AI would be to have the game adapt to your style as you play.

This advance can be distinguished from ryn's suggestions in that it's probably infeasible to actually implement.

That, and there should be a small random factor, so that the AI is not as predictable.

There already is a small random factor in the current AI. Give it the same situation and it will not always do the same thing.

David

[szopen]

It would be nice if it could deal with static, long-range units, like watch-towers as artillery

(ducks and runs)

[Gafgarion]

It would be nice if it could deal with static, long-range units, like watch-towers as artillery

(ducks and runs)

Yes, this would be an excellent feature for an AI in a Wesnoth Fork.

[ryn]

yes, it would, but I want to end up with something usable by the majority of players, so I'll concentrate on strength of play/proper use of abilities/good use of fog and shroud, and only then implement miscellaneous requests which don't won't help in multiplayer. My current list of priorities:
1. get something simple and working that will beat the sample_ai
2. teach it about more common abilities (poison, healing, cure) and improve playing strength by other means
3. (by this point it should beat the default AI, I hope) have an option to make it not use knowlege about fog/shroud
4. implement other requests.

Of course, this will take the majority of the summer and is not set in stone
If, for example, some scenario designer asks me for something, I may choose to implement it early. Making the AI not use info about foggy/shrouded hexes it shouldn't know is, for example, on the list because of Turin's post.

[kangounator]

it will be great if in a same game the IA "learn" : if it was losing ( for strategy reasons ) it don't make the same mistake in this game.

[ryn]

That's one thing that actually can be done, kangounator, but not until it can actually do what I wanted it to do in the first place.
edit: look up TD(lambda) if interested. This algorithm is applicable because my ideas are based on an evaluation function.

[Breeblebox]

I think the abilities are important. Having an allied archer range attack an assassin because it does more damage is just plain frustrating. I've seen whole groups of elves (The first scenario in HTTT for instance) essentially take themselves out of a fight by trying to shoot an assassin.

Poisoned.. poisoned.. poisoned.. poisoned.. d'oh!