Conceptual question on AIs

[Teiresias]

Hello everybody,

watching the progression of 0AD for some time now left a question open with me: How do the AI bots "fit into the big picture"?
For clarification, i see at least the following uses of a "computer player":

1. "Substitute" a human player e.g. when network is not available/too slow for multiplayer (=> aegis?)
2. "Assistants" - see discussion on "Player-Scriptable Micro Managers" (2011/12/17)
3. "Trainer": Allow newcomers to practice before joining multiplayer games against experienced players (=> jubot?)
4. "Specials": Special behavior designed to fit a custom scenario, e.g. single-player campains or replay of historical events (battle of Marathon comes to mind).
5. "Demos" Show features of gameplay alive, including a "getting started walk-through" and auto-testing (=> tutorialbot, testbot)
6. "Bots in their own right". Could include 'abuse' of 0AD as a "core war runtime with excellent GUI" (i recall during the high-days of AOK there was an AI tournament set up by heavengames.com ~approx. 10 years ago).

I haven't found a concept paper on this in the wiki, but i'm curious what the 0AD officials think of where the focus is - what are your "must haves/nice to haves/don't cares/don't want to see thats"?

AI development seems relatively quiet at the moment, so i thought it a good time to raise the question. As people were thinking of changing the implementation design (discussions on C++/genetic algorithms etc.), such clarification might help (=> discussion "Finalizing planning for part 1.0" 2013/02/11 - consensus of the associated log seems to implement on an on-need base).

Please pardon me if there is info on the web and i didn't find it...

[Lion.Kanzen]

we have some info in our Wiki

http://trac.wildfiregames.com/wiki/AIEngineAPI

is outdated but is better i can search for you.

[Teiresias]

Yes, I have seen the technical specification, my point was more about the purpose of the AIs (the "gameplay").

For example, in some elder threads there was a desire for a "real hard AI that does not cheat", which is probably category #1,#2. Let's imagine there was a perfect-playing(*) AI - what would one do with it?

Another issue is the "partial AI rewrite in C++" mentioned before. If the primary audience is multiplayer, a 100% hardcoded/scripted AI might be sufficient with respect to the performance troubles.

(* For tic-tac-toe, checkers and similar games a database of *all possible gameplays* has been built - one can formally prove an AI using that DB can enforce a stalemate at least => it is invincible)

[Lion.Kanzen]

ok there are best to ask this, Ieper, Historic Bruno, Quantumstate(Qbot AI programmer), Writii(Aegis Ai Programmer), Sandler17, Ykkrosh and Redfox (the guy that we have to Optimizate the game and one of oldest in this). you can send a Private message or contact in IRC chat.

http://webchat.quakenet.org/?channels=0ad

[Teiresias]

Thanks for the info&link, i will stand by a few days before doing so - eventually some of them will pass by this thread beforehand, it's probably rude to "hijack" them... (and i could image this question has no simple answer...)

Edit: somehow text was garbled up a bit

Edited November 19, 2013 by Teiresias