Star Wars Rebellion

[whyBish]

As far as the AI goes - how many games really have challenging AI?  Most games will have marginal AI with scaling 'cheats' for the computer player, i.e. Civilization.

I hated Civ/Alpha Centauri MOO3 et al. too for their crap AI. I did have fun a long while back (still in uni) making AI scripts for dark reign. That was an OK RTS, but had an (at the time) innovative system for AI creation. The AI was controlled by a set of parameters. The game world could be tested for a set of values. Scripts could be created that updated the AI parameters...
You could do fancy things like having a script that moved through appropriate parameter sets for various states, so that the AI would (for example):
- react differently if it was outnumbered
- calculate weak spots in the enemy line differently depending on the units it was planning to attack with
- Do growth early but produce some units (or defense) early if scouts detected a rush
- Switch back to an appropriate build strategy if it is partly destroyed
- Give text messages saying when it is "coming to get you"

That was fun, but unfortunately you still needed a handcoded AI. To date I haven't seen a game that actually uses AI, they just claim to use ArtificialIntelligence, when really they are using AutomatedHardcodedLogic (even if it is parameter based).

To be AI, it needs to:
-have an (automated) feedback mechanism (so that it can improve), such as adaptation/evolution/experimental design type techniques.
- It needs to be able to sense its environment (e.g. knowing that there is a king at D8, or knowing that it is outnumbered)
- Be able to affect its environment (e.g. move the king to a spot)
- Have a success measure for the feedback mechanism (e.g. knowing whether it won or lost (for evolution), knowing whether it was doing well or poorly (for adaptation))

The Civ (& co) AI failed on 1,2, & 4. We don't call people that fail on 1,2, & 4 Intelligent. They would be just lying on the floor in spasms, unaware that they were lying on the floor in spasms, with no clue that lying on the floor in spasms was a bad thing.

Dark Reign AI failed on 1 & 4. Which was a huge improvement, because the parameter set that defined what to do, could at least define what to do dependent on the current game situation.

Ironically, to make a real game AI, 1&4 are the easy ones to do. It is 2 & 3 that are the hard ones. This is because the evolution/adaptation/experimental design techniques require a (relatively) smooth success surface to traverse (It can be hilly as you like, but small perterbations in parameters should usually lead to small changes in the success function). The way you choose to show the AI the environmental inputs, and the way you choose to let the AI determine output actions determines the 'smoothness' of this surface. If the AI knows the position of every unit then the surface won't be (as) smooth, if the AI knows that it can see lots of your units and not many of its own then the surface will be much smoother in comparison. Of course, you can make up for a lack of 'smoothness' by allowing more time for the AI to adapt/evolve, but this brings you closer to brute force approach (and exponential time issues)

Am I off topic yet? :P