Interesting idea from the main forum

**easytarget** · 10-03-2013, 00:27

I PM'd to ask what other games he was referring to, I'm curious to hear as well.

And your player profile library and randomizing that seems like an interesting idea as well.

This is a bit like the online drop in w/o the hassle of hoping someone will drop in, you play to the strength or propensity of most TW players to play SP and just collect what they do and close the loop by feeding it back into the system to help improve the factions.

**easytarget** · 10-03-2013, 03:05

The game he was thinking of as a point of reference for this was Spore.

**Sp4** · 10-03-2013, 04:13

Sounds interesting but it also sounds like a lot of work.

**easytarget** · 10-04-2013, 00:27

I guess, I was sort of thinking the "light" version would just be using what humans use as a build order for settlements, provinces and stacks as a place to start.

I mean CA is already putting this in already, it's not like this is new code. We're just suggesting they tweak what's there because right now it loads up a script list that's brain dead resulting in countless encounters with starving AI factions, stacks that never build past 1st tier and settlements not properly built out.

**al Roumi** · 10-04-2013, 10:47

Probably unfair to pick at the OP (especially as they're not even here) but you couldn't really have a direct replica of your play as e.g. Rome with another AI faction. This is simply becasue the circumstances under which the player makes their decisions are surely varied in many different ways to the context of the AI faction in question.

That said, and this may be the real merit of the proposal, if the AI were to learn from how other people ahve played said faction and apply it to your game, that could eb interesting, but again - once the AI starts interacting with other AIs and the player, so many different options are available that the AI still needs to find its own way in each game.

So perhaps the most that can be applied is a sub-set of scripts for pre identified contexts/situations that the AI can "choose" to apply, but then again (again again!), this is surely how the AI has been coded anyway by CA. Choosing what to do when and why, flexibibly and responsively, is surely the problem AIs struggle with anyway, rather than the particular strategy to pursue, which can be scripted.

**MadKow** · 10-04-2013, 11:36

In photography there is one thing called matrix mettering that basically analises the image by dividing it in zones and compares it to a database of scenes and then chooses the best exposure for that scene. This happens while your fingers does "click", so it's pretty fast. Nikon claims to use a db of 10 000 scenes.
I mention this because i guess this sort of thinking could be, maybe, used to program the AI to take a "snapshot" of it's present turn, and deciding on its next move by comparing it to similar situations in it's database. The key here is to have a large enough database that covers as many situations as possible, and thats where the input from players experiences comes in.