AI Pre-Knowledge

The reality is that without some advantages like that the AI would be extremely handicapped against a human opponent. We're really really far away from having an AI be able to not 'cheat' in some way in order to provide a human a challenge.

I mean think about it this way. Right now chess from an AI perspective is mostly a 'beaten' game. Even incompetent Chess AI's can kill your run of the mill player. And we're certainly at the point where the best ones are more or less able to beat grand masters.

Look at Go. A bunch of white/black stones on a 19x19 board. We're only 'just' kinda getting to the point where AI can beat competent human opponents.

Now think about your typical turn based strategy game. It's literally orders of magintude more complicated. Board sizes are exponentially larger (a 19x19 grid in Galciv2 would be what? like half of a system). The # of resources to manage (economy/tech/military/base/etc) is a bazillon times more complicated. And as games progress, we as players demand even MORE COMPLICATED systems to interact with. We want more realistic diplomacy, more divergent and complex tech trees, we want deeper economic systems, more races, more events, more choices. Oh yeah and btw AI programmer, your AI can't spend 10 minutes calculating each turn either, more like 30 seconds, 1 minute on the outside. Oh and it need to run on a TI-84 as well.

Games can easily capture and quantify and track these concepts. But the problem mostly is that AI systems haven't evolved at the pace at which we demand. People constantly complain in almost every game that 'omg the AI is stupid, lazy programmers!. But that's simply the furthest from the truth. In fact, generally depspite the fact that AI is so critical, generally speaking the 'resource allocation' from a system perspective is really low. AI or fancy textures.... TEXTURES IT IS!! AI programmers are really constrained in what they can do, vs what they are expected to accomplish.

If we get to the point where AI systems don't need to 'cheat' to win, I'd be more worried about Skynet taking over

I agree pre-knowledge should be limited where possible, also that computational complexity needs to be carefully managed, and some degree of cheating will be needed on the highest difficulties.

In some cases AI implementation by using expert systems based on watching human players can deliver interesting AI gameplay while remaining algorithmically manageable.  It's obviously limited in scope but Multiplayer is a real opportunity here.  Stardock could routinely watch what top players do to generate many implementable AI ideas ... ideally continuously improving the AI over the games lifecycle.

GalCiv2 already did that with the metaverse. It looked at build orders and such and correlated new strategies for the AI depending on what 'worked' and what didn't.

Just reading up on that now, it's been a long time since playing GalCiv 2, and I didn't play Metaverse.  How did it go in practice?  Were the improvements in AI visible?

I see in GalCiv 2 that Stardock openly published they were reducing unfair advantages for the AI compared to GalCiv 1.  I assume they are continuing that journey.

https://forums.galciv2.com/138274/page/1/

Sort of an old post, but this might be what they're looking at for GalCiv3, especially given the very high system requirements.

You should take a look at the back story. It states that the other races sent out sub-light drones over many millennia to map the galaxy. While there may have been some coding practicalities in the past that required this in the back story, I think changing the back story is not something SD would do even if those practicalities have been overcome.