marigoldran

Join Date 06/2013

+41

Giving AI More Economic Bonuses to Counter Human Economic Specialization ISN'T GOING TO WORK. Instead Improve AI Economic Management

The problem with GC3 is EXPONENTIAL GROWTH. Thanks to the way the economic system works, the percent multipliers from starbases, population, tech, and buildings can make planets 100x or even 1000x more productive than before.

Consequently, giving the AI flat production bonuses will lead to either:

1. You gave the AI TOO MUCH ECONOMIC BONUSES, which means the human can never catch up and the game is no fun.

2. You gave the AI TOO LITTLE ECONOMIC BONUSES, in which case after a human specializes his planets making your flat percent bonuses effectively useless.

For example: AI gets a flat 100% bonus. Sounds good right? Unfortunately the AI can't specialize its planets and the human can, and on turn 50 the human has planets that are 10x more productive than the AI because the human knows how to specialize HIS planets.

Consequently, now you give the AI flat 500% bonuses, in which case the AI EXPANDS TOO QUICKLY and colonizes all of the planets on the map before the human even has a chance to, meaning the human will never be able to catch up or forcing the human player to stay on a minimal number of worlds.

In other words, FLAT BONUSES as a way of balancing GC III WON'T WORK. Either the bonuses are too large AND TOO POWERFUL IN THE EARLY GAME and the AI expands too quickly or the bonuses are too small AND POINTLESS IN THE LATE GAME THANKS TO ECONOMIC SPECIALIZATION, and the human over-specializes and crushes the AI.

Trying to add in FLAT BONUSES to counter exponential growth is a losing proposition. It's almost impossible to balance correctly. I have proven that this mindset is flawed. Much better to teach the AI how to specialize economically.

Locked Post

Advanced Search

Watch this post

Do not email me updates for this post Email me updates for this post

Successfully updated karma reason!

Much better to teach the AI how to specialize economically.

Suppose, that Galciv3 AI becomes strong..really strong?
Then we get the situation as in chess..playing again a grandmaster chess with his ELO rating..a normal human being can never win this.

Successfully updated karma reason!

exponential growth is always tricky to balance with cheating

i think some solutions in other games have been to add more bonuses at different stages of the game

for example, in the early game the AI can start with bonuses that help with the first few expansions and research

in the later game, the AI can get bonuses that help with military and improvements

it ends up being more fun that flat bonuses from the beginning

Quoting janhardo,

reply 1

Much better to teach the AI how to specialize economically.

Suppose, that Galciv3 AI becomes strong..really strong?
Then we get the situation as in chess..playing again a grandmaster chess with his ELO rating..a normal human being can never win this.

it's very easy to program a grandmaster-level AI to play poorly (or have penalties) on lower difficulty levels

Successfully updated karma reason!

Hi,

I agree wholeheartedly that the important thing is to improve the AI not to give it ludicrous bonuses. This isn't an easy process at all of course and I wasn't expecting when I bought GC3 that it would be much more than a good beta - which it could be suggested it still is. The tweaking that SD have been introducing from update to update is obviously well intended but can have unintended consequences. And the recent hiking of killer tech values in 1.2 has stymied the AI rather than humans at least on the gentle level (genius) that I've been enjoying. (The game I'm now playing feels like a sandbox - much more so than previous versions: all I'm doing after a couple of hundred of lethargic turns is the space station optimisation waltz until I can summon the (mental) energy to group my forces and take them out.)

The idea of a game like this is surely to provide an interesting and exciting challenge without it being impossible. Chess has been mentioned and I am a grandmaster. Chess engines are so strong nowadays or rather calculate tactics so well that we can't really play against them because human beings even very good ones make mistakes - drop the odd pawn or more - which they exploit. So most of us don't play against them at all except for some people for training purposes. Instead we use them to check our play afterwards or prepare openings beforehand. The analogy of a computer game is far from perfect because unless you play people you will be facing an engine. But you can always choose the level you're happy against rather than taking on the top one to be macho

GC3 has the potential to be a really superb game. It has nice clear graphics (as distinct from Civ5 say) which are easily comprehensible and personally I really like all the micromanagement (fiddling) within limits though God knows the UI needs a huge overhaul and they really should prioritise.

In any case, eventually I imagine the developers and modding community will produce a decent AI and this will be hugely more satisfying to play against than a weak one pumped up by bonuses (though the latter is infinitely easier to create.) It will take time though.

Cheers,

Jon

Successfully updated karma reason!

Quoting janhardo,

reply 1

Suppose, that Galciv3 AI becomes strong..really strong?

Then we get the situation as in chess..playing again a grandmaster chess with his ELO rating..a normal human being can never win this.

It won't.

Chess AIs have been worked on with enormous amounts of funding for about 60 years. Dozens and dozens of teams of academics working with supercomputers have been drafted in to work on the best chess AI possible, as it's been generally considered to be a key aspect of AI research. Stardock do not have those kind of resources, and GC3 is vastly more complex than chess. The last thing we really need to worry about is the AI becoming so good at the base game that the player cannot compete anymore, because that's not going to happen from AI strategy improvements. It's pretty easy to make it happen from handicap bonuses, which is yet another reason why handicaps are a shabby band-aid solution to poor AI behaviour.

Successfully updated karma reason!