Impossible difficulty suggestion

[Psillycyber]

Lately I've noticed that my games on impossible difficulty tend to follow a pattern.  Hanging on for the first 125 turns is nail-biting and sometimes downright impossible indeed...and oftentimes more stressful than fun...but then if I can consolidate a decent position, the game hits a sweet spot of fun and challenge for about 25-50 turns, and it continues to get easier and easier from that point on.  It seems like I rarely get a chance to use most of the late-game toys because games that last more than 200 turns are rare.   

In short, I think the game would be more enjoyable if impossible difficulty were easier in the beginning, but then just as hard or even (dare I say it!) harder later in the game.  

I wonder...to what extent is this difficulty curve throughout the game due to the fact that the AI always has 4 insta-built colony ships on impossible?  Also, correct me if I'm wrong, but the impossible AI also starts with more population and factories, correct?

I wonder how differently the game would play if the AI on impossible had to build its own colony ships like the human player, and started with the same homeworld resources.  

I understand that the AI might struggle with knowing when to build a colony ship...so, what if the AI only had a chance to have 1 colony ship insta-built at any one time instead of 4 on impossible?  And plus it would be nice too if this insta-built colony ship of theirs always had to start at their homeworld (or maybe at their world with the largest population).  It is frustrating to see the AI on impossible chain-settle planets on the frontier with a weak escort force and magically re-duplicating colony ships that seem to re-duplicate even while the AI fleet is en route to the next planet.  That sort of blatant cheating really brings the human player out of the game.  

I wonder how hard something like this would be to patch into the game....

By the way, I tried stepping down to hard difficulty with the Mrrshans, and it was a joke.  Total curbstomp.  No pressure at all until about 2375.  I had atmospheric terraforming by 2390.  Advanced soil enrichment by 2405.  Antimatter torps by 2410, which shredded enemy missile bases that had at most 7 layers of shielding.  Victory by 2425.  So, stepping down to hard is not the solution, unless I just want to mess around.

[haphazard1]

There is quite a sizable jump in diffictulty from Hard to Impossible, and if your skill level (like mine) is not quite fully up to Impossible it can be frustrating. Hard is not difficult enough, but Impossible is still too much. Playing stronger races on Impossible and weaker races at Hard can help somewhat, but it would be great to have a difficulty level between.

The free colony ships are certainly a big part of the AI's toughness on Impossible, as the AI races just rush the land grab incredibly fast. Silicoids are especially bad as they just swarm everywhere grabbing anything within range; only their pop growth penalty holds them back from totally exploding to dominance if they get a chain of worlds within range at the start.

I suspect that requiring the AI to actually build colony ships would cripple it fatally. But if there was a way to trim back the advantage at Impossible, it would help. You mention they get 4 free col ships on Impossible -- what do they get at Hard? The AI needs that early expansion boost to be competitive enough/tough enough, but maybe some balance in between Hard and Impossible could be found.

[RefSteel]

AIs get up to 4 free colships at every difficulty. Each turn, if they have less than 4, they have a chance to get a new one, generally en route to a colony destination if one exists. If I remember kyrub's notes correctly, the chance is based on their current planetary reserve - something on the order of a treasury/500 chance each turn. On Impossible, this rapidly turns into "one free colony ship each turn unless 4 are already present" as the AI builds up its reserves. On lower difficulties, it just takes longer to get there.

As for in-between difficulty levels, that's one of the things variant play is for. I have found in MoO though that Impossible is just a whole different game from every other difficulty. Things like the number of missile bases AIs build on their worlds have an enormous effect on the options available to the player in my experience.

[rho21]

The difficulty curve over the course of a game in 4X games has always been a tricky problem for developers because it's really difficult to write an AI that is even remotely competent at intricate strategic puzzles.

As a result, the AI will tend to struggle in the mid-game almost no matter how big the advantages it is given are. The whack-a-mole nature of mid-game MoO is an example of this. The AI produces really quickly, but then throws it all away on lots of mid-sized fleets that are poorly directed and assembled and often don't pose much of a threat.

The best hope for a competitive mid-game is to give the AI even more help in the early game, to get out to a reasonable size quickly. High-level Civ VI is a good example here: the AI gets loads of extra starting units (including an extra settler). If it starts near you and decides to throw everything at you around turn 20, there's often not much you can do to stop it. Unfortunately the AI is weaker than MoO's so generally falls well behind by the mid-game.

So to answer the original post, I don't think you'd get much joy out of removing starting advantages in favour of greater long-term production advantages etc. It would be really difficult to get the balance right, and I reckon whatever you did you'd end up with too many games where the AI wasn't competitive at all due to insufficient expansion.

I think there's lots of hope for the future though, in the form of Google's AlphaGo Zero. It's already great at several abstract strategy games, hopefully it won't be too long before they (or another comparable AI project) try it out on more complex games. Add another 10 years for hardware components capable of running it on cloud servers and perhaps we'll see some truly strong AI for Civ VIII.

And yes, I do believe Firaxis would invest in it, because at some point it will be cheaper (not to mention much stronger) than an AI written from scratch.

[RFS-81]

I doubt you'd even need cloud servers to run a challenging AI. The bottleneck is programmer time, not CPU time. Given that AI doesn't sell games, it has to become ridiculously cheap before game developers just start throwing neural networks at it. How long is it going to take until it's cheaper than writing a bad AI from scratch?

Relatedly, there's a company, Arago, creating an AI for FreeCiv, though unlike AlphaGo, it's based on expert players formalizing their strategies. But Arago is not a game company (nor working for the games industry), the point of the exercise is to showcase their product. "You've heard about AlphaGo - but our system beats humans at a much more complicated game! That's what you want to optimize your business processes!" Still interesting to see what comes out of it, of course!

[zyzyfer]

To OP: I find the same thing about my games. I tend to be pretty intuitive at knowing when to start pilling on missiles bases so the AI doesn't attack me at all throughout the defensive phase of the game.
What's worse is that one can bate the AI with scouts; it will attack a 50 missile defended gravity well (that it normally wouldn't), if a scout is there; kill the scout while in the process getting most of its fleet wiped-out by the missile bases, then retreating out with the remnants of its fleet.
You can steal as much tech during this phase as your heart-content; the AI even after declaring war can't follow through & break your heavily fortified planets..

I also find the AI doesn't handle repulsor beams with HEF well at all. My impossible games always tend to end once I steal/attain gauss-cannon, HEF, repulse beam, auto-repair & a nice bomb (my last game I managed to steal neutronium bomb).
I had 15 Huge designs to begin my galaxy assault using repulsor beams; gauss, auto-repair, neutronium bomb; I totally went to town with the galaxy. The AI just had no answer even though it was ahead in technology + fleet.

I think my next challenge will try playing with the weaker races on impossible i.e. kitties (something i haven't done yet).

[haphazard1]

Playing the weaker races is one option for more challenge. And as RefSteel said, variant play is another. Things like limiting specials used (no repulsors, etc.) or colonization limits -- taking some choices away from the player -- can be interesting. Look at the rules of the various Imperia that have been run here at RB, or some of the succession games over the years -- lots of ideas there.

Some strategies are very effective (huge auto repair designs, for example) and tend to get used because they work. Sometimes you do nto have the techs in your tree and have to improvise another solution, and those tend to end up being the games which are most memorable. At least that is true for me.

[Ianus]

Sounds like we need a Darlok SG. Maybe on a Small map just to make it interesting? Then we can sit back and tech up to Gauss cannons in peace

But seriously, anyone up to play?

[haphazard1]

Hmmm, Darloks on impossible. Any other variant rules, or should that be difficult enough?

I really ought to find time to get back to my SMAC exploration game, but some MoO would also be fun. Who needs sleep, anyway?

[Ianus]

No espionage? 'Loks on a small map is probably plenty but I'm up for most anything. Let's see what kind of interest we can get.