An idea: scaling AI boost

[Nexii]

I'd mentioned this idea in passing awhile back.  The problem with the AI is that it plays worse as the game goes on.  So what happens in every game is that if you can reach the early-mid game in okay shape, you've pretty much won.  I'm sure part of this is due to flaws in its general strategy such as ignoring advanced terraforming, how it uses and prioritizes military, and how it does diplomacy.  But a lot of it is also because as the game goes on, there's more choices to be made.  More things the AI can do wrong compared to a human player.  I feel this will be true no matter how much the AI is optimized although it could be diminished. 

The issue with just boosting up AI by super factions is that early game, these boosts are simply crushing.  Even starting near Spartan AI on Transcend can be instant death, let alone Alien faction or custom one that's stronger.  It's not so fun to just die off the start or have the chance of success based on where you start (luck).  So my idea is this: to make the AI boost scale as the game goes on, rather than a flat boost throughout the whole game.

As a simple example, a linear scaling.  The production boost would still be multiplied by difficulty the same as now.
turn 1 > 1% boost
turn 100 > 25% boost
turn 200 > 50% boost
turn 300 > 75% boost
turn 400 > 100% boost.
It could be a curve too I suppose if it was felt the AI needed more/less boost at certain points of the game.

[BFG]

Honestly, from what I've seen part of the weakness in late-game AI is due to the generalist strategy the AI generally tries to follow.  That is, they all generally seem to split resources relatively evenly between Explore, Build, and Conquer approaches so they don't do particularly well in any of the three, while a human player can devote most resources to one of the approaches while minimizing the other two, which provides a superior long-term strategy.  The other key difference between SMACX's AI and a human player: a human player can learn and adapt.  (To wit: in a recent game I watched as Yang tried, unsuccessfully, on 8 separate occasions to take out a nearby Fungal Tower with land units built in a peripheral base, wasting massive resources such that that base never developed.  A human player would have learned from the first failure and changed strategy.)

I don't know what the solution is, but I suspect it would require extensive rewriting of the AI, as I've never really liked the "rubber-band" type AI approach you've suggested.  BUT, that type of approach may be the only option short of a full rewrite - something which is probably beyond any of us.

[Buster's Uncle]

It appears to me that there's some sort of switch-off in the AI; after a certain point they mostly stop expanding -very notable with Yang and Morgan, who ICS a homeland area tightly like mad at first- and stop building base improvements of any kind, not that they ever do much - no further expansion & no real industrial capacity v. a slower-expanding human who (even conquest players do some) invests long-term in improvements = no chance.

[BFG]

Hmm, I've noticed that too.  I strongly suspect it has something to do with the total number of bases in-game (or at least the total number of AI-controlled bases in-game) as the shutoff doesn't seem to be specific to a certain turn/year.  The latest I've ever seen either of them build a new base is circa 2380.

Another weakness of the AI is that it doesn't do a very good job of optimizing units.  Humans know that they can tack-on certain bonuses (for example, why have a 1-12-4 unit when a 10-12-4 ECM unit costs the same?) to make their units more effective.  And they often are not capable of controlling those units effectively either.  For example, I've seen numerous AI plane crashes due to fuel, often caused by having to go around enemy units on the return trip; and (my personal pet peeve since I play a lot of Transcendence games) -- they cannot properly automate Aero Colonies, Gravship Formers or other advanced units.

[Bearu]

The AI difficulties with regards for the colonization and terraforming aspects of the aerial units would likely result from the fact that the game controls the movement and terrain improvements of both types of plans differently for sea and land units. The game specifically checks the current square of the unit and the chassis type of the unit to determine the terraform commands.

[PvtHudson]

One approach to scaling is long-known so called satellite-boosted AIs. If one gives AIs through scenario editor say 30 satellites of each (or some) kind, the AIs will gain raising benefits as they plant new bases and grow old ones.

[BFG]

The AI difficulties with regards for the colonization and terraforming aspects of the aerial units would likely result from the fact that the game controls the movement and terrain improvements of both types of plans differently for sea and land units. The game specifically checks the current square of the unit and the chassis type of the unit to determine the terraform commands.

Hmm...that would explain why the AI seems to properly issue sea-based terraform commands to a Gravship that's already over water, and land-based terraform commands to a Gravship that's in the base square or over land.  Interesting.

Still, I am a bit surprised that the programmers didn't foresee the need for an AI routine specifically for such unit types.  More advanced Builder games run into these routinely.