[ROTP] What are your ship design tactics?

[Ianus]

Well put! And I agree that an "adaptive" AI, even in regard to small-scale things like planetary attacks such as you describe would be revolutionary, particularly for MoO. We have all seen the AI make futile attacks against strongly defended targets and get nowhere only to repeat the same futile attacks again and again. While the AI is excellent at identifying and targeting weak points in an enemy's defensive perimeter it is NOT good at learning from past mistakes.

Although now that I think about it AI behavior makes it clear that the AI treats static defenses (missile bases) differently than mobile defenses (ships). I remember one game where I had confined an enemy race to a corner of the galaxy and had moved on to a stronger opponent elsewhere. Yet I found myself constantly distracted by raiding parties from the contained AI. In frustration I set ALL planets in range to build missile bases until I had at least 100 of them on each world. After that the attacks stopped because the AI could no longer identify a weak point to attack. Contrast this with its earlier behavior of continually attacking even though I had an unbeatable fleet which could (and did) move to respond to each attack.
Now this is probably a result of programming requirements as I can't even begin to answer the question of how an AI could consider empire-wide fleet strength while at the same time not be prevented from attacking by the hypothetical threat of a fleet positioned far beyond effective response range, but it IS a problem. Anyway, adaptive AI would be a good thing!

[Ray F]

This would also be fairly easy to implement (just save total BC of the attack fleet and the 6 numbers for the designs as long as they have not been updated) and might be more fun as compared to the optimal "either send nothing or something 100% invincible".

As a general rule, programming problems like this are far more difficult than they may appear to be.

[ignatius]

We have all seen the AI make futile attacks against strongly defended targets and get nowhere only to repeat the same futile attacks again and again.

That's it. The first inadequate attack does not break immersion (actually, quite the contrary as you would expect the first attack from an unknown alien force to be either way too weak or way too strong) - the repetition does!

Another example would the the endless stream of complaints by utterly defeated AIs from their last Zoo planet. If you're cornered and completely at the mercy of an essentially all-mighty opponent, it's probably wise to simply shut up and not draw attention to oneself. Flattery and cooperation - with or without an ultimate attempt at treachery, should an opportunity arise - are the way to go here.

And I agree that an "adaptive" AI, even in regard to small-scale things like planetary attacks such as you describe would be revolutionary, particularly for MoO.

Well, technically, MOO already has adaptive AI - when dealing with broken treaties. Not much and only in an area where there's no alternative. From what I've read, Ray already uses adaptive AI with tech trading.

There are not too many areas in MOO which lend themselves to the technique, though. It can only be used in situations which occur sufficiently frequently. Offensive and defensive battles (esp. defeats) are the obvious choice to harvest useful information. Some random ideas that come to mind:

- When a planet got successfully attacked, build additional missle bases in all colonies in reach of the attacking stack.
- With 6 active stacks, scrap the oldest stack which didn't find a target it could hurt (due to shields / repulsors)
- See which designs got attacked first by the human player and produce more of them (might be gamed, though)
- analysing the "what kills what"-matrix might help identifying working own designs / weapons
- Do a "war-game" replaying the battle with identical fleets both controlled by the AI. If the outcome is different from the actual battle, it is a hint that "more of the same" might not work.
- Simulated battles could also be used to decide whether to retreat or not (or to attack at all), etc.

ignatius

[HansLemurson]

Just make an AI that does only the things that I want it to do, and none of the things I don't want.

Code:

if(! movePlan.isStupid() ){
    movePlan.doTheStuff();
}

There.
Just saved you an afternoon of coding.