AI Improvements

Better ability autocast selection. (This will probably be one of the hardest things.)

If I have 100 Stilakus Subverters with Teleport Disable (Whatever it's called) and they are set to autocast their ability, they will all pick one ship and disable it.

I would like a better target process for abilities of all kinds. If it doesn't stack, then don't stack it.

Hopefully this makes sense.

Other then what is already stated, I would like to suggest that the AI notice what kind of ship is getting the most kills against their fleets, and then build ships that are the counter to it.

This should be top on the list! If they ever learn not to charge a SB semi right. We are all doomed. They have moments when they do it right, but add a fleet to next to the SB and they get a bit stupid.

The rest of the list is would be icing on the cake.

What about studying fleet compositions and tactics used in online multiplayer and applying them to AI personalities?

Personaly I find the AI fleet very diverse and very capaible to fight fleet to fleet.

If you took MP fleet composition, you end up with spam of one unit or another. *Cough*Carriers*Cough*LRF*Cough*

Only one suggestions... What about a self learning AI? No more easy-medium-hard difficulty but a AI who learn tactic for counter people who play with it...

Seem that the technic is there : http://www.aihub.org/modules.php?name=News&file=article&sid=203

Of course, this improvements will be more difficult to implement that open the actual AI to modding but it is the dev choice :p

I think you should get started on that right away, Thoumsin. We'll expect your self-learning AI to be presentable in a couple months.

Somehow, I doubt Skynet will have it's start with Sins.

One simple (?) fix might be to limit the number of caps and starbases that the AI builds.  The limit could be based on the amount of time into the game.  For the first hour, no more than 2 caps at once, etc.  The AI could also be instructed to focus on spamming out LRFs and flaks with a small percentage of carriers/fighters (in an attempt to mitigate bomber spam).

This is where things get really tough.  How do you program the AI to decide which side of the gravity well to jump into?  Can you teach it to "kite" its ships around so as to avoid being chased by Vasari starbases?  Can you teach it to make bomber spam when needed to counter starbases?  Can you teach a Vasari AI to build a starbase at the right place and to then manage it properly so as to kill an enemy (stationary) starbase.  Ironically, Entrenchment and the addition of the starbases really weakened the AI relative to human players.  It added a new layer of strategic complexity to the game, which was a very welcome addition, but it came at the expense of the AI.

In all seriousness, why not?  Why wouldn't real artificial intelligence start by playing games?  That way if it screws up it doesn't cause any actual damage in the real world.  (The IBM computers seem to specialize in playing Chess and Jeopardy.)  I certainly wouldn't want to hand it the keys to our nuclear arsenal (and not just not right away, but ever).

I think the AI should be taught to counter enemy fleets better - eg. if enemy has X frigates of type A and Y frigates of type B, the AI should build Z frigates of type C to counter most effectively. So, if the human enemy spams LRF, the AI will plug the numbers into its formula and conclude that it needs to spam flak.

I don't think it would be that hard to create a formula such as this.

It would also be essential to fix the suicide-on-starbases tactic the AI always uses. It would also be nice, as mentioned, if the autocasts were fixed so they don't all target the same frigate and waste nearly all of it.

Also: have the AI *prioritize* targets. If there's an enemy ship in the gravity well, the AI should assess its threat ("is it a siege frigate? do I need to kill it? or is it just a lone mostly-harmless scout or flak frigate?") and conclude when it would be best to ignore it and when it would be best to engage the enemy fleet.

They already do this. Granted its not a fast change but if you play on slow you can see them adapt their fleets to yours.

Sometimes, they just pointlessly autocast..  I mean, I understand why, but in some cases, POH for instance gets used improperly as a result because the enemy AI thinks "Ooh, my cooldown is done and I have the antimatter, so I'm going to spam POH on random units that have no effect on the battle."

And if they do manage to adapt to your fleet, then could they potentially do so faster?  So that hard isn't just the AI with extra money, but actually is better at countering you.

Not exactly what you are describing, but I recently played a game against an unfair AI and while besieging one of my planets, using a star as a jumping off point, it spread its ships into about two to five groups across the gravity well. However, since I had a supporting fleet (or two fleets by my organization strategy) ordered to not leave my starbase's minefield they all promptly were wiped out so something about that needs to be fixed.

As for my own wishes, I want a feature where you can create AI "generals" that serve the same function as real life ones, you give them nominal control of some of your ships (which I can override at any time) and give them assignments (spread your ships to defend these planets, capture this planet etc.) One of my favorite strategies is to separate my fleet and hit the enemy from all sides (I won't win on every front, but it confuses the crap out of the AI) and giving some of these detachments to an AI and then telling it what to do so I can micro the most important two or three fronts would streamline my operation quite a bit

Doesn't sound like it'd be that different from auto-attack (and auto-colonize) behavior, after warping in...

AI generals have been something that were suggested long long ago, but there are fundamental problems simply because you would have to do all sorts of things to improve the tactical ability of the AI.

If someone did make a learning AI and then gave it to some of the better players to observe their actions, then you probably could end up with a good, tactical AI, but at that point, you're just delegating all you're work and placing a TON of work on the devs.

What I'm thinking is auto-warping and strategizing, not just auto attack which I rely on anyway.

So what it comes down to is that to feed my laziness, the devs have to give us a modabble AI

Some other obvious improvements which shouldn't be too hard: the AI shouldn't queue up like 50 or so light frigates in a row at the very beginning of the game, even if it has the money (read: is Viscous/Cruel).

Something the AI is greatly lacking in is the understanding of how to expand quickly.

If I could, I would program the Cruel and Vicious AI to expand at the maximum possible colonization speed (eg. 5 or 6 planets by 10 minutes if map is a random small 1v1) using the method I try to use. If the planet types and number of enemy ships can be processed through an AI program, it would be doable.

The current AI's expansion rate is horribly slow, even on the harder settings - it often takes the AI more than 5 minutes to defeat the neutral ships on a planet and to move on.

Overall expansion should be reworked, I'll have a game with a long, multiplanet front and the AI will suicide against my one fortified planet (which I fortified because it was attacked) like a five year old screaming "But I want THAT planet!"

Yeah, the AI should learn NOT to move into range of starbases/stationary defenses if it doesn't have the overwhelming advantage.

If it's just anticipating fleet-to-fleet combat, though, if the AI builds units reasonably intelligently, I don't think there'd be that much of a need for a reworking on that front.

Well, the absolute maximum for vicious-level AI's and their more or less infinite funds would be immediately building enough military labs to build light carriers, and then start mass producing them.  From there, it would simply send them through and with enough, by the time they'd get through the well, the SC would have eliminated everything.  Send in colony frigates after that.

Somehow the idea of giving the AI the desire to mass produce LC's seems like a bad idea..  Of course, at such a high level of difficulty, I guess it sort of makes sense...  Be prepared to mass produce flak...

Right, those AIs get a huge income bonus. I was thinking more along the lines of teaching it how to expand quickly given very little research and extremely limited funds (like humans). I guess for that to be more useful there would have to be more of a differentiation between AI income bonus and AI playing ability.

It would be great (but probably unrealistic) if you could control two sliders or buttons when creating AI, to modify their income multiplier and/or their skill - otherwise, this quicker-AI-expansion idea of mine wouldn't be worth as much due to the harder AI's ability to simply spam attack units.

If not already mentioned then more commands to give to A.I. allies, like I'm working with the Advent in a very long game, I helped the Advent destroy and nuke a planet for colonizing but they wouldn't send a ship to colonize it, the enemy culture wasn't a problem as after 10 minutes of waiting I just said **** it" and colonized it myself.

That would definitely be nice...