The AI was buffed a while ago in entrenchment to make them take a but load more anti structure cruisers when attacking starbases. Too bad they will sometimes still have them attacking from withtin the range of the starbase.

Put it this way, the AI fights from the Empire tree.  The only ship that's micromanaged are the seige frigates. 

I agree about the assault specialization, but it's the AI's fault on choice of targets.  The offensive SB doesn't really need weapons, just a presence with another fleet which is why it has the phase stabilizer option to bring in a new fleet quickly.  AIs have only so many options to choose from and doesn't really know how to change it's strategy other than running away to regroup.

They need to add an algrorythm where the AI makes an assessment of the strength of an enemy starbase, support structures, any enemy fleet in the gravity well, and its own fleet and then decide that it either makes sense to attack the starbase or to keep away from it completely.  That might at least end the problem of the AI-suiciding-on-starbases.  Of course, if you want the real thing, you have to come play it against other people onilne.  It's lots of fun when you can trap a guy's fleet with phase jump inhibitors and have your Vasari starbase combined with subverters crush it.

Yeah, AI target selection could certainly be a bit better.  They probe for weaknesses pretty well (sometimes excessively, imo), but then they don't always have fleets constructed that can get the job done.  I suppose humans can be much the same, but we learn over time as opposed to a static AI which doesn't so a better way to analyze those systems would probably be the most we could reasonably hope for. 

I would also suspect some specialized scripts would be interesting to see in action - prebuilt strategems based upon race/map size combinations that would occasionally get used.  Then again, if you're playing against the AI it might not be able to be flexible enough to adapt to a human response.  In addition, the scipts would have to be balanced against each other to make sure that one AI isn't any more dominant than another, which would require additional testing and terefore expense.  Maybe something best left for the next iteration of Sins.  Ah, well.

I guess part of the algorythm would also need to include instructions to keep ships themselves away from TEC starbases since they could explode and wipe out the fleet.  (The only really safe way to take them down is from a safe distance with a swarm of bombers.)

Not to pry, but what's preventing you from trying it online?

One thing that surprises me is just how many people say that their Internet connections aren't up to snuff for online multiplayer.  I believe them, but I do find it kind of sad that it's still such a big problem today.  Perhaps it's just hard to get halfway decent broadband Internet access if you live out in the sticks.

I suspect that the developers won't put too much effort into improving the online multiplayer aspects of the game, such as the user interface and player-skill-based matchmaking, following the logic that you don't want to spent 33% of the development budget for a feature that only 2% of the purchasers will use.  However, ironically, improving the accessibility and desirability of the online multiplayer aspects of the game might very well be the best thing they could do in Sins2 because getting people to play it online against other people will solve all of these gripes people have about the AI, perhaps retaining people's interest in the game longer.

I'd like to see them come up with an optional matchmaking system that would match up players of nearly equal skill and experience against one another.  This would become feasible of thousands of players could be brought online at once.  The game also needs an online multiplayer tutorial and other means of encouraging single player only folks to come online, perhaps active invitations to people who have Internet connections but who aren't logged onto ICO to come join games for new (to online) players.  (It would know if someone has been online or not and thus whether someone is a new player by CD key, perhaps only issuing invitations afte someone has played 20 games in single player.)  "Game X has 8 new people in it and only needs 2 more, you are invited to join."  It could track people by CD key allowing them to play, say, 30 games on ICO without having to create an ICO account, making it very easy for newbies to play against other newbies.

Easing people into online multiplayer and matching people up based on skill might really be the best thing they could do in terms of improving the game's AI.