Darvin's AI Wish-List
Sins Forums
Let's cut to the chase: we all know the artificial idiocy is pretty bad currently. Not to insult the developers, as I'm well aware of the incredible difficulty in creating an AI, but there are many areas where it could be improved. Even addressing a couple of these would be very nice and much appreciated by the community. Entrenchment greatly reduced the difficulty of the AI, since it already had trouble beating a strong defensive formation and starbases often made it possible for players to create something the AI just didn't know how to break. I'd like to see diplomacy restore some of the challenge to the AI, and hopefully make it less of a joke.
Repair Bays – Use 'Em
Over and over again I have stressed the importance of repair bays to players who are learning the game. They are relatively inexpensive, but are easily more influential than your other static defense options. Yet our good friend the artificial idiot won't use them. I occasionally see one or two, but they're few and far between and often later in the game. Not good enough.
Worse yet, capital ships that do retreat from battle often find themselves with nowhere to repair. This means that in those rare circumstances where the AI does successfully preserve their capitals, they end up in another battle five minutes later still sporting heavy damage.
The bottom line is that the repair bay should be a very high priority research item. It should always be researched within the first 10 minutes of the game and be deployed near any battlefield. Structures (particularly starbases) should be built nearby repair bays, and injured units should fall back to the nearest repair bays.
Know when to Retreat
Too often I see the AI take critical casualties in a battle when it could have retreated. This is particularly bad against starbases, which for some reason it's loath to retreat against (presumably because it's going to take some hull point damage on the jump... but why you'd prefer to be destroyed rather than take some damage and live to fight again I do not know). The AI needs to know when to fold and regroup.
Certainly there comes a point where the AI needs to be launching desperate and risky attacks, but there's a difference between suicide and desperation. A desperate attack is one where the AI is still capable of dealing some damage to the player, whereas a suicide attack is one where the AI has little hope of actually dealing lasting damage but stands to lose significant military assets.
Sensible Unit Priorities
Too often I see an AI will prioritize replacing light frigates or siege frigates when it's being attacked relentlessly by long range frigates. The AI needs to better prioritize its unit choices based on its present situation. There's nothing wrong with having light frigates, but if the enemy is massing long range frigates then this should be the lowest priority on its replace list.
This really also deals with the love of siege frigates. The developers have never addressed this, indicating to me it's intended to be an intentional “handicap” of the AI, however it really impacts the enjoyability of the game. I don't mind the AI producing some siege frigates, and as much as I find it stupid the suicide attacks are not the biggest problem. The problem is that replacing those siege frigates is a high priority for the AI, so in a major battle that will decide the outcome of a game, a siege frigate pops out of the nearest frigate factory when it should have been a combat unit.
Capital Ships – Preserve 'Em
While occasionally the AI will get capital ships out of the danger zone before they die, more often than not they will begin their retreat much too late if at all. Capital ships are huge targets and have plenty of opportunity to get out of the line of fire if they're taking too much damage. The AI needs to recognize when to fold its capital ships. This is not to say it should retreat its entire army, just the threatened capital ship (something it doesn't do, presumably because it's loath to sacrifice fleet cohesion).
Of course, retreating a capital ship is often superfluous if it just sits there in the next gravity well. It doesn't take a player long to realize he's more than capable of pursuing into the next gravity well. The capital ship needs to retreat to the nearest repair bays, or its escape may yet fail. If there aren't any nearby repair bays, then my first point clearly hasn't been addressed.
Capital Ships – Discriminating Selection
The AI currently picks its capital ships somewhat randomly. The result is usually that its capital ship choice mismatches its situation, and more sinisterly it often lacks a necessary capital ship for its dangerous situation. The AI needs to think about its needs and classify each capital ship according to its role and what abilities it offers to fulfill those needs.
So, if the AI is in heavy combat situation, the Antorak Marauder should be a low priority. Its abilities are not heavily applicable to the current situation. On the other hand, the presence of a level 6 Marza may demand a counter-ability, which should accordingly raise the Antorak's priority. The ability selected should be affected similarly. The right abilities for the right situation should be chosen.
That is, if there are two squads of fighters buzzing around, flak burst should be a low priority for the Kol. On the other hand, if hundreds of bombers are flying around, flak burst should have such a high priority that there is virtually a 100% chance the next capital ship will be a Kol and it will pick flak burst. The situation demands it, and if the AI doesn't react accordingly, it may a well surrender... which is another issue entirely.
Better Siege Capabilities
The AI's biggest problem currently is that it's really bad at killing starbases, or really any location that's well defended. At very least it needs to recognize when it shouldn't be getting within arm's reach of a starbase. Better yet, it needs to identify that the biggest threats are the repair bays that need to be taken out first. Constructors should be a high priority as well. If there are nearby enemy reinforcements that could severely tilt the tides of battle, removing PJI's from the field should be their first move. While better performance in combat in general would be a good thing, the siege is a place of particular weakness.
Know When to Bypass
With the right set of position, assets, and opportunity, it's entirely possible to create an impregnable defence. Perhaps a stronger fleet could kill it, or perhaps you'd have a fair shot if the enemy wasn't camping there with the right units. In any case, the AI's currently fleet clearly is insufficient to take down this defensive position. This might be a situation where the AI should bypass and start causing havoc on undefended inner worlds.
Better Balanced Expenditures
The AI is very bad about allocating its funds. It has a bad habit of buying a capital ship slot and then not using it for over fifteen minutes. Or worse yet it will research trade ports near the start of the game and therefore open itself to a rush. The worst offence, however, are those ridiculously heavily upgraded starbases. There's nothing wrong with starbases or trade ports or extra capital ships slots, but if you put your money in the wrong place at the wrong time, you're going to lose, and quickly.
What we need is an AI that has a better sense of balance between its fleet, economy, and defence. This needs to consider its short-term and long-term needs. A starbase is well and nice, but it needs to be placed well, especially if it's going to be heavily upgraded. This needs attention in particular, because if the AI invests a huge sum of cash in a starbase and compromises its fleet, there are many options I have to avoid the starbase and render that massive investment quite modest.
Surrender, Gracefully
The current surrender behaviour is a problem. Well known is the fact that it surrenders on a hair-trigger in the late game, often throwing in the towel as soon a player fleet arrives on its doorstep looking for that climatic battle. Less well known is that if you rush the AI, it will almost never surrender. In fact, it will often force you to bombard every last planet it controls, long after its fleet and all its factories have been annihilated.
The former problem (surrender on a hair-trigger) isn't due to the fact that the AI is giving up too early. Rather, it's due to the AI giving up too late. I've played several long games against the AI and subsequently reviewed the replays to get a better understanding of its surrender behaviour. I've also paid close attention to their empire size, fleet size, technology level, and economy (the big four). My findings are not surprising: the AI's defeat was a forgone conclusion long before they actually surrendered. Usually they were at a significant disadvantage over an hour before their eventual defeat, and their position had only gotten worse since.
In other words, the AI has been significantly disadvantaged for a prolonged period of time, and the straw that breaks the camel's back is usually a massive fleet knocking on its door. We shouldn't complain that the AI drops dead when we come knocking; rather we should complain that this silly situation occurred in the first place. What I propose, therefore, is a graduated surrender behaviour.
The AI should detect a growing disadvantage, rather than suddenly throwing in the towel because it runs into a fleet three times bigger than its own. The big ones to watch for are fleet size, economy, number of planets, and technology level (roughly in that order in terms of importance). As its disadvantage grows, it should become more desperate (desperate != suicidal, as I have already discussed). Depending on its situation, this may mean trying to prop up a competing economy even if its means sacrificing its own defences, or else an all-out attack before things fall apart for it. This desperation should increase over time, and culminate with the offer of a surrender.
Unfortunately, I know there will always be some players who play against low-difficulty levels and will be indignant when the AI (which has just failed for the twentieth time to break a defensive choke point) surrenders. Some players are just in it for blood, or else are still learning the ropes and often need to try out massive super-fleets against weaker opponents to get a sense of the relative strengths of units. As a result, a player should be able to reject a surrender offer.
The second issue with surrenders, which is less well known, is that the AI often hangs out too long in smaller early-game scenarios. I'm talking to the point at which it has a single dead asteroid and no units and it still won't surrender. The problem is clearly that the surrender logic doesn't scale well. For some reason my early-game fleet isn't “threatening” enough to convince the AI to surrender, even though it literally has no ability to fight back.
That's a long writeup, but it needed to be said. In a game like Sins, surrender has a very important role. We have many economic assets in this game, but our ability to fight back is usually compromised if only a small portion of these are taken out of the picture. There is no reason to drag out the game (potentially for hours) to clear this up. A desperate counter-attack is a reasonable precursor to surrender in these cases. If someone wants to beat the AI to a complete pulp, let them reject the surrender, but for the rest of us give us a “desperation mode” AI that goes through its death throes then surrenders without making us trudge through its territory long after it's lost the ability to suitably threaten us.
We'd still get our asses kicked by unfair AI sometimes 
I know quite a few posters here can play 1v3-4+ against the hardest AI difficulty... I want to see some replays! I wouldn't mind some pointers lol!