Lafiel
Yes, sometimes you want your ship to be pincered. It's the only way something like the Twincruiser will bring all its guns to bear.
The most important thing for AI development is to abandon the notion that we can have a one size fits all AI that is in any way satisfactory.
Some ships want to put the front towards the enemy and care nothing for their target's orientation. Some ships have very wide or distributed fire arcs and don't care what facing they point at the enemy as long as they're at the right distance and not staying in line with its potential spinal railguns. Some ships are fast and reliant on shields or weapons that reload slowly enough they don't want to hang out and be shot at while they're reloading and want to pick a range, draw a tangent to a circle around the target at that range, and do flybys.
Targeting should probably usually be random. Any fixed scheme can be exploited. You can dangle reactors that look exposed buy are heavily shielded to bait the AI into terrible targeting decisions. You can make a ship that rapidly spins or fans back and forth so that the predictions for any component not at the center of mass get thrown way off. You can make a ship with massive central defenses and exposed wings. Sometimes the least buried reactor is buried enough that the fastest way to reduce the danger an opponent poses is to pick off its weapons. Identifying which is the case is too hard to be practical, but you can always have the ship randomize its target prioritization rubric when spawned or at sufficiently long intervals that if the previous rubric was working the battle should be won by now.
Targeting priority randomization should be able to be overridden by a "shoot where the railguns or fixed axis ion beams are pointing" option and electrobolts and EMP missiles should probably ideally independently target unoccluded powered components when not overridden by a designer dictate to shoot where the railguns of fixed ion beams are pointing.
