, US UAV X-47B Northrop-Grumman J-UCAS RC bomber 5.jpg
The AI would build an imperfect picture of the world around it, like how human pilots do, but people can make sound judgments based on common sense and experience (difficult to quantify and translate into a computer program), but an AI could just ignore things that are not threats or concerning to the mission. Such as being mindful of areas it is not supposed to enter, the ground or navigational hazards, or threats such as enemy radar, anti-aircraft artillery or missiles. When fighting an aerial dogfight, the AI could be programmed to concentrate on the most pressing concerns in order of importance, such as enemy aircraft, flak and SAMs.
And as for predicting the trajectory of enemy flak, that would be the AI's forte as long as its sensors are effective (and the sensors it is sharing with its network). If the AI indeed thinks and reacts with logical decisions hundreds of times faster than a human can, it could weave through flak and obstructions like an ace in the prosecution of its mission.
And determining who to attack, the AI would be governed by the rules of engagement, just like human pilots are, but the AI might make unfortunate judgment calls.