Right now, most drones are fighting non-robotic targets, and doing it with an emphasis on stealth (for example, flying so high that forces on the ground can't s

Right now, most drones are fighting non-robotic targets, and doing it with an emphasis on stealth (for example, flying so high that forces on the ground can't see or hear them). This makes latency less of a factor, since the human operators can take their time in selecting targets, and are fairly confident that they won't face any sudden attacks. As drone-on-drone combat becomes more common, you're definitely correct that more decisions will be made by the drones, without human intervention. I can imagine drones having a bedrock set of commands that don't require human approval (such as, "if another aircraft is shooting at you, drop what you're doing and shoot back"). Humans would still provide overall goals, such as "blow up that bunker", but the drones will have increasing autonomy in how that is carried out. There's an interesting parallel to the Mars rovers here, precisely because of the massive latency, even though Mars isn't an actively hostile environment. I don't recall the source, but I remember reading about Curiosity having much more autonomy than previous versions. It's given the same bedrock set of commands ("don't drive off cliffs", "don't smash into large rocks", etc), and human operators give it a similar set of high-level goals ("drive towards that weird outcropping over there"). This lets it explore much more efficiently than driving a few feet, waiting on an Mars-Earth-Mars round-trip for instructions, and so on.