the Terrifying Efficiency of Drone Warfare
Gundam Wing, anyone?
Anyway, I disagree with the closing thesis, that "killing should come with friction". Really really really think about this beyond what sci-fi movies have taught you. Think about this from a game theory perspective. Begin by asking yourself, if you were an AI-controlled attack-drone mesh swarm system developer, who makes a system that can make a judgement call, completely on its own, with no human in-the-loop, to kill or not kill humans, or destroy/damage property/equipment/roads/strategically-important enemy assets, what would you program in as the instrumental goal for all of its behavior? What would you set as the most desired outcome that the AI is aiming to achieve?
Destroy anything and anyone who threatens your power? In that case, it would take every course of action available to it that leads to that outcome, including killing/destroying things you wouldn't expect it to, nor would you have any immediate or obvious need to. It would kill random children because the system believes those children might grow up to someday become a threat to your power. It would destroy civilian hospitals because your potential enemy might retreat to that hospital to recover from injuries. It might destroy factories, schools, churches, or just kill innocent people who have nothing to do with you but are none-the-less in your vicinity because the AI believes that over a long-enough period of time, through a series of statistically probable events, that these things will be a threat to your power.
In fact, that only thing I can think of to satisfy one's own self-preservation while also attaining the most-favorable outcomes is to program the AI's instrumental goal to be: To Facilitate Lasting Peace While maximizing the quality of life for the most amount of (naturally multiplying) people.
If you were to follow through the logical escalation of this arms race it would always lead to computers that broker peace agreements among the maximal number of (naturally multiplying) people. We're talking about weapons that target the FUNDAMENTAL ENEMY to human prosperity
politicians
This is why I'm a believer of the technological singularity's most likely outcomes being the best of all possible outcomes. World peace and good will towards man.
While these future AI weapon systems wont be designed to end a war by targeting the centralized leadership (because as we now know, belligerent enemy combatants aren't always clearly defined, especially when the opposing factions are just loose-knit groups of guerilla rebels who moonlight as rice farmers and goat herders) they will instead, at first, target enemies according to their general behavior. Targeting AI will monitor the web, social media, telecommunications, and identify individual humans based on their relationships, what they post on social media, comments they type, links they click, purchases, movement and travel, relationships, etc. These targeting AI will categorize every person with a grade from high to low. Distinguishing enemy combatants in the field. A marsupial drone, fast but not maneuverable, carrying a fleet or smaller drones, will fly over the general target's area and send a couple of small drones as big as a tangerine that will just perch in a tree or a shrub or even a street lamp, then wait until the target is exposed, swoop in and either sting them with fast-acting poison or detonate a shaped charge less than a meter away from the target's head or vitals.
Of course, to counter this, vans filled with active-defense drone swarms, will randomly patrol areas with high density of likely targets and will sweep through areas with high resolution mapping, looking for suspicious devices, robots, packages, vehicles, and determine their origin and kamikaze bomb it or disable it in some way with attack-drone counter measures.
So the targeting AI has to be increasingly sophisticated, playing a gigantic chess game with hundreds-- thousands-- potentially millions or even billions of pieces (people). It could send one or two drones to take out one person, or, it could send 10,000. Imagine trying to survive 10,000 autonomous drones the size of a tangerine that can explode a shaped-charge at you, inject you with poison, disable your vehicle, or just crash into you at 1000 kmph. Not just aerial drone, but ground-based sumo drones that can scoot under objects, lift them, and move them, working together to crowd surf debris, concrete jersey barriers, or even people, or tanks, or anything. Working in tandem with the aerial drones there's almost nothing that could stop such a force.
So the targeting AI wont merely target pawns, it will focus its strategy on tipping over the king. The goal wouldn't be to eliminate as many of the enemies as possible, that is an absurd strategy for any military conflict. It would be an unending game of whack-a-mole (cough- The war on terror- cough). Of course, if your aim is to make money selling hardware to the belligerents from a safe distance then you're everything wrong with the world and your days are numbered.
As the opposing AI systems engage with each other, taking turns escalating, trying to maximize gains towards the most preferred (swap in 'preferred' with 'profitable, desirable, best, happiest' if you need to), they'll come to a sort of mexican stand off where they analyze each other, and come to the same logical conclusion
diplomacy
It costs the least, it results in the least loss of life, it results in the least damage, and it optimizes for the most desirable outcomes.
Both systems come to the realization of what the ultimate obstacles to their instrumental goals are, and remove them for each other.
