Since Nuremberg, the individual decision to refuse to carry out orders contrary to national or international law has been a vital bulwark against atrocities of all types, including nuclear war.
Stanislov Petrov saved the world by refusing to follow procedure, because he (correctly) believed the data to be in error.
Tiananmen Square didn't repeat itself in Tahrir only because the commander of the theatre operations refused a direct order from President Mubarak to use tanks to retake the square.
After the refusal, the order was relayed directly to the tanks within and among the square. Officers were calling their fathers, brothers, mentors, even one who called his former teacher, an American NDU Professor, all asking what they should do. Eventually, the armed forces defected, and Mubarak fell.
Robotic soldiers (AGVs) have no such fail safe. They have no conscience, no larger sense of purpose, no allegiance to King and country. They can be configured to kill whatever they are told, without even a moment's hesitation.
Having soldiers that can coldly eradicate populations serves as the perfect tool of the autocrat, and is the perfect foil of the volunteer warrior from a democracy.
One risks nothing, benefitting himself, the other risks everything, benefitting others.
That said, robots are merely tools: physical extensions made by humans, tied to mathematical algorithms written by humans, executed on silicon designed by humans. They can have whatever value system we choose to imbue.
The ruse that we shouldn't accommodate is any attempt on the part of the operators to shift the blame for their actions onto the system.
A drone doesn't kill 16 people at a wedding any more than an AKS-74 killed a dozen Ukrainian citizens in Crimea. People control these actions.
It's like we spend all of our time obsessing about the puppets, and never question who's holding the strings.
"Will robot soldiers be good or bad?" isn't the question. The question is, "Are the soldiers controlling the robot good or bad?"
One huge issue foreign policy people have with drone strikes is how "uneven" they are. War is a horrible thing and when it can be prosecuted with less risk, it seems more likely. And while robot armies fighting each other might seem like a "good" thing, the people dying most often in Syria and Iraq are not soldiers on either side, they are just trying to live out their days.
Arming mechanical devices to kill people is never a good thing.
E.g. in ww2-with-robots people would have been killed but the nanking massacre would not have happened.
If you consider that battle frenzy is likely to de-humanize soldiers much more than "setup camp and hold ground" operations, it seems likely that the number of war crimes goes down even more, once you remove humans from active battle.
Of course, I cannot exclude rampaging robots taking revenge on puny humans, but that seems unlikely.
It just seems to me that we already have "robot" soldiers -- tanks, plans, bombs, artillery. When the purpose is to terrorize and demoralize an enemy's population, they're not going to have robots doing it.
Why not do this (or just use only robots). For the same reason when you read "videogames for real wars" and instantly see how stupid is it.
You should watch the documentary series "Terminator".
Why do you say "thankfully"? Isn't it better to have robot armies face each other than human armies? Wars will probably continue to happen no matter what.