Or, N robots could integrate and act upon commands from M operators, where M >> N. Decoupling the control logic from the pointy end could allow you to use the (fast, automatic, and expensive) pointy end at a proper speed without a reduction in control accuracy.
The idea of one robot having one (potentially flawed) operator seems exceedingly dated to me. In conjunction with AI and dynamically created battlefield models, we could relegate the human brain into making those decisions that human brain is good at ("that's a terrorist, and that guy is a civilian") and let machines do their job at what they excel at ("let's use drone 17, it has its barrel only 7 degrees away from the target, already moving in the right direction, and can hit it within the next 700 milliseconds").
I think it might be other way around. It's harder to kill if you or your friends are not in danger. And if you can see in all gory details and contemplate what effect your weapon has on the body of an enemy.
We're still in the beginning. Removing the "witnessing the carnage" part could still be accomplished one day. As I mentioned, using our brains just to simulate humans on-site controlling stuff at the battlefield may not be the best use we could come up with. Compared to what I had in mind, current drones are just glorified RC models. Thanks for the interesting link, though. This risk is always there, I guess.