While the first thing that comes to mind is a machine that instantly targets and destroys, I wonder if this could be something more methodical. Since "friendly" human lives aren't on the line for the decision maker, these could be used to slow down the process of determining whether or not to use lethal force.
For example, much larger sets of data could be used that just "Looks like a bad guy with a gun and I think he might want to shoot me." With facial recognition, individual enemy combatants could be tracked, and autonomous lethal force only authorized after confirming the target has been actively involved in a prior action.
I'm probably being overly optimistic, but without adrenaline, threat of immediate bodily harm, etc; the option to slow things down and not just react under fire is a new luxury human soldiers aren't (reasonably) afforded.