Discussing walking robots with a colleague made me realise that military robots are a really bad idea, beside the obvious ethical reasons. I’ll leave to others the discussion about moral issues of building machines that can kill or hurt either autonomously, or at least with great emotional distance, and concentrate on the technical part: this can only go wrong.
Now consider a robot soldier in the field: it will receive data from the its sensor and a data link to its chain of command. Those data sources are both high volume and insecure, the enemy can not only try to tamper the data link, but also send all kinds of noise unto the sensors of the robot. If you only consider the vision system of the robot, this is a system that needs to handle a high volume of complex data: a human eye has a bandwidth of 10 Mbit/second, so one can assume that robotic eyes will be similar or better. A single exploit in one of the vision handling routines, and you have a rogue robot.
Some will argue that the military can afford to write much more robust code than civilian systems, that they can afford much better engineers. I have my doubts, the Stuxnet debacle has shown that critical system are not better designed or protected than consumer systems, and US military drone were shown to send their video feed unencrypted. More importantly, the military tend to consider information (intelligence) and threats in two separated categories, the mere fact of having a robot look at some funny looking picture or hearing some strange audio sequence and then crashing or going mad, inconceivable, that would be something out of Monty Python show. History shows that armies adapt to a threat after they have been whacked by it…