Singer made some good points. But the problem in realistic situations is usually the time component: Go to war killing 1 innocent now to save 10 innocents over the next 5 years? What makes this devilish is the difficulty in accurately predicting the next 5 years.
Using robots for combats will not improve our accuracy in predicting future outcomes. All it will do is reduce the costs of going to war (both, human/political and financial costs), and further desensitivize us to warfare (we’ve all experienced this from watching TV).
Some good background material here:
Gianmarco Veruggio – Roboethics
Noel Sharkey – Robot Ethics (Part 1)
Ronald Arkin – Robot Ethics (Part 2) (coming up)