Quote Originally Posted by Cojafoji View Post
suggesting that the military revamp it's ROE, to possibly include a more definitive requirement for recognizing weapons, or perhaps some sort of recognition software that could be loaded into these cameras which would help to define possible weapons more accurately?
I was about to post on the troubles of software recognition, and that got me thinking about the nature of the situation in general: in almost all situations, you're dealing with incomplete information. It's usually possible to create an algorithm that makes the correct decision in every case when given all the relevant information, but by definition when the information is incomplete there will be times when you make an incorrect decision. In terms of software, if you had some sort of image recognition suite that for example highlighted people carrying weapons for the gunner to shoot, it would be a matter of time before that suite made a false positive (or a false negative, allowing a combatant to get away), and then the scandal would be on the people who developed the software. Is the programmer to blame for the deaths of the people caused by the failure of his software?

But in general, people will do the same thing. Actually, the human eye is far, far better at object and pattern recognition than software. The human eye tends to be more reliable at picking out something that looks like a gun. It's still imperfect, though, relying on incomplete information, and to complicate it further the human brain has built-in biases for situations- it's pretty common to distort what you're seeing in order to make it fit with what you expect to see.

In this case, a soldier was looking through a grainy black and white camera at an area where US ground forces were being attacked and saw people carrying what looked like weapons.


Now, once again, a wrong decision is a wrong decision, and those involved rightly bear the blame. But I still don't think we can condemn them entirely because of the simple fact that from a probabilistic standpoint, this sort of misidentification is bound to happen again and again. The only solution is to continually refine imaging technology, always provide as much information as possible to people making decisions- in this case, simply provide a larger, clearer image. But by nature it is impossible to always get all the information and so by definition there will always be mistakes.