Content recognition, and contextual/cultural analysis is extremely difficult for a computer to do on its own. Sometimes you really have to have someone fluent or from the culture in question to pick up on what's really being said.
But shifting all of the translation and some analysis work to a computer, even if it was possible (which it isn't, at least not yet) would be a nightmare for everyone involved. It'd just add another layer of Male Bovine Feces to work through but one that's completely unresponsive. I can tell you with near 100% certainty that it would also seriously piss off the voice and non-voice collectors who are hanging their asses out in some very hot places a lot of times to get the linguists and analysts their product. You have to be able to talk to the analysts if they or you have questions, and you can't do that with some dumb-ass computer. And I can see that kind of extra bullshit getting people killed.