Some years back, in a chapter of my Future Imperfect, I mentioned a scene in an old movie ("The President's Analyst") where it turns out that the bad guys have been tapping every phone in America and suggested that it was now practical to do it, with computers instead of people listening. The main cost of wiretaps in the past was labor, and computers work cheap.
Artificial intelligence is not yet good enough to understand phone calls, but it does not have to. Speech to text software can convert conversations to a machine readable form, keyword searches can spot the tiny minority of conversations of possible interest, and human beings can take it from there. I argued at the time ("my brief for the bad guys") that one could plausibly argue that until a human got involved, no warrant was needed, and at that point the success of the keyword search provided the legal justification to get one.
Which raises an obvious question: How good a prophet was I? Can any of my readers who have followed Snowdon's revelations and their consequences more carefully than I have say how much of my proposal was implemented by the actual bad guys? Were they using speech to text and keyword searches to decide what a human would look at? Did their legal justification include the claim that no warrant was needed as long as no human listened to, or read, the content of the call? Did they argue that, once a call of possible interest was spotted, the successful computer search provided the probable cause needed to justify a human search?