Imagine that the government is clever enough that it can come up with a filter that has a 99.99% chance of tagging a non-terrorist communication as non-terrorist, and a 99.99% chance of tagging a terrorist-related communication as terrorist. I think we'd all be impressed.

Further guess that as many as 1 out of 10,000 (0.01%) of all the communications under surveillance are terrorist-related. (I hope that's a gross overestimate.)

If you do the math, out of 100,000,000 communications, you'll have 10,000 genuine terrorist communications, 9999 of which you'll correctly tag as terrorist (the other one will slip through the cracks). Of the 99,990,000 non-terrorist communications, you'll correctly tag 99,980,001 of them as non-terrorist, leaving 9999 as false positives. So because terrorist communications are rare, even this very good filter will produce a false positive for every true positive.

And if that 1 out of 10,000 calls is in fact an overestimate, the number of false positives will outnumber the true positives. If only 1 out of 100,000 calls were terrorist-related, then the false positives would outnumber the true positives by about 10 to 1.

And that's ignoring the manpower issue of who's going to drill down into those 19,998 communications tagged as terrorist by the filter, half of which will be a frustrating waste of time.