[bigiain]

Sure, I'm speculating out my ass here (it's what everyone does on The Internet, right?)

Seems to me though, the most recent numbers I've heard for commercial state-of-the-art facial recognition are barely capable of 99% accuracy. I don't know the error rates of fingerprint and retinal scans, but I'd put good money of the combination of passport and fingerprint or passport and retinal scan being several orders of magnitude more accurate that face recognition we have available right now.

(And I probably should have left out the "border areas"comment as part of a different argument - my beef with that is not "how many illegal crossers might you find within 100 miles of a border", but "is it worth reducing the rights of everybody, legal as well as illegal, just because they live/work/travel within 100 miles of a border?" that includes everybody in CA west of a line thru Sacramento, Fresno and Bakersfield.)

[remarkEon]

Why is the accuracy of the technology relevant at all? I think you're assuming that we already have fingerprint and retinal scans of everyone entering, which is quite obviously not true. We might, however, have a rough facial footprint of a known bad actor. I'm fine with this technology being employed in such a manner.

[bigiain]

The accuracy matters (at least it seems to be so to me) because if all you have is "a rough facial footprint of a known bad actor" and you use a technology with a 1% error rate - given that there's probably something like several hundred million airport border crossings a year in the US - _somebody_ is going to have to deal with a million false positives a day, which doesn't seem like a win given I suspect the number of bad actors for which facial features are know but cannot be detected with the in-place passport/fingerprint/retinal crossing system is probably in the single digits per year...

The aggregated individual cost to the 1% false positives - when deployed against a population of several hundred million travellers a year - seems outrageously high to me.

[remarkEon]

Easily solved by simply fingerprinting and retinal scanning the positives and the "unknowns", which is essentially the status quo. Nothing changes except our confidence level that we are actually engaging the right people. The cost, to me, is simply in terms of how expensive implementation would be in terms of dollars.