Hacker News Reader

Comment by rbetts | Hacker News Reader

> A number of unknowns make this tech a questionable solution for its stated purpose. We have no idea how many hot list criminals were scanned and not matched

I don't think the article's author had sufficient information to make any claims about success rates. Counting "not identified" as an error is wrong if the system only expects to identify hot-list faces - which is also more protective of privacy for those not on a list.

View on HN

There were five alerts generated due to a positive match, and five of them were false positives. We have no idea how many of the ~500 individuals on the watch list were scanned and missed, and we have no idea how many faces were scanned, but it's still a 100 percent failure rate.

You are all demonstrating you don't know how FR works, nor the article author. FR is a technology that helps to identify, it is not "the authority" and was never intended as such - but hard explaining that to any journalist, or people predisposed by fear journalism. When an FR system scans people and produces "false positives" - that is what you want. The system is tuned sensitive to generate false positives on purpose, so a human can then make the final determination. The point being, a human cannot look at thousands of faces and remain objective, but they can look at a dozen and tell if they have the person they are seeking. The FR system does the labor of throwing away the obvious non-matches, and shows the operator possible matches for final confirmation. That is their design. FR systems are not tuned the other way, because you are trying to help find someone, not be the authority on finding them. Such a system will never know their "false negatives" when dealing with the public - but comparing their ROC Curves ("false negatives", "false positives", "true negatives" and "true positives") is exactly how they are measured and compared when selecting which FR system is better. The key take away is: they are not intended to be an autonomous agent, they are used by a human to help scan more people than they can alone, and it is the human who makes the final identification.

>You are all demonstrating you don't know how FR works, nor the article author. FR is a technology that helps to identify, it is not "the authority" and was never intended as such - but hard explaining that to any journalist, or people predisposed by fear journalism.

I'm not afraid of facial recognition. I fear what "the authority" can do with such a system, both intentionally and accidentally. High false positive rates may be necessary to get the best facial recognition, but they're also a reason to oppose facial recognition.

Reply on news.ycombinator.com