Hacker News Reader

Comment by kevinavery | Hacker News Reader

They say that Face ID falsely recognizes 1 in 1,000,000 people (1 in 50,000 for Touch ID), but it's a lot easier to find the people who look like a given target to unlock their phone. Could be an issue if a mutual friend knows someone who looks enough like you to unlock your phone.

View on HN

Just because someone "looks like" your target to you doesn't mean they're similar enough to fool whatever algorithms the iPhone is using.

Yes. So how much do look-alikes change that number?

It's 1/1M for strangers. Is it 1/10k for family members? 1/100 for close look-alikes?

The difference between FaceID and TouchID is that you don't have pictures of millions of people's fingerprints available publicly online...

Exactly. Someone mining Facebook could build a directory of doppelgängers near each person. A little social engineering could get a bystander to unlock the target's phone without them even realizing they are facilitating a crime.

Ah, HN-level paranoia.

That's what they used to say about Stallman.

Ironically people who look like you to humans tend to be a further away to biometrics.

I've worked on a few facial biometric systems in the past and the false positives were really shocking about 65% of them were cross sex and virtually none of them looked alike.

People tend to look at different flesh centric facial features and things like hair style, eye color and overall superficial appearance is what we see. Biometrics see distance between various points like eyes, nose bridge cheek bones etc we tend to actually overlook those fine features when it comes to doppelgängers.

From friends that still work in the field gate analysis seems to have the lowest rate of false positives yet so maybe IPhone X2 will require you to dance to unlock.

> From friends that still work in the field gate analysis seems to have the lowest rate of false positives yet so maybe IPhone X2 will require you to dance to unlock.

Please sneeze to unlock phone

Fast DNA analysis can be even worse like 1 in 10,000.

It's not clear whether the false positives would mainly be among "similar faces", or it probably could happen due to signature collisions on entirely different faces? Like, a weakness in the algorithm which causes two entirely different faces to hash to a very similar signature.

Reply on news.ycombinator.com