wzamen01
7 HN mentions · mentioned in 7 months, 2013–2023
2:16 · 3.2M views · Dec 10, 2009
A short 2009 amateur video showing an HP MediaSmart webcam whose face tracking follows a white face but fails to track a black face. The uploader presents it as humorous.
HN cites it as the standard example of algorithmic bias in face detection and of how skewed training data produces biased tech.
7 comments from Hacker News
Given the rate of false positives when "cold hits" are used with DNA databases, facial recognition (which is presumably higher entropy than DNA) has the potential to send a lot of innocent people to jail. I also have some concerns of inadvertent "software racism". See for example, this video: this video . Obviously facial recognition isn't inherently incapable of recognizing people of different races, but if it's designed and calibrated on subjects of one race, it won't necessarily work well for subjects of different races. The last thing we want is a facial recognition database that thinks e.g. all Arabs look alike. And of course, there are a ton of ways that it can be thwarted easily and inconspicuously, even if it works very well. Are all surveillance cameras going to be fitted with IR filters to ensure that a bright IR LED necklace doesn't blind them? And how hard is it for a criminal to hire a makeup artist?
I'd like to ignore the privacy aspect. I want to discuss the false positive problem. Is the technology reliable to have a low false positive rate? Probably face recognition is good enough to classify the photograph of your friends in Facebook, and the friends of your friends. It has to choose a person in a set of ~100. What happens when someone is falsely identified? Is the suspect released after a ID check or he is moved to the police station to a full check? Can this be a proxy for racial profiling? There was a horror story about a face recognition software that didn't detect someone with black skin. This can have the reverse problem. this video
Functions are tools for humans to cause the computer to achieve a desired output. If the function isn't generating the desired output, change the function. We can talk all day about how the history of facial recognition has used things like the eye-nose-mouth T-shape or high-contrast contours to lock in on the target, but if at the end of the day the solution works for people with light skin and not people with dark skin, "HP computers are racist" for every meaningful (i.e. user-experienced) definition of the concept. (reference: this video )
My first thought was about the diversity of faces used in the demo and how ten years ago, computers didn't think black people were humans. this video But after that, I was reminded of the paranoia (or not?) around Zoom and that, for an extreme example, the CCP was mining and generating facial fingerprints and social networks using video calls. It seems like this technology is the same concept except put to a useful purpose.
> I can't remember which company it was that launched a camera with face identification features, but that didn't recognize any face that wasn't lilly white like every single engineer that worked at that company It was HP this video