There are different types of face recognition in smartphones today. One kind is the biometric recognition used for unlocking phones, and can tell people apart. The second is just the recognition that a face exists in a certain location, often used by filters. It seems like they wanted to mix these two kinds, but one is very simplistic and the other is very difficult.
Distinguishing among different people is a much harder problem, and can have a bunch of false positives. The facial recognition systems used by police seem to make the news frequently saying that they mistook one person for another. Those systems will have more processing power than a pair of smart glasses.
That’s assuming that they’re doing the processing locally. If they’re uploading data to AI datacenters for processing, then it’s really hard to justify calling them “smart glasses”. In the computer field, it used to be common for them to use terminals that sent all processing to a large server, and the terminal itself did no processing. They were called “dumb terminals.” It would be weird for smart glasses to actually be dumb terminals.
But anyways, if the processing is done locally, the error rate is going to be high. The glasses would tell you that Steve is over there, but it’s actually Doug. And worst of all, the mistakes it would make, just like the police’s facial recognition system, will inevitably come off as extremely racist.
So, it doesn’t surprise me that they cancelled the feature that would make all of their users be called racists. I’m not sure why they’d be mad about it. It seems like a doomed feature to me.
Based on my understanding:
The glasses would send data to the user’s phone, which would maintain the local end of a bank of faces, and relay that to a central server for full id.
Similar to how Facebook builds profiles on you across the web on connected sites regardless of whether you are logged into Facebook.
They’re already calling LLMs “AI”, calling dumb terminals “smart glasses” doesn’t seem that much of a stretch. And I think they only scrapped it after unfavourable reporting, not because they couldn’t make it work. Not saying they could make it work well and not be racist in the first place, but when did that ever stop them?
Edit: they only scrapped the code from the consumer product, I’m sure they’ll keep developing it. Another report mentions an internal memo about releasing it during turmoil to minimise backlash. They’re going to push it sooner or later unless law is passed to prevent them.
There are different types of face recognition in smartphones today. One kind is the biometric recognition used for unlocking phones, and can tell people apart. The second is just the recognition that a face exists in a certain location, often used by filters. It seems like they wanted to mix these two kinds, but one is very simplistic and the other is very difficult.
Distinguishing among different people is a much harder problem, and can have a bunch of false positives. The facial recognition systems used by police seem to make the news frequently saying that they mistook one person for another. Those systems will have more processing power than a pair of smart glasses.
That’s assuming that they’re doing the processing locally. If they’re uploading data to AI datacenters for processing, then it’s really hard to justify calling them “smart glasses”. In the computer field, it used to be common for them to use terminals that sent all processing to a large server, and the terminal itself did no processing. They were called “dumb terminals.” It would be weird for smart glasses to actually be dumb terminals.
But anyways, if the processing is done locally, the error rate is going to be high. The glasses would tell you that Steve is over there, but it’s actually Doug. And worst of all, the mistakes it would make, just like the police’s facial recognition system, will inevitably come off as extremely racist.
So, it doesn’t surprise me that they cancelled the feature that would make all of their users be called racists. I’m not sure why they’d be mad about it. It seems like a doomed feature to me.
Based on my understanding:
The glasses would send data to the user’s phone, which would maintain the local end of a bank of faces, and relay that to a central server for full id.
Similar to how Facebook builds profiles on you across the web on connected sites regardless of whether you are logged into Facebook.
Smart when compared to glasses, dumb when compared to terminals. It’s not that nonsensical.
They’re already calling LLMs “AI”, calling dumb terminals “smart glasses” doesn’t seem that much of a stretch. And I think they only scrapped it after unfavourable reporting, not because they couldn’t make it work. Not saying they could make it work well and not be racist in the first place, but when did that ever stop them?
Edit: they only scrapped the code from the consumer product, I’m sure they’ll keep developing it. Another report mentions an internal memo about releasing it during turmoil to minimise backlash. They’re going to push it sooner or later unless law is passed to prevent them.