Stephen Schenck / Android Authority
TL;DR
- Anker brand eufy solicited its camera users to submit footage of crime to help train its AI systems.
- Users could get $2 per approved clip, maxing out at $40 per camera.
- If they didn’t have enough crime happening around them, eufy encouraged users to film themselves pretending to be package thieves.
It’s sometimes difficult to remember, especially when we’re confronted with all the incredibly impressive feats it can pull off, that AI, despite the “intelligence” in its name, is still profoundly dumb. At least, it doesn’t have any real understanding of concepts, and only manages to give the impression that it does because it is so very, very good at matching patterns. Getting that good requires a lot of training, which is one of the biggest expenses faced by the companies building these models. And this week we’re learning about a clever way one company looked to crowdsource an effort like that — even if it cut a few corners along the way.
Don’t want to miss the best from Android Authority?
Our story gets started all the way back at the tail end of 2024, when Anker’s camera brand eufy was looking to improve its AI offerings by training them to be better at recognizing bad guys — specifically, package thieves, and people going around trying to open car doors. In its community forums, eufy solicited users for “donations” of videos meeting that criteria, paying $2 for each approved clip (via News). Users could get compensation for up to 10 videos per camera for each of those criminal activities the company was trying to classify — up to $40 a camera.
What were you to do if you weren’t living on the most crime-ridden street on the planet? Fake it! After all, this kind of machine learning is only concerned with what it can see, not your intent, so going through the motions of theft should (in theory) generate the same kind of footage as you’d get filming real criminals. The company even encouraged users to optimize their pantomime displays by positioning themselves to be filmed by two cameras at once:
Don’t worry, you can even create events by pretending to be a thief and donate those events. You can complete this quickly. Maybe one act can be captured by your two outdoor cameras simultaneously, making it efficient and easy.
While we can technically appreciate why this seemed like a reasonable approach, it’s also one that we can imagine could give users some some apprehension — for as close as these faked acts might appear to the real deal, isn’t in conceivable that a system trained on only real footage might have an advantage? Or maybe at least reduce the incidence of false positives?
Right now, though, we’re mostly just curious how well that all worked. Have any of you eufy users noticed better AI theft detection over the course of the past six months or so? Let us know down in the comments.
Thank you for being part of our community. Read our Comment Policy before posting.
