(Thread) A couple of weeks ago, the #clearviewai reporting by @kashhill shook the world. Now we show how easy it is to download bulk-scale data from Instagram – and how advanced open source #facerec technology already is
@kashhill We downloaded 200k Instagram pictures within a couple of days, from one computer, without an account. We then searched these pictures for candidates for the Swiss federal elections and found a couple of interesting "matches".
@kashhill We used the open source Dlib library for facial recognition (github.com/ageitgey/face_…). It was able to recognizes faces that are partially hidden, in the dark, etc. I would argue the AI behind it is not very different from the one that big players like Amazon use.
@kashhill I personally think this is and remains a big problem. There's not only #clearview out there. Everyone can do it (it took me about 2 weeks for the prototype). It's dangerous. Here's our article: srf.ch/news/schweiz/a… (in German)
Regarding the bulk-scale download, a Facebook spokesman told us: "The scraping of public information is a challenge for all Internet companies because it is technically very difficult to detect.". Of course, we violated the terms of service (which we were also reminded of).
Disclaimer: We've already deleted everything.
