Evaluation of Cohort Algorithms for the FLoC API
Google,
Mar 04, 2021
This feels like exchanging one model of ad-tech surveillance for another, and the EFF seems to regard it that way, but Google is touting it's new flock-based Federated Learning of Cohorts (FLoC) ad campaign mechanism as "a privacy preserving mechanism proposed within the Chrome Privacy Sandbox." Their white paper (17 page PDF) describes "different methods for generating cohorts, showing clear trade-offs between privacy and utility." Readers are assigned one or more cohort IDs which are used to deliver ads. Cohorts are composed of people with similar interests, and the use of the cohort ID means individual users are not tracked from website to website. The key question is how to assign individuals to cohorts (in online learning, they simply join courses themselves) and the white paper discusses various mechanisms for doing this. Via Protocol.
It is an interesting idea, though. Imagine a similar approach taken with online courses. People identify themselves as being members of this or that course, and as they visit different websites different resources are offered to them to assist in their learning. Maybe a generalized system to allow all educational institutions to each define their own cohorts, or for groups of individuals to self-define cohorts, called Open FLoCs (I can just see edtech people rushing to the patent office now). I would not be a fan of a Chrome-only approach, though. And while Google can line up 1,000 websites to participate, I'm not sure anyone in the education system has that much clout.
Today: 5 Total: 94 [Direct link] [Share]
