Ted is writing things
I’m about to do something they always tell you not to do when you give presentations, and that’s “speak about myself for a little while”.
At Google, and in the rest of the industry, people mostly used notions like k-anonymity. The idea is to people into buckets of sufficiently many other people that they’re “basically anonymous”.
It wasn’t always k-anonymity. It could be other notions based on an intuitive idea of what anonymous data should look like. The general idea was to try and see what could go wrong with a certain method, and if we couldn’t think of anything, we’d say: it’s probably good enough.
In academia, though, especially among computer scientists, everyone seemed to have converged on another notion: differential privacy. DP, as we like to call it, is not just a new criterion to decide whether data “looks” anonymous enough. Instead, it’s a completely different approach, grounded in math. It tells you: there’s bad news and good news.
A lot of people seemed to have seen this new concept and gone like: this is it. Not only is the math satisfying, but this is the right notion, on a political or even moral level. That’s how we really protect the people in the data.
