• Posts
  • RSS
  • ◂◂RSS
  • Contact

  • Essentialness of Data

    July 13th, 2021
    privacy, tech  [html]
    In discussing online privacy, people will sometimes say things like "if the data isn't required to operate, you don't need it." If I can turn on tracking protection, stop sending data, and nothing breaks, then clearly it wasn't needed, right? But consider these three cases:

    • Learning from implicit feedback: dictation software can operate without learning what corrections people make, or a search engine can operate without learning what links people click on, but the overall quality will be lower. Each individual piece of information isn't required, but the feedback loop allows building a substantially better product.

    • Incremental rollouts: when you make changes to software that operates in complex environments it can be very difficult to ensure that it operates correctly through testing alone. Incremental rollouts, with telemetry to verify that there are no regressions or that relevant bugs have been fixed, produces better software. Even Firefox collects telemetry by default.

    • Ads: most websites are able to offer their writing for free, without a paywall, because they can get paid for showing ads. Collecting more data makes ads more efficient, which makes them more profitable for the sites, which translates into more competition to provide users with things to read. (more)

    Instead of pushing for "don't collect data", I think it would make a lot more sense for advocates to push for "only collect data privately" and work to make that easier (carrot) or mandatory (stick). None of these uses require individual level data, they're just easiest to implement by sending all of the data back to a central server and processing it there.

    (What does "private" mean? Ideally it means that no one reviewing the data can tell what your, or any other individual's, contribution was. This is formalized as differential privacy, and is typically implemented by adding noise proportional to the maximum contribution any individual could have. In some cases k-anonymity may also provide good protection, but it's trickier. And this is only the beginning; privacy researchers and engineers have been putting a lot of work into this space.)

    Comment via: facebook, lesswrong

    Recent posts on blogs I like:

    What I've been doing instead of writing

    I’ve been too busy with work to write much recently, but in lieu of that, here’s a batch of links to other stuff I’ve been doing elsewhere. The thing I’m most excited about: Wave raises $200m from Sequoia, Stripe, Founders Fund and Ribbit at a $1.7b valuati…

    via benkuhn.net September 11, 2021

    Some reasons to measure

    A question I get asked with some frequency is: why bother measuring X, why not build something instead? More bluntly, in a recent conversation with a newsletter author, his comment on some future measurement projects I wanted to do (in the same vein as ot…

    via Posts on Dan Luu August 27, 2021

    Songs about terrible relationships

    [Spoilers for several old musicals.] TV Tropes lists dozens of examples of the “I want” song (where the hero of a musical sings about their dream of escaping their small surroundings). After watching a bunch of musicals on maternity leave, I’m wondering h…

    via The whole sky July 17, 2021

    more     (via openring)


  • Posts
  • RSS
  • ◂◂RSS
  • Contact