4 ms·
try an external source of true randomness for choosing your test topics. choices that seem random to you may be totally predictable. i know that's wild, but al
by a-dub 4y ago
try an external source of true randomness for choosing your test topics. choices that seem random to you may be totally predictable.
i know that's wild, but also often true. humans are bad at randomness. there may be no direct leak at all of your test topics, they might just be guessable based on everything that is known about you, people like you and things you've been presented or looked at.
- kevincox 4y agoYou also have to factor in confirmation bias-type effects where when you are looking for something everything seems related. If you are seeing dozens of ads a day on Instagram and suddenly you have some "random" topic in your head you will mentally connect them. Maybe this could be counteracted by something like: 1. Generate multiple random topics and only send one across WhatsApp. Count "related" ads for each. 2. For every other random topic don't send ti across WhatsApp and see if you still find "related" ads.
- a-dub 4y agothat would work. i would be utterly shocked if an experiment like that still found related ads. (you'd probably also want a prior on the baseline prevalence of various topics.) assuming that's all true, good news is that they're not doing any totally illegal spying. bad news is, they don't have to. > If you are seeing dozens of ads a day on Instagram and suddenly you have some "random" topic in your head you will mentally connect them. ahh yes. incidentally that sort of behavior is often linked with the onset of poor mental health. there are some interesting questions around the limits of personalization and impacts on mental health. if the machine behaves like magic, specifically on a personal level, does that encourage magical thinking?