4 ms·

This seems like a fairly easy thing to test rigorously (in a statistically significant way). Just create a fake account and do a search and see what the effect

by tw1010 8y ago

This seems like a fairly easy thing to test rigorously (in a statistically significant way). Just create a fake account and do a search and see what the effect is on other places on the web. Are there services that does this automatically, so we don't have to rely on anecdotal "I searched for X here and now I see recommendations for that on Y, and I think it was because of X, but maybe it's just coincidence"? A site that rigorously confirms secret data links between e.g. google search or gmail plain text and other places (e.g. youtube or even amazon recommendations). If it doesn't exist, it seems like a cool thing to create, as a sort of service to the world. It'd probably reveal a lot of hidden data links that aren't obvious (or which companies promise doesn't exist).

Cyphase 8y ago
That could turn into an arms race of the target companies trying to detect when it's a fake account versus a real one, and the analysis service trying to make the accounts seem more real. Of course, if it became public that the target companies were trying to dodge the analysis, that could be damaging to them. Then again, automated accounts would likely be against the TOS. The analysis service could work by watching the traffic of real people who download a browser extension, but then privacy from _them_ becomes an issue. Perhaps some kind of distributed data-collection system could be created, where an extension will analyze your own traffic, strip out as much personally identifiable information (PII) as possible / as you'd like, and then submit that to a central repository for aggregation or further collective analysis.
- FloatingVertex 8y ago
  An arms race like that would be a help for privacy, there is no way of filtering out enough bots without a relatively high false positive rate.
  - Cyphase 8y ago
    But "filtering out" would just need to be reducing ad targeting, not suspending the accounts, so some level of false positives might be fine if you can get a low enough false negative rate. The goal of the companies in this hypothetical is to limit systematic statistical detection of ad targeting, not to suspend automated accounts. If a small enough percentage of human users' ads are less/not targeted, no big deal.

by tw1010 8y ago

Cyphase 8y ago

That could turn into an arms race of the target companies trying to detect when it's a fake account versus a real one, and the analysis service trying to make the accounts seem more real. Of course, if it became public that the target companies were trying to dodge the analysis, that could be damaging to them. Then again, automated accounts would likely be against the TOS. The analysis service could work by watching the traffic of real people who download a browser extension, but then privacy from _them_ becomes an issue. Perhaps some kind of distributed data-collection system could be created, where an extension will analyze your own traffic, strip out as much personally identifiable information (PII) as possible / as you'd like, and then submit that to a central repository for aggregation or further collective analysis.

FloatingVertex 8y ago
An arms race like that would be a help for privacy, there is no way of filtering out enough bots without a relatively high false positive rate.
- Cyphase 8y ago
  But "filtering out" would just need to be reducing ad targeting, not suspending the accounts, so some level of false positives might be fine if you can get a low enough false negative rate. The goal of the companies in this hypothetical is to limit systematic statistical detection of ad targeting, not to suspend automated accounts. If a small enough percentage of human users' ads are less/not targeted, no big deal.