Timeline for Do advertisers listen as we talk?
Current License: CC BY-SA 4.0
16 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 28, 2020 at 16:39 | comment | added | E. K. | Sure, a non-accidental activation sent off is not a concern. But that still leaves the accidental ones, where just whatever could be sent off. I don't really see how anyone would not be concerned about that, but then again I'm a bit paranoid. lol | |
| Jul 28, 2020 at 16:37 | comment | added | Tim | @E.T. I see what you mean. The wake word is still always detected on device. I don’t think the wake word being sent off is a concern. | |
| Jul 28, 2020 at 16:37 | comment | added | E. K. | Since I can't edit my original wrong comment, would it make sense to delete it? What's the common etiquette? I guess the discussion is interesting anyway. | |
| Jul 28, 2020 at 16:30 | comment | added | E. K. | @Tim yes but they seem to then send the wake word along to the server anyway if online to analyze potential false positives. So I don't see how that helps much given a notable accidental activation rate, as the article suggests is the case. In any case, my initial comment might not have been correct at least based on the article I linked to support it, but that doesn't necessarily matter. Also, just because that one article doesn't confirm it doesn't mean they aren't sometimes sending random sound bits outside of visible activation too. (While the device has online connectivity, of course.) | |
| Jul 28, 2020 at 15:04 | comment | added | Tim | @E.T. The fact that all my voice activated devices can activate without an internet connection is pretty strong evidence that they locally process the wake word. | |
| Jul 28, 2020 at 15:00 | comment | added | E. K. | @Tim oops sorry, I already deleted it because at least given the article it looks like you might be right. Although until I see it in Apple lawyer writing I still wouldn't trust that personally, but in the end it doesn't really matter anyway. Accidental activations aren't really better than intentionally sampling into the void sometimes, IMHO | |
| Jul 27, 2020 at 11:02 | history | edited | LSerni | CC BY-SA 4.0 |
added 35 characters in body
|
| Jul 27, 2020 at 8:23 | history | edited | LSerni | CC BY-SA 4.0 |
added 574 characters in body
|
| Jul 27, 2020 at 7:38 | comment | added | LSerni | @MartinFürholz I probably explained myself poorly. What you describe is true, but it is what I tried to express with "Finally, your "propensity" for a given item might depend on some complex demographic..." (okay, not so complex after all, I suppose) | |
| Jul 27, 2020 at 7:29 | comment | added | Martin Fürholz | I think the "The less obvious" part is a fallacy. If you discuss a product with a range of friends, chances are extremely high that at least some of them will 'google' for the product. And the advertises 'know' the connection between your friend and you through various methods, e.g. you have used the same Wifi, you have sent them an - unrelated - link with a tracker ID, they have visited your Ebay sale page, you have sent them a link to the product in something as 'innocent' as a messenger, which often process the link (to render it in an adapted form), etc. | |
| Jul 27, 2020 at 7:14 | history | edited | LSerni | CC BY-SA 4.0 |
added 2 characters in body
|
| Jul 26, 2020 at 19:48 | comment | added | Tim | @E.T. That’s accidental activations. The wake-word is always detected locally. | |
| Jul 26, 2020 at 15:38 | comment | added | E. K. | @Dave this is sadly not how this works in all cases: theguardian.com/technology/2019/jul/26/… it's sometimes sent off anyway. The companies just would like you to believe it's always on-device, but it's only most of the time. | |
| Jul 25, 2020 at 23:37 | comment | added | Dave | I think it's worth pointing out that the "wake words" for voice-activated devices are checked on-device (which is one reason why they're relatively inflexible, as well as why the devices don't chew your internet bandwidth 24/7). Only if a wake word is identified (correctly or incorrectly) does any audio leave these devices for more intense processing / eavesdropping. This is known about not just because the companies have announced it and researchers have verified it, but also because human audio interpreters working behind the scenes have leaked info about what they hear. | |
| Jul 25, 2020 at 17:04 | history | edited | LSerni | CC BY-SA 4.0 |
added 328 characters in body
|
| Jul 25, 2020 at 13:17 | history | answered | LSerni | CC BY-SA 4.0 |