Hacker News Reader

Comment by lumb63 | Hacker News Reader

Gizmodo should be ashamed of themselves. Running an article that says “there’s little to no evidence suggesting that ‘your devices are listening to you’” is obviously and provably false.

Any device that responds to voice commands must be listening. Always. This is obvious.

We also have proof that companies like Amazon do collect the data, store the data, and listen to the data [1].

They try to make this sound like a conspiratorial idea in contrast with debunked ideas. I question their motives when the ideas they call false are so clearly true.

[1]: https://time.com/5568815/amazon-workers-listen-to-alexa/

View on HN

Amazon is listening to voice recordings sent to it explicitly to process. This is much different than the device passively listening to you when it is shut off. And just by the amount of human labor Amazon puts into transcribing the clips should be proof enough at how useless this data would be for automated systems.

The idea that "wake words" equate to your voice being sent off to the internet is pretty easy to disprove. These devices will respond to the wake words even when disconnected from the internet.

> Any device that responds to voice commands must be listening. Always. This is obvious.

Processing sound is required, but recording or sending to a server are not. I think the latter two is what most would be concerned with.

Why would they not?

Generally such assistants work by continuously listening for a 'wake word', the processing for which happens on device and without recording the audio to disk or sending it to a server. After they hear the wake word, they then start sending the recorded audio to a server to recognise the request (usually after playing some audio cue to let you know they are expecting). The reason for this is twofold: firstly, the obvious privacy concerns. Secondly the cost of streaming that much audio continuously and processing it server-side is very high.

If someone were to show evidence that any such products were in fact sending audio back without being triggered by such a wake word (either continuously, triggered by something else covertly, or without giving a cue that it was listening), it would likely be quite a scandal.

It is technically true that devices with wake words are always listening, but that's not the whole truth. In that idle mode they're only listening for the wake word. Only once you've said the wake word does it start recording. This is evidenced by the fact that you can't set a custom wake word on them (you may only select from a small list on which they were trained).

Nobody has provided evidence that they're always recording and sending (only conspiracy theories). It's possible those theories are true, but there's currently no evidence.

Reply on news.ycombinator.com