Instead of guessing, you people need to learn to use Wireshark and find out for yourself.
No, they don’t just listen all the time with an open mic and just send all audio to the cloud. Anyone in cybersecurity would definitely notice that and sound the alarm. There’s probably tens of thousands of people watching what these companies and their tech do all day long.
They can get all the data they need through other means, like trackers. Most of us aren’t consciously aware of the metric shitton of bread crumbs we all leave behind on the net.
Wireshark may or may not help you here. The proposed mechanism is abusing the wake words, which are processed locally on the device. Each marketing wake word could be processed, set a flag and go back to sleep with no network activity. Periodically a bit array of flags would be sent to the server with any other regular traffic (checking for notifications, perhaps). The actual audio never gets sent. I’m not saying that Facebook actually does this, but it’s a reasonable explanation for the behaviour seen in the Vice article.
No, they don’t just listen all the time with an open mic and just send all audio to the cloud. Anyone in cybersecurity would definitely notice that and sound the alarm.
How would they though? The mic is already known to be always on, and what the servers/back-end are doing with the mic input data is not viewable/known by us on the outside. So how would those ‘cybersecurity’ people know?
If you’re monitoring the traffic, and you start speaking, and you suddenly see packets spewing out of a device every time you talk, that’s a good indication. There’s indirect methods to analyze it without necessarily being able to see the actual data.
Poking around the PCB with an oscilloscope to see electrical signals will probably be useful too.
If you’re monitoring the traffic, and you start speaking, and you suddenly see pac6kers spewing out of a device every time you talk, that’s a good indication. There’s indirect methods to analyze it without necessarily being able to see the actual data.
Its already established that the mic always hot, and that data is always being sent to the server.
What they do with the data is not seeable by us. That is the point being discussed, do they listen in to conversations and market off of that data to us.
I would only want the phone to listen when I actually ask it a question, not 24/7.
If the phone does not listen 24/7, then how does it know when you are asking a question? It should discard all information until the wake up word is called in theory.
Only way it could work if you have to press a button to start listening to your question. This was the case in the past, however people wanted to ask questions while showering or something since they introduced this “improvement”.
If the phone does not listen 24/7, then how does it know when you are asking a question?
I pushed the microphone button on the keyboard editor when I want the microphone to listen to me.
For example, when I comment here on Lemmy, I use the voice-to-text option to type out my comments, via the microphone.
It should discard all information until the wake up word is called in theory.
But even with always-on listening mode, it shouldn’t actually be taking any of your data for advertising (or legal issues for that matter) and using it, unless you explicitly authorize it to do so.
And it has to be very explicit, not buried down in some long multi-page license somewhere that only a knowledgeable lawyer would be able to know and find.
Oh, and you should be able to opt-out of that mode as well.
It most likely is. And if it’s not your phone, then it’s your car (assuming it has been built in the last few years)
Instead of guessing, you people need to learn to use Wireshark and find out for yourself.
No, they don’t just listen all the time with an open mic and just send all audio to the cloud. Anyone in cybersecurity would definitely notice that and sound the alarm. There’s probably tens of thousands of people watching what these companies and their tech do all day long.
They can get all the data they need through other means, like trackers. Most of us aren’t consciously aware of the metric shitton of bread crumbs we all leave behind on the net.
Wireshark may or may not help you here. The proposed mechanism is abusing the wake words, which are processed locally on the device. Each marketing wake word could be processed, set a flag and go back to sleep with no network activity. Periodically a bit array of flags would be sent to the server with any other regular traffic (checking for notifications, perhaps). The actual audio never gets sent. I’m not saying that Facebook actually does this, but it’s a reasonable explanation for the behaviour seen in the Vice article.
How would they though? The mic is already known to be always on, and what the servers/back-end are doing with the mic input data is not viewable/known by us on the outside. So how would those ‘cybersecurity’ people know?
Anti Commercial-AI license (CC BY-NC-SA 4.0)
If you’re monitoring the traffic, and you start speaking, and you suddenly see packets spewing out of a device every time you talk, that’s a good indication. There’s indirect methods to analyze it without necessarily being able to see the actual data.
Poking around the PCB with an oscilloscope to see electrical signals will probably be useful too.
Its already established that the mic always hot, and that data is always being sent to the server.
What they do with the data is not seeable by us. That is the point being discussed, do they listen in to conversations and market off of that data to us.
Anti Commercial-AI license (CC BY-NC-SA 4.0)
Tell me, how have you established this? What were your methods?
By calling out for the Google assist, without having pushed any button first. It’s always listening for the activate/initiate key phrase.
Anti Commercial-AI license (CC BY-NC-SA 4.0)
That doesn’t mean it’s sending anything out through the network connection. The wake word is locally processed.
Doesn’t mean it’s not, either.
This old article from Vice seems to have proven it, back in 2018.
Anti Commercial-AI license (CC BY-NC-SA 4.0)
Yeah, that’s really creepy. I would only want the phone to listen when I actually ask it a question, not 24/7.
Anti Commercial-AI license (CC BY-NC-SA 4.0)
If the phone does not listen 24/7, then how does it know when you are asking a question? It should discard all information until the wake up word is called in theory. Only way it could work if you have to press a button to start listening to your question. This was the case in the past, however people wanted to ask questions while showering or something since they introduced this “improvement”.
I pushed the microphone button on the keyboard editor when I want the microphone to listen to me.
For example, when I comment here on Lemmy, I use the voice-to-text option to type out my comments, via the microphone.
But even with always-on listening mode, it shouldn’t actually be taking any of your data for advertising (or legal issues for that matter) and using it, unless you explicitly authorize it to do so.
And it has to be very explicit, not buried down in some long multi-page license somewhere that only a knowledgeable lawyer would be able to know and find.
Oh, and you should be able to opt-out of that mode as well.
Anti Commercial-AI license (CC BY-NC-SA 4.0)