The microphones are in the houses 24/7, connected to power and the internet. That is creepy.
There were also several creepy situations involving these devices:
* Google home had a hardware flaw which actually made it record 24/7.
* all assistants activate by mistake when they think they recognize the activation word. Siriously.
* An Amazon Echo recorded a private conversation and sent it to a random contact of the owner
Even current smartphones do not turn on their microphones outside of a call, unless one has made the huge mistake to turn their assistants on, or there's an app recording - and at least on iOS that's controllable by permissions and marked clearly when the app is in the background.
Anyway, smartphones are creepy too.
One cannot know with absolute certainty, only with reasonable certainty.
On the other hand, an assistant is listening all the time by design.
If one needs to know with absolute certainty, then they cannot use a mobile phone, nor probably a reasonably modern regular phone.
These days, I'm not sure at all. All we have is the law as a deterrent, and that's weak, especially to governments.
ISTR there are documented instances of law enforcement and/or intelligence doing that sort of thing to phones, and the battery drain causing it to get noticed.
Hooking it up to an outlet is a legit different thing.
If battery tech or power reduction tech progresses sufficiently, including just by continuing on its current curve (most power reduction rather than battery tech), cell phones will rise to the same level of concerning.
(Cell phones of course leak location data full time, and some similar things like that. I'm not saying the current cell situation is awesome either. Only that there is a legit reason to be more concerned about gear hooked up to actual power lines than portable gear.)
Google's actual smartphone line (Pixel) ships with passive song recognition, default disabled. I enabled it because I was curious and it sips so little power that I left it on, but to function it has to be listening to the audio all the time so that it can say to itself "Hey, that is Sabotage by the Beastie Boys".
It doesn't use the network, or the screen (unless you power the screen on to find out what you're listening to) but it passively remembers every recognisable song it heard, essentially forever. So that means it is _listening_ it just doesn't tell anybody about it.
We have not historically been there, which informs people's about cell phones, and we are not quite there yet. 4 or 5 years maybe.
(Although I have some interesting thoughts on how to deal with that. It would be fun to take one of these "transcribing phones" into a movie theatre, then fire the MPAA at Google for copyright infringement.)
Does it in some sense "wake up" to do this? Yes, I assume so.
We know that modern phones fan constantly listen to the microphone with minimal power impact, since that’s what these voice assistants are doing. That’s specialized hardware, but adding a little something to save the data wouldn’t be hard.
Even if we take it as a given that it can’t be done on batter power, phones get connected to outlets pretty regularly. Have you ever heard someone refuse to charge their phone in their house because it’s creepy?
I don't know that an old-school land line phone isn't always listening. But it was made by some disinterested party who just makes hardware. It was a simple logical device designed to rely on a simple mechanical switch. It could be listening, but it's understandable by someone like me and there's no incentive for anyone involved. I don't really know if I'm in control, but it's reasonable to think I am.
I don't know that a modern voice assistant isn't always listening either. It is an incredibly complex device usually designed by an entity whose business makes hundreds of millions of dollars based off of the inspection of the personal data of people like me. It is a complex full functioning computer. It uses a statistical algorithm that I know will incorrectly activate sometimes. I cannot understand or predict when it will listen. I cannot understand or predict when it will send a recording somewhere. It receives updates frequently. Because it's a full computer with an internet connection, those updates could make it do a huge variety of things. I know I'm not in control.
Even something like an Apple Watch is the same thing with the addition of reading your vitals.
This is not the same as listening, never mind always listening. A home assistant must always listen in order to recognize the activation words.
This is a reasonable approach for someone whose thread model allows them to use a smartphone...
The people making those complaints are probably not crazy; it might be a good idea to try to understand their actual concerns instead of jumping to this inaccurate (and insulting) misrepresentation.
> they were streaming constant recordings
That would be a waste of bandwidth. Obviously a constant stream isn't happening. While a trivial noise gate and high-efficiency encoding could easily[1] allow almost constant recording of voice for trivial amounts of bandwidth, that isn't really the concern.
> constantly listening
It has to be constantly listening, or it wouldn't hear the wake word. However, this too isn't really the concern.
> in a targeted, continual sense
Dan Geer, on "targeting"[2]:
>> There is no mechanistic difference whatsoever between personalization and targeting save for the intent of the analyst. To believe otherwise is to believe in the Tooth Fairy. To not care is to abandon your duty.
Targeted surveillance is by definition rare and often hard to avoid. I'm far more concerned about non-targeted, sparsely-gathered data that surveillance capitalists like Google/FB/etc can use to refine their model of your pattern-of-life[3]. While this touches on a few important issues, yet again this isn't why I consider always-on, remotely-controlled microphones to be a serious problem.
The big problem is that you're normalizing the idea that remote 3rd parties might be recording in areas traditionally considered private. Kyllo v United States[4] created a bright-line rule that asks if a technology (in general, not any specific product) is considered to be "in general public use"[5]. If it's not, using that technology is considered a "search" for 4th Amendment purposes, and a warrant is required if the police want to use it. When a technology is "in general public use"[5], the police no longer need a warrant to use their own voice-activated internet-controlled microphones to see the "details of a private home that would previously have been unknowable without physical intrusion"[6].
I don't care what these devices may or may not be recording right now. I do care that normalizing of the idea that voice might be recorded inside the home by a remotely-managed device is eroding my future 4th Amendment rights even if I never buy an Alexa/etc.
[1] https://news.ycombinator.com/item?id=16346416
[2] http://geer.tinho.net/geer.source.27iv17.txt
[3] https://en.wikipedia.org/wiki/Pattern-of-life_analysis
[4] https://caselaw.findlaw.com/us-supreme-court/533/27.html
[5] used multiple times in ruling[4] (esp. section II of Justice Stevens' dissent)
[6] see [4], 2nd paragraph
That’s not a conspiracy theory, and it is practically indistinguishable from constant listening for the normal person.
Source?
This seems like a distorted representation, to support your argument, of what most the complaints and responses were. The gulf between 24/7 recording and "occasionally unintentional sending" is absolutely massive and you just merge them into "see, we were right".
The complaints were (and still are) conspiracy theories that they were streaming constant recordings and constantly listening; which they aren't; almost nobody is pretending this would be okay.
I've also not seen anyone who knows what they are talking about seriously claiming that /no person/ would ever listen to the audio samples - only that there isn't somebody _listening_ to your samples e.g. in a targeted, continual sense.
In fact the only part that does sound slightly worrying is that the evaluators get location information along with audio; this would probably be enough to reverse any ID anonymisation, and if the metadata is kept - requests by law enforcement, though I would hope the contractors couldn't control/request specific samples coming into them. In any case, Alexa is potentially worse than this because I can see a history and playback old recordings on my account, which means they are kept fully associated. Apple might do this with Siri internally, I don't know.