67
votes
Are phones really listening to us at all times?
Had an interesting conversation with my colleagues this morning. We were pretty split whether phones listen to us for advertising or not.
On one hand, we anecdotally see Google news and ad suggestions based on what we say. We know our mics are on at all times for voice assistant and music detection. But we also read online talking about how there is no evidence about the phones recording us. It's hard to trust anything nowadays.
Probably not. The "X ad come up after I talked about it" is a combination of selection bias (do you remember all the times where the google ad had nothing to do with anything you talked about? probably not, because that's a routine experience) and not realizing that you had revealed this information through other means.
The logistics of constantly recording your audio and sending it back the mothership just isn't practical. For one, it's quite taxing computationally, and would be noticeable both in your battery life, and in your data usage. Secondly, it would be impossible to keep this hidden with applications like WireShark or LittleSnitch or deeper router-level traffic analysis programs existing, and these devices being by hundreds of millions of people (if not billions).
Also, not realizing that humans are ridiculously predictable even on seemingly-creepy stuff.
User: "But how can Google know that my girlfriend and I want to have a baby? I swear I didn't search for anything baby-related!"
Google: "Well, I dunno man, maybe it's the fact you're 28, have a stable income and no child yet, not to mention you have been browsing for wedding rings and stopped going to the pharmacy every two weeks to buy condoms; I mean, c'mon, all of this is stuff I just got from your Maps history"
In the land before smartphones ~2003 target made a model based on consumer purchases that predicted when a customer was pregnant and could roughly estimate when the child would be born. They used this model to send targeted ads to mothers to be about baby products.
A man in Minneapolis was furious that target was sending his highschool aged daughter ads for baby products and complained to the manger that they were encouraging his daughter to get pregnant. A week later he apologized because his daughter was pregnant and he didn't know.
I could only imagine the data and information tech companies can extrapolate about me from the vast amount of information I willingly give them if target could predict pregnancy based on vitamins and lotion sales in ~2003.
Here's the full article for those intrested:
https://web.archive.org/web/20210305085130/http://www.nytimes.com/2012/02/19/magazine/shopping-habits.html?pagewanted=6&_r=1&hp
I like to think of this as well in terms of the birthday paradox: https://en.m.wikipedia.org/wiki/Birthday_problem
The take away lesson is that coincidences happen more than we think. It’s because it’s not the odds of any given event that matter, but the odds none of the things we would recognize happen.
It’s a risky cognitive bias. The religious see a message from God, the spiritual from the universe; others start to imagine conspiracies. I’ve read through the comments in this thread, and I think even the skeptics are overestimating how well Google, etc can target ads at you. They’re not so sophisticated, but when they get lucky they make money!
Not in this space or an adjacent space, so coming from pure speculation land, but wondering why raw audio would need to be sent back for purposes of profiling for ads. Google play services might have a local audio processor that sends summaries or fingerprints of some sort back home as diagnositic data. The local processes could look for keywords or key phrases. It wouldn't even have to run all the time. It could sample every x minutes and only stay on if it detects something.
You would notice this from your phone being hot all the time. Processing audio effectively makes my macbook fan start up after a few minutes - if your phone was doing it constantly, you would have very little battery life, and a hot pocket.
Yeah makes sense. Wondering two things: 1) If we assume software is too clunky to efficiently parse and process voice, how does they "hey google/siri" voice command work without causing same issues? is there special hardware on the SOC for this function? if so, what's stopping Google from coding more than the "hey Google" command. ie. why not code keywords/phrases as well 2) perhaps the audio processing you're doing is more computationally expensive than whatever Google is doing. or their swe have figured out clever ways to make the processes more efficient for their hyperfocused application
I'll refer to "Google" for smart speakers, mostly because that's the flavour of speaker I have, but it could be any other company that sees you, the human, as a product.
Smart speakers (and to a lesser extent other hardware) are almost always listening but almost never processing. There is a "wake word" that they are listening for; listening for a specific wake word is much computationally cheaper than constantly processing everything that is being said; the smart speaker itself has very little going on; it is basically just a speaker and a microphone and a bit of code that listens and specifically searches for the wake word.
It is definitely possible to have multiple wake words - my home system has three different sets - and it is certainly possible that google could have many secret wake words. The main question I have is - why would they do that? What do you think that they would gain from having a secret wake word that creates a different, secret behaviour? The information that would be gathered is almost always something that google already has.
Google already knows 95% of what you do online, and 95% of where you go, and 95% of what you purchase, and they're not going to glean the other 5% from trying to listen to banal conversations on a speaker.
Google almost certainly has access to better ways to process audio than I do, but if they had a good way to process audio offline, that would be a feature that they sell, and not a way to spy on people. Processing audio in a natural way is difficult, which is why it is done in the cloud and not in a local device.
Basically, having access to super efficient ways to process, store, and understand audio information would be worth more to Google publicly than it would to use secretly to spy on people to try to ineffectively find the last dregs of information that they don't already know about you.
Or to put it a more succinct way: if you are the sort of person to buy and use a smart speaker, then the company you bought it from already knows everything it needs to know about you, and it doesn't need to secretly spy on you to learn more, because they are already publicly spying on you an incredible amount.
I'd just like to point out they DO sell it, they make a huge deal about the Tensor "AI" chip, a major selling point of pixel phones which they state can do real time language translation on the chip. "...the many ML capabilities it enables, such as advanced speech recognition,[1] real-time language translation, the ability to unblur photographs,[2] and HDR-like frame-by-frame processing for videos" all of those features are processed locally with the Tensor processor.
I don't have an opinion on whether or not they listen in, it's clear they just don't need to, it would be really bad PR and we're leaking so much data every day it'd be almost pointless. However, I did think the claimed capabilities of the Tensor chip are relevant to this discussion and should change our understanding of how the process would have to work.
Allowing efficient language processing without sending large amounts of raw audio data back to an external server by doing all that processing locally, radically changes how we should be investigating the possibility. The "well it's not sending hundreds of mb of data back and forth so clearly it's not doing any language processing." Argument isn't as valid as it was 2 years ago and that's an important consideration.
That is a really important point, and you are correct; I was mostly just running quickly over the top of the issue.
My experience with the Tensor chip has been underwhelming, to the point that I didn't even consider it to really fulfill what I described - a good way to process audio offline - but that could just be what's currently available in a retail space, and it is always become more and more feasible to be doing that parsing on the device, and sending very small and relevant chunks of information back to the mothership. It will also certainly improve in the near future, and that's an important thing to consider.
If I were a three-letter agency and had somehow pwned some criminal's phone, I'd pick specific keywords I wanted to know about (like someone else's name), and set up the phone to start recording using that as the keyword.
Processing real-time is hard, so maybe I'd just record a few minutes and then save it all to disk, then process it later when they're browsing some website and wouldn't be suspicious about the phone getting hot.
It is special hardware that can only recognize that basic phrase. It does technically listen all the time, but if it were the core OS of the phone (that could actually do something more complex with the data), it would drain the phone in a few hours. Instead, this hardware wakes the CPU upon having heard the magic phrase, that will instantly start the listen and actually process it.
In general, having the CPU sleep is the biggest battery life win, that’s why a faster CPU might still be beneficial. Not because you take advantage of it, but because every task will finish a bit sooner, letting the CPU sleep more.
Hmmm, yes my phone has those symptoms.
While I don't discount your phone listening constantly for keywords or the like (similar to how Alexa does), I think you are right in that companies can build your shadow profile in so many other more efficient ways.
For example, my friend was moving, he rented a Uhaul and got an invoice from them in his gmail. Later, when driving through a city on the way to the new house, me following in my car, both using the same address in google maps (he has a Pixel phone), he got redirected because there was a bridge too low for his Uhaul.
He never explicitly told google the size of the truck or that he was even driving it. They still pieced it together based on the data they gathered in the background.
To add to that, ad predictions can come from a lot of different pieces of information. For example a common way is through geolocation, if someone close to you searched up something then you're more likely to get an ad for it.
I've had way too many coincidences for that to be true.
A few weeks ago, my wife and I were watching the last season of Barry, and we randomly had a conversation where we repeated "No ho hank is the best character" a bunch of times, verbally. No Google searches or watching old clips or any kind of related digital trail.
The next day, my Youtube front page was full of "No ho Hank is the best character" titled Mash up videos.
That's way too coincidental.
And no, my YouTube feed did not usually contain that kind of content. It was all woodworking and PC hardware videos previously.
But now, No Ho Hank is mixed in there because I watched one video... damn targeted advertising.
I'd note that almost certainly Google knew you were watching the last season of Barry. But I've never even watched the show, and I know that NoHo Hank is considered the standout character of that season. Is it really surprising that, knowing you watched the last season, Google decided to surface some Barry-related content to you, and that Hank was the primary reference?
Sometimes I’ll be watching a show and we’ll say something like “wow, x actor is so tall, I wonder how tall he is” and I’ll pull out my phone and type half the actors first name before google is suggesting “x actor height”.
It hasn’t been listening, but it “knows” I am streaming this show, and that a lot of people have this query while watching the show/actor, and therefore that might be my query too. It’s just very, very clever guesswork
If your phone was recording you and sending that back to someone for targeted ads you would be able to see the data usage. As far as I’ve seen, numerous people have analyzed outgoing data from phones and have never found any evidence that audio was being streamed to anyone. It would also kill you battery.
I think this depends a lot on when it sends data back and how often it's listening (in theory).
I'm not sold that it is, but I'm not sold that it isn't. I've seen some very hard to explain coincidences, and it does make me wonder if they just piggyback the data with something you'd have to send anyways.
Just from time and ip addresses I can know when you’re home and at work, if you have a vacation home you frequently visit, and to which country and where you holiday. And lots more.
There is a lot that can be figured out with very little.
Oh sure. I live in that world believe me.
It doesn't explain how the hell I started getting cat food ads after talking about cats with some friends over. I do not own a cat. I have not gone to a pet food store. I do not browse for cats. I don't even go down that aisle at the grocery store.
There's one or two other instances like that I can think of which are very hard to explain from any other form of traditional tracking.
Did your friends use your WiFi while they were over?
Nope. Because they weren't over at my house, as I was watching my parents (who also do not have cats) and they don't have the wifi password for that and I know I didn't give it to them because its one of those crazy alpha/numa's I'd have to ask for.
Common WiFi would have been the simplest explanation (e.g., seeing two users connecting from the same IP address, so chances are that they share some of the same interests). It could still be relevant if you've connected to the same WiFi networks on a somewhat regular basis; an association could have already been made, and even rough location data could determine you were both in relative proximity, so attempt some cross-interest targeting.
But I'd lean toward it being a case of the Baader-Meinhof Phenomenon (a.k.a. Recency Bias or Frequency Illusion): it's an uncanny phenomenon that occurs in ways we know isn't related to big tech. The most recent time this happened to me, I was talking to my wife about Tesla cars and how they tended to be rare in our area—possibly due to how cold it gets in the winter—and then saw 3 different Tesla in two days, including one driven by another parent at my son's daycare while I was there picking him up.
I know that I've seen other Tesla, before, but they were non-events because I didn't have a recent discussion with my wife about how they seemed to be uncommon. Likewise, you could have been shown cat food advertisements before, but they were non-events because you didn't recently discuss cats with a friend, and our brains regularly filter out all kinds of things if they don't have any recency relevance.
All i can say is that I'm very aware of all of this, and that's why I chose this example because i'm about as certain as I can be that it's not recency bias and it's not common wifi.
That still doesn't mean they're sending data (could literally just be coincidence that some cat ad decided to target me due to some new campaign), but it's one of the stronger examples I've experienced that has made me wonder.
And there is also the fact that outside of it probably being illegal, it is something most companies would love to do. There's other discussions in this topic about how much data it would be and what not, but the simple answer is that it's insanely valuable and would easily pay for itself. There was a time where the data might have been "too much" to get reasonable use from it, but that's long over.
Your friend searched for cats after the talk.
I have a lot to say about this topic that's very polar opposite to the general thread, but I think instead of trying to distill a bunch of jumbled anecdotes together, I'd like to simply refute the notion that the processing power and sheer size of data is too much for this kind of data mining to happen.
Most phones today quite literally have an 'always listening' capability, which is how any of the 'Hey Google' or 'Hey Siri' hands free phone accessibility options work. This has been available as [edit: recently as] at least 2015 and probably earlier. Apps also have plenty of cached data that takes hundreds of MB's of storage that they're updating all the time. I'd like to pose the following example conversation:
This conversation is 300-400B. Not kb's. Bytes. This is .001 percent of a kb. The phone does not need to process any of this data in-house, it just needs to find any and all time to beam 400 bytes of data up to Google.
Google then has loads of information to pick from that: The name of a couple, their baby, and favorite restaurant. All of this is 100% possible, and has been for some time, and it absolutely irks me when people try to make this out to something that it is impossible.
I have plenty of personal anecdotes, friends in advertising, and my personal sub-par technical knowledge I could use to rant for days about the fact that yes, you are being listened to. I don't want to get bogged down in arguments I don't have the energy to go back and forth on. I will simply say, with the utmost certainty that the phone in your pocket has all the capability and more (even older models) to send telemetry data to a third party, and that party (with a whole heck of a lot of incentive to sell that data for money) influences the ads you see.
But it's not the storage or transfer space that's the issue. It's the on-device text-to-speech processing which is currently stopping this.
Even current-gen voice recognition is just not that good to be able to work in a pocket and pick up ambient conversations. Google assistant can barely understand me when I speak deliberately and directly to it. When someone pocket-dials you, you don't get a nice, clear audio signal of what's going on in the room they're in, you get a load of muffled noise. Same.
Secondly, even if all the conversations going on around the phone were clear enough to be processed to speech, that processing is going to destroy battery life. If I accidentally turn live captioning on, where my phone attempts to caption all speech it's playing, battery life plummets. For my phone to be converting all audio, all day, into text is just not possible.
Just for reference my current phone is a Pixel 6, which has Google's Tensor chip designed specifically for this kind of processing. And still, it just wouldn't be able to do it 24/7. It's not constantly doing that listening for watch words, that's a much simpler bit of processing being done on dedicated hardware.
For further context, this was actually something that FiveEyes security services did for a while. Capturing all phone conversations going on all day every day (a quantity of data which is dwarfed by all conversations happening near phones). They ditched it in the end because it was just too much. Actual countries didn't have the resources to do a fraction of the processing needed, this is not a thing which can be done in everyone's pockets.
Additionally and most importantly - someone would have noticed. Android is open enough and this belief that we're constantly being "listened" to pervasive enough that someone would have spotted all this processing going on, constantly. From an optics point of view it's WAY too much of a risk for Google to take, getting caught doing something like this. Even if half the people vaguely think it's already happening, actually having it confirmed would be devastating.
There really isn't. Rather the opposite. I guarantee you that Google are NOT selling your data. It's far too valuable to them. They want to be the people who have the best dataset, so the advertisers will go to them. They can sell your "data" once, they can sell access to your eyes via that data (advertisers never see any user data) multiple times every day.
I don't think anyone doubts that it is capable of such activity. More about whether Google/Apple actually bother doing so.
Like, one specific conversation is 400b, but over the course of a day (let's say, a busy workday, 8 hours), that 10 minute conversation becomes 19.2KBs or so. Which becomes 7MB a year. Okay, still not much to consider.
Now consider how many people own an android (since Google is more incentivized to do this). In the US, Google tells me 129m phones in 2020. Okay so... 900GB of data. Still paltry for Google, a consumer can fit that data on a modern hard drive.
But here's the real question: how do we process this and get useful data to sell? We'd need to geotag the data to ensure the proper location (which may or may not be moving, so you need to feed in more than a simple "X was here on Day Y), link the data to the user (so, voice recognition unless you are fine processing 3rd party noise) and their phone number, and then aggregate all that into buckets of data people want (in other words, filtering out a lot of noise and useless info), hoping there's not too much unsalvagable data.
Now could Google do all this? Sure. But on the other hand, that's a lot of effort when Google already gets access to your phone number, contacts, location data, IP address, and much more simply by asking a user to enter that info when buying a phone. All properly sorted out, even less of a data footprint than that soundbite, and all immediately usable. Very few opt out so it's not even worth considering.
I can't say that Google doesn't do this, but I think they are too lazy to bother.
Just out of curiosity, what bitrates and audio quality are you assuming that brings that conversation down to 400 bytes? Or are you assuming that the phone -- which can't understand the wake word a third of the time when I'm facing the microphone head-on -- can do a perfect translation of speech-to-text when it's sitting in my pocket?
400 bytes would be the raw text data. No way they're talking about audio file sizes.
This technology doesn't rely on your phone being able to magically perform speech-to-text 24/7 without destroying your battery life. There is a reason these all rely on usually non-configurable wake words, and that is because it is a lot computationally cheaper to listen for a specific short phrase that then turns on the speech-to-text input.
Though I think the bigger deal here is that Google's (and certainly not Apple's) privacy policy doesn't actually say they do this. While it is always possible that they could be lying, it would be a pretty enormous scandal if either of them got caught. And it wouldn't be particularly difficult for some enterprising security researchers to sniff that out if it were happening.
Suffice to say, that while it is hypothetically possible your phone is recording all your conversations and sending them to advertisers, there is a very clear lack of evidence that this is the case.
Well actually....
Mac OS in the days of rainbow Apple logo already had voice commands. They were simple, but also started with a key phrase that the computer was listening for. For example, "Computer, open Finder", and MacOS would open the Finder file browser for you.
This was in OS8/9 days so late 90s ish?
The tech has been around for a long time.
This only works without drawing a bunch of power because there is a more basic process that is only looking for that phrase and isn't registering all words said.
Rather than totally disagreeing with you, I'd like to offer some plausible alternatives that would function similarly. 1) alternate phrases could be added so that they are registered while still using low power 2) specific apps with mic access could just fully listen whenever possible regardless of power usage (Facebook used to be notoriously hard on your battery) 3) monitoring might only happen intermittently or while charging
The "Reply All" podcast had a pretty interesting episode on this https://gimletmedia.com/shows/reply-all/z3hlwr/109-is-facebook-spying-on-you
tldr: Facebook collects so much data in other ways that they don’t have a need to listen to you through a microphone
I don’t use Facebook, Instagram, Twitter or tiktok. I bet I have other apps on my phone that collect weird data (I do use google maps, for example) but I have never had one of these weird coincidence ads. So yeah, extremely anecdotally I have also always assumed meta et al just know so much about you that they know what you want.
Something interesting is that that kind of prediction, and getting you what you want before you know you want it, is something that used to be prized in servants. But we’re all a little creeped out by it when it isn’t a person doing the predicting, so I wonder how it’ll be when/if we get to a point of having robot servants.
An important difference between a faceless algorithm and an actual person, is that you take for granted that the real person isn't going to hear you as long as you whisper, or move away from them.
A real person can't know what they haven't been given access to.
You can easily limit what others know about by choosing your whens, wheres and hows in life.
This mass data collection, faceless, impersonal thingy crosses that boundry. A lot.
Theory of mind is the name for that thing where most of us will realise at the age of 2-3 years old that other people can only know what they know, but that depends on their own personal perspective. That's when toddlers discover lying. Them trying to fib their way out of trouble, or pretending they're someone else because of a costume etc is a massive cognitive developmental stage.
These mass collection services walk all over the human cognitive boundaries we expect from our surroundings.
I said elsewhere that I understand in a very broad sense that a private company knowing everything about you isn’t ideal, but I don’t really understand why it isn’t or what bad things they can actually do with the data. I can’t tell if I’m stupid and naive or if other people are paranoid, or if it’s somewhere in the middle. Seeing ads I might be interested in vs ads I’m probably not interested in is at worst benign and at best helpful; what is the big worrying downside that I should be thinking about?!
Because it makes it trivial to discriminate, punish, and harm people based on the whims of a few individuals.
On the corporate side-
You can easily decide one day that you're not going to show the best deals to group X. Lets say cat lovers. Evidence you own a cat? Oh ok, well then our targeted marketing system has decided that you wouldn't be interested in deals a/b/c. Or maybe you just can't shop on our storefront. Or perhaps your orders go in a lower priority queue. The vast majority of this would be extremely hard to prove, and most of it wouldn't be illegal even if you could.
On the state side-
The state gets the monopoly on "legal" violence. That's basically what defines a state. Soooo when they decide that it's time to round up all the cat lovers, they're going to go to that corporation, cat lover or not, and let them know that if they don't hand over their database of ALL information on their customers, they'll kill everyone they've ever known. And then they have a nice list of cat lovers, where they live, what other interests they have, and so on.
Frankly, it's despairing how common this attitude is. I don't mean this as an insult, because I think your viewpoint is the vast majority, and it just...blows my mind? I know people who are "cat lovers" that actively distrust various sides of the current political spectrum and are worried that should certain things happen they will be targeted and possibly even harmed. And even they think "well who cares if Amazon knows blah and blah about me. they don't care about me". To which my cynical thought is "yet...they don't care yet."
The entire thing feels like everyone asking why someone wants to build their house with reinforced concrete rather than a beach cabana because they haven't seen a hurricane. It's the kind of thing that by the time it happens, it's too late, and we are WAAAY past the point of too late in my eyes. There's already some very disturbing stuff going on in the "east" (especially china) and the "west" is not nearly as immune to these abuses and woefully ignorant on the dangers given how many current examples there are.
I'm not even going into how you can cobble profiles together or use statistics and digital fingerprints to narrow down even more information than you'd expect from something trivial. This is just first layer "said you liked cats in the survey" stuff. It only gets worse.
It’s not that I don’t understand or have never heard these hypotheticals, I’m just on the (probably naive) side of thinking it’ll never be like that. This comes across as a little paranoid, to me, and not far off tinfoil hat stuff.
I’d be interested to hear some real world examples, though. Maybe I think this because I don’t know about them, or because I don’t live in the US with its SCOTUS-sanctioned discrimination and religious right dictating laws. I’ll be the first to admit I’m probably woefully uninformed (hence me asking in the first place!) but it just all feels a bit like catastrophising.
Well...historically..it's almost always been like that for the overwhelming majority of human history, and seems to regress to that naturally, and even now there's a significant chunk of the population living exactly like that. There's horrible treatment of people all over the world, it's just not as common that they have the tech advantages of more technologically advanced countries.
Still, you have the US, which LONG before the current SCOTUS and religious right's power, was doing such wonderful things as spying on the competition for political reasons, and spying on it's entire population for "whatever the fuck they felt like" reasons, and that's before we even get into the quagmire of state and federal enforcement agencies doing all sorts of shady shit.
The UK is notorious for it's surveillance state issues, making the US look positively liberal in comparison (which is somewhat helped by the theoretical right to privacy in the constitution they sometimes have to dance around).
My favorite china example is simply this story - https://en.wikipedia.org/wiki/Xu_Xiaodong as it's all very easy to confirm,but there's obviously been usage of data to track certain religious minorities, who they are possibly putting in camps and harvesting their organs.
Singapore is another interesting one from a political science standpoint given their extremely high standard of living and their potentially just as extreme brutality when it comes to punishments.
These stories repeat across the world. Data collection is a tool. Tools are not good or bad, but we like to limit the tools so that should someone bad get access to them it's something that can hopefully be stopped or fixed. Should a bad actor get in, they're going to be able to do incredible amounts of harm very very quickly.
Thanks for a really well thought out and informative response! Certainly some stuff I hadn’t considered here. Also some things I don’t agree with, but I can see your point.
I think I land at not being personally concerned, but I can see why I should be concerned on behalf of others and why I might be concerned in the future.
I’ll try and remember to come back when I have time to formulate my thoughts properly! I think disagree would be too strong of a term, more like don’t totally agree that this issue is down to data collection from phones (eg the point about being able to drag up something you said 20 years ago - this is more a function of things on the internet being potentially preserved forever than it is about data collection). But as I said, I’ll try and construct a properly thought out response later on because it is something I want to know more about and am very open to being challenged on!
Facebook heavily uses your location data. Whenever I go anywhere, a few hours later I start getting friend suggestions for people I vaguely know who were in the same area. I barely use Facebook (maybe 10 minutes max per day if at all).
I am pretty sure it's primarily based on location data + proximity over anything else.
They don't. For one it's highly impractical to do so like some other comments have mentioned, but on top of that: why would they? There are far better ways to get information that don't involve literally spying on people, like just buying data from another company. That data is also going to be far more generally useful than the audio from any specific individual.
Let’s put it this way - I don’t think I’m being spied on 24/7, and I generally trust at least Apple because they haven’t had data leaks and their security and privacy audits come back positive, but if some whistleblower revealed that yes, in fact, your phone is recording and spying all the time, I wouldn’t exactly be surprised.
I have to say I haven't actually researched the topic but I'm quite confident they don't. First of all, voice recognition isn't that great at the moment (just look at Siri) and constantly having to have the mic on and analyse speech would take up quite a few computational resources and probably wouldn't even work well. It would also be illegal to listen on you all the time without your consent in a lot of jurisdictions and someone would've leaked it by now anyway.
Instead I think what you might be noticing is the frequency illusion (or Baader-Meinhof phenomenon). Once you talk or learn about something, you start noticing it everywhere. The things you say are typically based on your interests anyway which Google will know based on your search history etc. You'll just pay more attention to the targeted ads after having talked about a specific topic recently.
After talking over WhatsApp to a friend of mine, about how a different friend of line had lost a family member and was going to travel for the funeral, I started getting funeral service ads. That same day.
None of us were physically in the same location. Or even near each other geographically.
I do not usually see ads for funeral related industries whatsoever.
While I don't think anyone listens on to every single word, there seems to be some pretty clear patterns they're able to get from whatever they do collect.
They don't need to listen for that though. It's far more likely that they noticed one of your contacts (perhaps even someone you contact often, though idk the specifics of your friendships) was googling funeral services and other similar topics recently, so they started serving you ads for the same thing.
Nope. As others pointed out, the network traffic from doing that isn't doable in the current arrangement without being noticeable. Imo the truth of it is almost worse - by way of being able to analyse such a vast expanse of data, folks can infer everything about other folks, giving them sometimes a clearer picture of the person than what said person might think even of themselves.
As an example, let's say you and I meet up and you express to me that you're a person who just fuckin loves bicycles. I talk to you, and a day or two later I start seeing ads for bicycles.
The easy thing to think is that the phone was listening. But, by way of monitoring network traffic, we can be pretty confident in saying that didn't happen. So how did they know? Because they can see everything else. Pretending I am such an entity, what I can see, is that your friend and you were at x place at y time, and after that you and your friend looked at content about bicycles, so I took the shot and served you an ad about bicycles.
Its like being able to understand the shape and composition of a stone, through analyzing all available information that came up after that stone was cast into the pond. We can look at the ripples and infer the rest. And the part of that which makes it worse, in my opinion, is that because you can successfully observe in this fashion, you can see the Real Person. As in, I might tell you I am all about x y z, but my purchasing habits may tell a completely different story, and I myself might not be willing to acknowledge that difference. The entities doing the observation, however, always have access to "what actually happened", so to speak. They can act on information even you yourself are unwilling to consider. Thats the shit that really weirds me out, to be honest.
I think your bicycle example can be simplified. From the machine's perspective:
I am thoroughly convinced it is simply based on location data amd interests of any person within extended proximity of another. Then the general assumption is that maybe the thing one person has shown extreme interest in was talked about, therefore send ads about it to any other person whose location data matched.
Edit: This is why the funeral and cats examples elsewhere in the thread don't seem that amazing to me. The machine obviously knows the interests of each person, and knows their social relationship to one another. So even if OG funeral person was never talked to, they have been researching funeral homes, and they are friends with the other two people who had an independent conversation about another topic. The conversation doesn't matter. What does matter is you are all obviously friends, and one of you is arranging a funeral, why not send ads to the friends about funerals? Maybe one of the two friends will see an ad, then recommend it back to the person whonwas researching funeral homes?
Whilst there is little to no evidence to date showing that they are listening and using ML inference to understand key words to improve ad targeting, there is evidence they listen for ultrasonic tones in ads - https://arstechnica.com/information-technology/2017/05/theres-a-spike-in-android-apps-that-covertly-listen-for-inaudible-sounds-in-ads/
In short, no they are not. The experience of having ads arise after a conversation is due to a, perhaps more insidious, ability of advertisers and technology to use your network of friends and acquaintances against you.
If your friend had been shopping for a couch, that information is tied to them digitally. They visit you and you both discuss their new couch. That evening, you see ads for couches. It's natural to think that this is because your conversation was listened in on and data from it scraped and turned into targeted ads.
In actuality this is due to a variety of proximity/foaf (friend of a friend)/networking/location based connections. If your friend and you both had location on, if your friend connected to your wifi, if your friend is in your contacts, or if you are both in the same geographic area then advertisers will connect the two and will serve ads from one of you to the other.
I'm home sick today and a cursory search didn't bring up any great sources for this and I can't hop on databases to pull up any articles, so hopefully someone else can supplement this with a citation, otherwise I will try to edit once I'm back on my feet (and back on an actual computer).
While I understand and even possibly agree that it's unlikely that phones are recording, the number of creepy moments I've encountered where I'm 99.9999% certain I gave absolutely zero indicators that would be accessible or triangulate-able by Meta is so many at this point that I'm not sure I care how they're accomplishing it, but that they're doing it at all.
Here's a fun thing about Meta (I shy away from calling it a fact, but have experienced it several times) - they actually make connections based on people who view your profile, not just on what you are doing. So if you have a person in your network who has done something relevant - say, buy a bed - then they look at your profile, you may get ads related to buying beds.
They also do other things by people looking at your profile. You may think "I wonder what that person I knew in high school is up to" and search for them, but decide not to friend them. Down the line, that friend may get a notification for them to add you as a suggested friend. So never look up anyone on Facebook if you want to hide from them.
The amount of ways that Facebook / Google / Amazon can track information about you is absolutely insane; you may be 99.9999% certain you gave zero indicators, but that little indication came from somewhere, and it can be really insidious. But it can also just be totally based on demographics, and you can win the creepy lottery, but it's just a complete coincidence.
It's at the point that while I am supremely confident that they're not actively spying on you using cameras or microphones, I don't really blame other people for thinking that they do.
The flaw with the idea that phones are listening to us all the time, as opposed to just scraping contextual data, is that there are enough people involved in data collection that it'd be impossible to keep everyone quiet.
There's people out there sharing Pentagon secrets just for clout on Discord, police and government officials are caught out accessing citizen data for frivolous reasons constantly, electronics details are leaked before announcements so often it's just considered to be normal, and finally just look at the important people asking questions at the TikTok and Facebook hearings.
There's no way the kind of people who would benefit from constant listening could keep quiet about it.
I want to add to this, because I think it’s likely the most convincing piece of evidence. Systems like this take a lot of people to build. Folks have been making claims like this for at least a decade. In that time, hundreds of engineers, product managers, designers, etc would have touched these systems. And these are ethically controversial features, at best! There is a fairly robust history of Google engineers speaking out, both internally and publicly, about ethically controversial software built by Google. In just the 18 months I was at Google as an engineer, employees pressured the company to cancel its contract with the Department of Defense, and spoke publicly about and pushed for the end of investigations into building a Chinese government-censored search product for Chinese citizens.
There’s simply no explanation that I can imagine for the idea that Google would be unable to keep these very damaging stories under wraps, but would be able to effectively contain a decades-long conspiracy to build and maintain a global surveillance network, processing absolutely unprecedented amounts of audio data for building shadow profiles of half the people on the planet.
I’ve only had one incident that really, really gave me pause.
My friend and I were talking on the phone, and he mentioned he was going to move to a city closer to me. Awesome, was really happy for him. He didn’t know an address yet or anything like it, we just had that conversation about new job + new city.
I immediately got inundated with ads for real estate for sale in the city he said he was moving to (which is not where I live), particularly on Facebook from what I recall. As far as I can remember (which, granted, my memory is faulty) I had not looked up anything related to our conversation when the ads began appearing.
It definitely freaked me out a little bit.
That’s a fairly easy one to explain, I think. Through your phone call your friend was linked to you, digitally; they had been searching for real estate so some ad algorithm somewhere “thought” you might be interested in the same thing as the friend you just spoke to and pushed ads to you
It's exactly this and a reason I won't have Meta products and I say no to all the Google opt-ins.
The big thing for me was WhatsApp. Once Meta got their hands on I bailed for Signal. WhatsApp wants to know everything about your location. It feeds back the SIDs of WiFi, IP info, GPS, etc. If you go to the doctor's, they know and would feed you ads. But then they know everyone else who was there at the same time AND everyone who is in your household or associated with you. It's damn right scary. What's worse, they're all at it now.
Ahh I have to say I forgot about WhatsApp. I do use that, it’s basically ubiquitous in not-America. My impression was that it’s far less invasive than other meta products, and there not being topics to follow (pages, hashtags etc) mean they’re getting a lot less of a picture about us. I wouldn’t be surprised to find I’m wrong about that, though, it’s not something I’ve read into much.
I’m not very privacy conscious - I don’t use those social media apps for different reasons - and I sort of still don’t fully understand why I should care so much. On a broad level I understand that a private company knowing everything about you isn’t ideal, but I also don’t really understand the nefariousness of it - they can show us ads of things we might be interested in, which actually seems kind of a good thing (preferable to ads you aren’t interested in, anyway) but what else?
Im not sure if those are big and/or complicated questions so no pressure to reply, but if anyone wants to tackle it I’m open to learning exactly why I’m wrong and stupid for thinking this way!
I was on a WhatsApp call with a friend, talking about how another friend of mine had just lost a family member and needing to travel to attend the funeral.
After that call I started getting ads for funeral services and related businesses.
It was very obviously not the same type of ads I normally get, nor was it a topic I tend to talk about.
The connection was very clear.
I have actually started leaving my phone away from wherever I have face to face talks with friends. Particularly if it's a sensitive or private chat.
While I can handle seing ads for caskets and funeral themed flower arrangements without being emotionally hurt, the entire context was very uncomfortable.
WhatsApp is owned by Meta, correct? Sounds like something they could easily track and then target you with.
My creepy incident was recently. My partner and I were talking, in person, about how our old landline that we only use as an intercom (for the past few years) needs replacing. Less than an hour later, those were the Amazon ads.
However, I'm of the mind that it's a billion monkeys at a billion typewriters kind of thing, this is just the one that flew close enough for us to notice. I can see how it'd freak someone already suspecting spying, though.
I know it's probably likely that they're not BUT, I have had too many coincidental situations to cause me to question it. For example, one time my wife and I were in Home Depot looking at vinyl fencing and talking about potentially buying some. We hadn't searched for it before going to the store, we just randomly stopped to look at it in the store. Later on that same day vinyl fence ads started showing up in our Facebook feeds.
I think the main issue, that everyone is skirting around, is we can't trust anyone to tell us the truth here.
And, we can't trust them because every time we turn around we learn another way that the surveillance corporate-state is watching us and lying to us about it. They simply cannot be trusted.
So, for all the people saying it's not practical, there's not evidence, etc. They're correct. But, they're also missing the point -- we absolutely cannot trust anyone.
I don't buy into the "listening" bit but my wife was planning a trip last week and I started getting ads for the place she was looking. It was creepy but I sorta assume it's just an ip or chat based fuckup form some ad platform.