I think you partly may just be biased by happening to notice ads more if they fit the topic you sent to your wife or being lenient in deciding if ads meet the categorization. And if you dwell on an ad because it seems to match, you may get more similar ads.
Here is how I think you could design a more robust (but less fun) experiment:
- Come up with a bunch of topics, write them down on slips of paper, put the paper into a hat
- Each Monday, draw three topics from the hat, send some WhatsApp messages about the first, Messenger messages about the second, and don’t discuss the third. Don’t put the topics back in the hat.
- If you see any ads relating to one of the topics, screenshot them and save screenshots to eg your computer with a bit of the topic
- Separately, record which topic went to which platform
- After doing this for a while, go through the screenshots and (each of you and your wife or ideally other people) give a rating for how well the ad matches the topic. To avoid bias, you shouldn’t know which app saw the topic.
- Now work out average ratings / the distribution across the three products (WhatsApp vs Messenger vs none) and compare
A simpler protocol to realize that the Baader-Meinhof phenomenon is probably what's happening:
- pick said topic, something you never cared about before, talk about it but don't write any messages containing it;
- for 1 month record every ad you see about it;
- send a message about the topic;
- for another month, record every ad you see about it
Comparing the number of occurrences will tell you what is happening.
> pick said topic, something you never cared about before, talk about it but don't write any messages containing it;
This does not work. How did you come about the topic? Answer: it was in your brain, because advertising, trends among your peers and social connections, online trends real or astroturfed, etc.
That's why you end up with people thinking their phone is "listening" to them.
It also has to be a topic that advertisers would pay to target you with. You can't talk about something super obscure that advertisers don't care about - like steam engines.
Thanks for this!! Now my email is full of offers to buy historical steam engines, steam engine parts, and engineer hats. Amazon is even advertising a subscription for coal deliveries!
The problem is, your smart TV could be spying you too, if it's capable of voice commands or videoconference. If you discuss sex toys near it, at least some related keywords could make their way to targeted advertising.
My wife and I routinely use ad blockers, private browser windows, browser profiles, and try to use as little ad-supported products as possible. This doesn't stop targeted advertising, I guess because most devices we use connect through the same IP. A couple of days after she starts looking up a city we want to travel to, I'll start receiving ads from airline companies or travel agencies, and even tours/cruises to said city/region. Fighting tracking and spyware is nearly a lost cause unless you become a digital Amish.
Smart TVs in general use IP address to try target devices across households, which is against the privacy policies of a lot of ad tech providers because IPs are not redactable/resettable by an end-user.
The best way for small ad tech providers to compete with "big tech" has been to cross lines that the bigger companies won't cross, this is an example for why there are a lot of profitable ad tech companies in the connected TV / video ads space.
Even if you use a VPN, the TV itself likely has a unique ID for ads, so someone just needs to see one request with both the true IP and the unique device ID and then remember that for the future. It's all very shady. TVs are very far behind the level of user control that phones and browsers provide because there's less scrutiny and its more fragmented across manufacturers (all of which want to get in on ad tech).
You can usually find some opt-out of the identifiers if you dig deep enough into the menus, because multiple laws and regulations require them.
The first thing I did when helping my parents set up their new LG OLED TV's at xmas was to disable all the ads and tracking. It's exhausting how much pressure they put on you to opt-in, and how many layers there are, constantly implying the TV will be nonfunctional without it.
But sure enough, it works just fine with no ads, no "free tv channels", and no voice functionality.
Have you ever checked back to see if updates had re-enabled some of those or introduced new ones? You'd hope they'd let you know if they started getting ads all the time, but the tracking stuff is much less obvious.
LG will send out updates that require you to accept new license agreements when you turn the tv on next. It’s very obvious about what they are tracking but very obnoxious in pointing it out. The parents that OP refers to probably just clicks accept all and moves on.
We have an LG tv and one of my family members hit accept all after an update and now my remote listens to us. To fix this properly I would need to factory reset which loses all of our streaming settings. I actually don’t because I have a separate ISP only for our TVs so there’s a bit of separation between our streaming use and phone activity
If anyone wants to try this, a friend sent me one link to a device called Levo which does “herb oil infusion” aka it lets you make weed brownies easily. I clicked the one link my friend sent and now I get ads for Levo constantly in my YouTube adroll. Though I should say this is obviously on Google’s ad network specifically and I have no idea if this applies to other networks.
No I realize the differences. I am just saying that if someone wants to see if their encrypted app is resulting in ad serves, they could try discussing this product in encrypted chat only, using the methodology described in the comment I was replying to above.
These kind of stories are always fun to analyze using the Socratic method.
-How did you learn about the product?
-Have you ever searched for it?
-Did a friend of yours tell you about it? Do you think they searched for it?
-Are a lot of ads for it playing on TV channels you like? Could instagram know you like those TV channels?
-Is it something your neighbors got? Do you think there has been a spike in shipments of this product to your neighbors?
Eventually people start to “get” that scanning the text of messages is way more helpful for humans than it is for computers. They’ve got other data they can use.
I also have a theory that sometimes when people say "we were talking about <product> and I never even typed it into my phone or anything, and suddenly I started seeing ads for it the next day!" that the person in the story may not have looked up <product>, but someone else in the conversation might have Googled it or browsed an Amazon listing or something and they have some kind of connection in their ad profiles whether it be that they know these 2 people interact a lot, they're in the same geolocation, same wifi network/IP address, etc.
I'm just not convinced of the always on microphones in phones listening for and processing every single thing considering how much battery drain that would cause, whether the processing is done on device or they're sending all that data to a server to be processed.
> I'm just not convinced of the always on microphones in phones listening for and processing every single thing considering how much battery drain that would cause,
"Roughly twice per second, a Roku TV captures video “snapshots” in 4K resolution. These snapshots are scanned through a database of content and ads, which allows the exposure to be matched to what is airing. For example, if a streamer is watching an NFL football game and sees an ad for a hard seltzer, Roku’s ACR will know that the ad has appeared on the TV being watched at that time. In this way, the content on screen is automatically recognized, as the technology’s name indicates. The data then is paired with user profile data to link the account watching with the content they’re watching."
None of the people I know who use those devices knew that was happening, but the info was out there at least. When so many people are watching everything you see and do and say who can ever know what every company is doing or what the source of any one ad is?
> "Roughly twice per second, a Roku TV captures video “snapshots” in 4K resolution. These snapshots are scanned through a database of content and ads, which allows the exposure to be matched to what is airing.
There were users under the impression that Roku was unaware of the content it was displaying? Like 4K snapshot or not, if I know a user is watching an NFL stream, I know that ad played.
> There were users under the impression that Roku was unaware of the content it was displaying?
Sure, they expect Roku would know if they launched Disney+ or Netflix, but not that they would knew exactly what movie you were watching or what specific scenes you viewed and for how long. Same with personal videos cast to your screen via roku. It's pretty reasonable they'd know you were streaming content from your other devices, or which apps you were using, but less reasonable that they'd be watching over your shoulder taking notes.
I don't think it can be observed because it's likely a bug. It freaks users out. People uninstall the app and start threads like this.
It's sort of like getting mugged once and then setting up a camera in a bunch of alleys to prove that muggers exist. You can even set up a camera of yourself running into dark alleys every night, but the odds of reproducing a mugging is still extremely low.
There's a certain kind of precision that convinces me it's real though. Precision is common. I look at a book on Amazon, and a FB ad for that book appears.
But I get rejected for a loan via WhatsApp and then used car ads appear for that model of car that I applied a loan for? That's a bit on the nose.
> But I get rejected for a loan via WhatsApp and then used car ads appear for that model of car that I applied a loan for? That's a bit on the nose.
Aside from random coincidence, I could see this happening if you provided your personal information (especially email) for the loan application. It could have been shared to multiple underlying lenders alongside a data vendor who ultimately provided interest targeting (which can include car models) to an ad network.
Getting an ad for that specific model could also have been due to other online activity, such as checking the KBB.
> Getting an ad for that specific model could also have been due to other online activity, such as checking the KBB
I suspect op may have researched the car model and got retargeted: some ad networks keep track of specific products you've shown interest in (not generic interest-areas like Google ads) and track you via a cookie. You may be visiting a completely different site that uses the same network, and get ads on the exact product you've spent minutes reading about.
This is facebook. They've been caught recording people and selling that for advertising, they deny it because technically your audio is transcripted not recorded and they can send only some keywords back so whole conversations aren't sent back to them.
Plenty of research and news stories about this if you care to search. The speculative part of my comment is about the transcription which I'm speculating because of their fervent denials despite evidence which technically their wording in their denial statements is correct.
If I had to guess, your whatsapp messages are e2e secured but keywords are sent to facebook when they match some condition. So if you message "happy birthday" to someone, they won't see that but the fact that the keyword "birthday" was found even if the word isn't included is sent to fb. That way they can say they're not snooping your messages.
We did an experiment. We talked about how hard it is to find highlighter yellow nail polish. Nobody in the house is a purchaser of nail polish nor did we do any searches for highlighter or yellow or nail or polish. A day or two later my wife got an Instagram ad for highlighter yellow nail polish. It could have been a coincidence or maybe they were listening.
Or maybe some combination of things we did previously led naturally to thinking about that yellow nail polish. I'm thinking about something like the trick where you ask somebody a bunch of addition problems that have 14 as the answer (what's 10+4? 2+12? 3+9? etc...) then ask them to name a vegetable and they will almost always say carrot.
"I remembered the time I was in my fraternity house at MIT when the idea came into my head completely out of the blue that my grandmother was dead. Right after that, there was a telephone call, just like that. It was for Pete Bernays--my grandmother wasn't dead. So I remembered that, just in case somebody told me a story that ended the other way." -- Richard Feynman, "Surely You're Joking, Mr Feynman"
Someone did this trick with me in the '80s, but the numbers were to sum to 13. I still said "carrot", however. Wish I would have thought to use a different number than 13 when I tried it out on others.
It's an association thing. I thought carrot too. Maybe it's because of the tedious nature of processing. Maybe it's because the orange bar at the top of the screen makes me think carrot.
This is a message board about tech... comments aren't welcome anymore - we need evidence to participate?
I think it's interesting when a bunch of people chime in and say "Hey, yeah, I had some crazy thing happen to me, I'm in tech and understand how this stuff works, and there's a very small to zero chance this happened through some other parallel construction by the tech company, they just straight up listened to my conversation and showed me an ad".
This is what kicks off a handful of you to go packet sniffing and write up a blog post looking for this behavior. So yes, evidence is welcome but it doesn't seem like we are quite there yet.
In general I agree, but I think when you are being explicitly asked for a "source" in response to an allegation that it is settled that FB has been "caught recording people," I would prefer to not have anecdotes in reply.
I mean… this is a conversation, not some sort of formal debate? Someone is telling you "hey, this happened to me," and your response isn't "have you considered this other explanation?" but rather "I won't discuss this further unless you do a bunch of research and present the results to me."
I'm happy to continue discuss it (not sure where you are getting the idea that I'm not from), but I think it is also fair to point out when someone asks for a source to a claim that something has been proven/caught and instead the replies are a bunch of personal stories where people think something is happening.
To me, that is indicative that, contra the original claim, no such thing has ever been proven.
It's not verboten. But, candidly, it is kind of rude. What's the difference between someone at, say, the EFF "proving" something happened by running an experiment and writing about it publicly, and someone on Hacker News doing the same?
I disagree that it is rude to point out something is an anecdote.
The proof has to do with the technical details, not the authority figure posting it. If someone from the EFF wrote a blog post with the same content as these HN posters, I would be similarly dismissive of this as "proof."
They're saying "facebook has been caught multiple times doing this", which is not a personal anecdote, but an assertion that proof exists and is available.
> Edit: Holy fuck there are (paid?) Facebook shills all over this like flies on shit.
From the HN guidelines: [0]
> Please don't post insinuations about astroturfing, shilling, bots, brigading, foreign agents and the like. It degrades discussion and is usually mistaken. If you're worried about abuse, email hn@ycombinator.com and we'll look at the data.
Don’t iPhone have an indicator when the mic is recording? Also, this feels like it would be insanely easy to test by capturing the payloads sent to FB; you could even use something like Charles Proxy to do it.
FB having access to microphone makes sense for plenty of other completely innocent reasons (for example, if you can record a video from inside the app).
If this was actually true, I can’t help but feel that someone would have proven it technically by now instead of relying on these types of self experiment and anecdotes, especially given how commonly this is touted.
> 3) If a background app starts then stops recording audio while the screen is off, would you have an indicator that it recorded audio?
Yes. iOS displays an indicator if an app has recently used the mic.
> Note: Whenever an app uses the camera (including when the camera and microphone are used together), a green indicator appears. An orange indicator appears at the top of the screen whenever an app uses the microphone without the camera. Also, a message appears at the top of Control Center to inform you when an app has recently used either.
The phone was sitting between two people having a conversation, one of them "swiped it open" meaning it was off to begin with, then was immediately displayed an ad for that conversation, and upon hearing this the tech-savvy person in the house understood what happened, confirmed it with the mic access to facebook in the settings, and then disabled the behavior.
Considering the original claim was "zero indication that the mic is hot" and now it's "zero indication that the mic is hot if the screen is off", I'd say that the goal post has moved considerably.
But if you want to know if Facebook is listening to you through the iPhone microphone, you should probably look at the screen for the indicator. iOS apps can't start recording on their own in the background, there's no API for that. If they are listening to you, they'd have to start the audio session in the foreground, which would allow you to see the indicator.
I wrote the original "Wife swipes open the phone" comment, so that's the context you seem to be missing. Sure you can see a little dot on your phone when YOU run some experiment today and look for it, but was that indicator available in the exact situation where the targeted ad was displayed? No.
Also, this incident happened in the past and we know there have been dramatic API changes on both Apple and Facebook products. The limits of the API today don't reflect the capabilities that were available to developers in the past. I doubt Facebook is hacking the App Store process to use hidden APIs. It was probably just available in the past and my wife granted the facebook app complete access to the mic, so they took what they wanted.
I'd make sure to disable that permission today too, just in case.
One last thing is I just opened my iPhone again and hit record. I honestly didn't see the tiny orange pixel at the top of my phone until you pointed it out. I was basically looking for the green video indicator light to show. So I'm technically wrong about NO indication, you're welcome.
Android has a notification now when the mic is recording and has had the ability to deny microphone and lots of other access for a long time now. Thankfully it sounds like iOS is catching up
And don’t forget the battery. A mic recording 24/7 would drain the battery much faster and would not go unnoticed unless specialized hardware is used like the one for “hey siri” and “ok google”.
Try any voice recording app for a few hours, now use the facebook app for the same number of hours. The impact on battery life of a mic actively recording alone is very noticeable, so noticeable that your phone has a special chip just to recognize patterns similar to “hey siri”.
It would not need to record high quality audio and could maybe even take advantage of that same chip? Just thinking out loud here - smaller, crappier audio would also be easier to send back unnoticed (or instead of even recording audio it could be transcribing on the fly to a text file using something super basic and easy with low accuracy)
> This way too I can troll people on the internet when they suspect this is happening and I can say "bUt ThE bAtTeRy LiFe!" to defend Meta: my corporate overlord business daddy.
Please, stop with the sarcasm.
Okey, let’s say they manage to record us without a huge impact on our battery life. Now, how do you send these recordings or even the extracted keywords from a popular app, a client installed on devices controlled by the users and susceptible to reverse engineering and network traffic analysis without anyone noticing it?
This isn’t evidence. Even if Facebook was not listening to your conversations, there would be some rate in which you would just randomly be served an ad related to a topic you were discussing. There needs to be evidence that it is happening at a rate too high to be attributable to chance.
Sounds like a good way to engineer it... anything to improve the bottom line even if insanely-targeted ads only trickle out to users. How about limiting who sees this feature to also limit the risk of being detected? Maybe just do it once a year to everyone, or never to specific "tech-savvy" users that they have completely profiled.
So if were walking past a playing blaring the piña colada song, I’d see ads for alcohol and umbrellas? If coworkers around me are talking about activities I’m not interested in, I’d see ads for those?
They have far better information that shows I’m not interested in alcohol or extreme sports. Audio in the background is so low-signal that it isn’t worth showing ads based on it.
Even just transcribing speech something accurately is not something that was possible until the last couple of years. Yet this conspiracy theory has been around for a decade or more.
Yeah, it's probably not a coincidence that your wife is talking about X and is recognised by Facebook to be in a group of people that are interested in X.
Would they argue that the message goes first into a neural network that outputs potential product labels based on the message and that it all happens client-side? That's the only way I see it possible for them not to violate the E2EE.
An important thing you’re missing is the control. You should record every ad.
You need to know if you got 3 topics of ads every day and 1/3 of them are related to that secret topic, OR if you get 300 topics every day and 1/300 are related to that secret topic. If it’s the former, it’s suspicious, if it’s the latter, it’s way less suspicious.
The control is the topic you pick that you don’t discuss on WhatsApp or messenger. The idea is random differences between topics should average out over many trials.
I still think it’s important to consider the volume of topics that show up that aren’t being explicitly looked for.
I’ve gotten Instagram ads for ketamine and I absolutely am not discussing or searching for it. I probably wouldn’t even notice a random topic if it’s not so absurd. I’m sure there’s tons of topics I don’t even realize I see.
The reason for the control I suggested is to try to counteract the bias people have towards noticing things they recently thought about. I think the question of what adverts people are shown in general is interesting but quite separate.
You can’t come up with the topics yourself either, because the topics you will think up are different based on your demographic / type of person you are, and ad networks basically try to guess that.
You can if you first come up with a list of topics, and then once you have that list, randomly assign each of those topics to one of the three categories.
There will be some bias in what they choose to screenshot right? Meaning, the unrelated topic might show up in the feed but they don’t screenshot it because it doesn’t fit the narrative?
Also, what we’re interested is if the text changed what was shown. If I saw ads for X last week but didn’t notice them, then spoke to a friend and noticed them and took a screenshot, it would appear to confirm the theory. Even though I was always seeing ads for X.
Ultimately, I don’t think people who are convinced of this theory will change their minds so it’s a moot point.
Yeah, that’s the biggest flaw in the experiment I proposed, I think. This is the reason I try to have the hopefully independent grading of ad-topic-relevancy blinded to which system the topic was communicated over. It may be that one sees many vaguely related ads for the WhatsApp topic due to some selection bias but a similar number of actually related ads.
I call this the gaslighting explanation: “no, it definitely wasn’t the messaging product owned by an advertising behemoth. You must have searched for it somewhere else.” Obviously the OP remembers where they’ve seen the product. If they has seen the product elsewhere, they wouldn’t have started this thread!
Here is how I think you could design a more robust (but less fun) experiment:
- Come up with a bunch of topics, write them down on slips of paper, put the paper into a hat
- Each Monday, draw three topics from the hat, send some WhatsApp messages about the first, Messenger messages about the second, and don’t discuss the third. Don’t put the topics back in the hat.
- If you see any ads relating to one of the topics, screenshot them and save screenshots to eg your computer with a bit of the topic
- Separately, record which topic went to which platform
- After doing this for a while, go through the screenshots and (each of you and your wife or ideally other people) give a rating for how well the ad matches the topic. To avoid bias, you shouldn’t know which app saw the topic.
- Now work out average ratings / the distribution across the three products (WhatsApp vs Messenger vs none) and compare