My own creepy Android anecdote: A few months ago, I had a stiff neck. I asked my wife if we had any of that Biofreeze left. She said yes. Within 30 minutes, for the first time ever, an ad for Biofreeze came up on my Facebook feed. I have never searched for anything related to Biofreeze on my phone or laptop, and I asked my wife if she had (she hadn’t.) No ads for Biofreeze have shown up since that one time.
Dictation isn’t doing the kind of language processing smart assistants do, though.
I am in complete agreement with enalzi. Claims of always-on audio monitoring are universally based on someone seeing an ad, without any consideration of confirmation bias or coincidence. Meanwhile, the lack of actual evidence like constant voice-bandwidth data streams from devices, debug information on the device showing a process continually reading from the microphone, the high battery drain that would be associated with those activities, or the fact it would be massively illegal to conduct such audio monitoring are not considered or disregarded. As a factual matter, I am quite comfortable stating Google, Facebook, et al. do not perform always-on audio recording.
The way he phrased it made me think that wasn’t what he was talking about but I guess it could be.
With the proper software, a full time stream of audio going to google/amazon/FB would be as obvious to skilled networking types as firing a flare gun in a darkened theatre.
There are people/companies who actually do this kind of network analysis for a living, and have looked for it. Its not there.
That said, as network bandwidth climbs and mobile device processing power keeps climbing, such things will become more practical, but its not going to change the legality, or the visibility to tech types who know what they are looking for.
I think it might be worth noting that there were three pretty-obvious autocorrect goofs here; should have been, I presume:
I’m betting you received those suggestions due to having your location services on.
i wonder about this topic from time to time. i don’t think amazon has a room full of humans listening to us go about our days, but could they be transcribing conversations to text for later use?
first of all how does alexa know when you say “alexa…”? is she always listening and if so where do all the phrases that do not start with “alexa” go? instant delete? or is there a software cache that keeps it for a few minutes locally?
tin foil hat on could it be saved to an encrypted text file that looks like something else that’s useful to the device?
how about this. i don’t know much about how this works but could it go something like this?
speech to text dictation is getting pretty good, so it would be easy for alexa to save everything you say to a text file with a decent degree of accuracy. at least enough accuracy to parse what you’re talking about if key words are flagged for review. in the code there’s an if/then… if device hears and writes “alexa” to the text file, then the device interacts with humans. otherwise the device just sits there mute.
would it be possible for alexa to save everything said in the room to a text file and have the text file occasionally upload as part of the usual traffic? this way the eavesdropper is not sending massive amounts of audio data, just some text/code that won’t raise any eyebrows.
back on the eavesdropper’s server, the text files are scanned for keywords and nefarious actions are taken by ‘them’.
I’ve had odd experiences like this. One anecdotal example:
One of my coworkers was streaming music from his phone through the Bluetooth speaker we have. He was streaming 1980s pop music that day, and the song, “Video Killed the Radio Star” came on, and I mentioned that I hadn’t heard that song in years.
After work, at home, I visited YouTube, and there was the video for that song, front and center in my recommendations. It was particularly strange because I almost exclusively watch videos from modern metal bands, so “Video Killed the Radio Star” couldn’t have been even closely considered a “related” video.
This has since happened with other songs, from other artists I would not normally seek out on YouTube, but happened to be streamed by one of my coworkers.
For everyday purposes, none of these companies have humans listening. This is what AI is for. However, in order to make the programs better, a tiny fraction of interactions are checked by humans to see how well the software is doing to make adjustments. Some providers allow you to opt out of that.
One reason there are trigger words/phrases to initiate an action is that the devices can be set up to locally recognize the trigger. Note that generally one cannot set any phrase you like since this requires processing at the cloud end to set up and the device might not work well with the new data from that.
So an Amazon has only 4 options: Alexa, Computer, Echo and Amazon. Here’s an article about some of the issues involved in going to a cloud-based wake system.
Keep in mind that a lot of these devices are incredibly simple. There is nowhere near the processing capability to do dictation. You need a decent phone or computer for that. My Amazon Wand is so simple it can’t even handle a wake word. You have to push a button and then start talking. Ditto Alexa enabled Fire TV remotes.
For always on devices, yes, the device is listening. No, it is not recording. As the bits come in from the microphone, it does a match for the wake word. Anything not matching goes into the bit bucket about as fast as it comes in. (It’s not tecnhically “deleted”. It gets overwritten by whatever comes in next.) Only when there is a match does it send the audio stream to a cloud server which does the real work. Again, no reason to save anything locally.
My phone and computer ads predict with stunning accuracy things that I have just bought. They offer them again and again.
But, about a year ago I was suffering from weepy eyes and sniffles, while browsing. And the next day I started getting dropdown ads for allergy relief and eye drops.
Yes, it was allergy season (for me, actually, though, this can be any season). But I put a piece of tape over my laptop camera and it’s been there ever since. Because that was just freaky.
For the people saying the bandwidth would be a problem, the full audio for your conversation is already passing through a server somewhere. What if it’s just being stored on that server, and Google is paying to data-mine it?
For the past two days I’ve been trying to get my Android phone to listen to me talking, and then manifest some kind of ad to show that it’s been listening. Every hour or so I say, “I really could use some new shoes. Where should I look for shoes? Where could I find some good walking shoes?” And: "Man, I’d could go for some ice cream–like Ben and Jerry’s. Where can I get some Ben and Jerry’s ice cream?
So far, nothing has come up. Nothing about shoes, nothing about ice cream. If Google had a blanket practice of listening to EVERY conversation, regardless of the person, regardless of browsing history, then they’re not getting anything out of it with me.
Good way to tell if you’re haunted, too.
My husband and I recently tried something similar. We decided to loudly discuss frozen yogurt, pilates, and I think one other topic we definitely never searched for, with our Android phones out. We made sure to use those key words repeatedly, and as many related words as we could think of. We waited months, but the froyo/pilates/whatever ads never materialized.
We’re not talking about telephone calls; we’re talking about cell phones eavesdropping on IRL conversations.
Now, the telephone calls … well, that’s a whole 'nother ball of Snowden/NSA/PRISM/PanOpticon paranoia, so let’s leave that be for now. But it’s always been prudent not to put any information into a phone call/telegram/email/text that you wouldn’t want to see on the front page.
ETA: “The front page” refers to an archaic information-delivery system known as a “newspaper,” which apparently involved a great number of dead trees somehow.
It wouldn’t have to upload continuous streams, just sporadically save words it overhears and upload them in one brief burst every once in a while. That would also be consistent with the “sometimes it seems to know, sometimes it doesn’t” stories we’re getting.
This is possible, but it’s really unlikely for a few reasons.
For one, the number of people required to build and run such a system would be pretty large. Conspiracies happen, but the successful ones are often pretty short-lived and involve just a handful of people. I’m going to say that at least one of the dozens or hundreds of engineers required to build a system that did this and funnel all the voice data around would have some misgivings about such a thing and go public. Or one of the thousands of security researchers that might stumble upon such evidence.
The PR hit Google would take would be disastrous. I’m not saying that companies always act in their long-term self-interest, but this would be such a horrendous misstep that it kind of strains credibility that they would do this.
And, as pointed out here: they don’t need to do it. You actually can’t learn anything that interesting about people by recording their phone conversations that you can’t infer from the piles and piles of data that we provide in other contexts.
Again: there are really easy ways to test this theory. Talk about something around your phone but don’t search for it (and make sure your friends don’t search for it) and see if ads show up. They won’t.
Google knows where you are all the time. Even if you turn off location services, it still knows where you are based on the wifi networks that are available to your phone, even if you don’t connect to them. It knows who your friends are because you email them or have them in your phone contact lists. Google thus knows when you are meeting with your friends in real life. If after the meeting, Google sees that someone Googles or emails about some destination, it rightly or wrongly assumes you may have been talking about that destination and starts serving you ads for that destination. Sometimes, it will guess wrong. I might have dinner with friends and then start researching a trip to visit my in-laws without ever mentioning the trip to my friends. My friends might start seeing ads for sunny California but they won;t have any clue that they are getting the ads because of my search history, so it won’t stand out as confirmation that “Google is recording our conversation all the time!”
The critical thing to realize is that it’s only marginally less creepy that Google knows the content of our conversation without needing to listen to us.
Any chance that she searched for GoreTex online? The most obvious route to me is that she searched Amazon for Gore Tex, whether at work or at home. The computer or phone that she used has also connected at one time or another connected to her Amazon account, so it knows who she is. Some other device that she uses has also connected to your home network. I could anthropomorphize what Amazon “thinks” here, by saying things like “Amazon knows your wife is searching for Gore Tex and it thinks you might want to buy her a Gore Tex present.” Of course, computers don’t really think like that. What Amazon really knows is that when someone searches for Gore Tex, Amazon makes more money when it starts serving up ads for Gore Tex to everyone with a close tie to that person. A person who lives with her is about as close as they come.
Saintly Loser, you probably have your friends’ contact info in your phone, thus stored in Google’s servers. Or, even if you don’t, they probably have your contact information stored in their phones, and thus saved in Google’s servers. Google knows who your friends are even if you don’t touch social media. And any of your friends searching for, emailing, watching videos, etc. about Joshua Tree after the conversation will trigger the ads for you.
Exactly. Do you both have the Facebook app on your phone? Facebook knows where you are and when you are together. A meeting followed by your colleague’s search for these old acquaintances was enough for Facebook to think you might be interested in them too.
This could be a coincidence. Or it could be a result of another search you did for something like “sore neck treatment.” Or it could be because you looked for Osteopaths online.
Google knows that you have allergies based on your searches and data history from years past. It didn’t forget just because you did. It knows from the weather reports when allergies will be bad. It can put two and two together to know that people with allergies might be interested in antihistamine ads when allergy season comes back.
I have suspected for a while that I am getting targeted ads purely from things I have mentioned verbally near my phone, without actually searching or otherwise interacting. I really need to do some proper controlled experiments, by choosing a subject completely at random (maybe by opening a hard copy encyclopedia, or throwing a dart at a world map?) and then having conversations about that thing.
What if Google has a level of advertising that takes an advertiser’s keywords and applies them to the “wake word” algorithm? If they get said then the targeted ads are injected to the speakers’ Google ad rotation.
This eliminates the need for constant saving and parsing of all words spoken (it works exactly the same as the wake word), advertisers won’t spill the beans on it because they want it to continue, and it explains why talking about BIOFREEZE gets you an ad (biofreeze had paid for the ad tier) and talking about FROYO doesn’t (nobody had paid for the frozen yogurt keyword).
Eh? 
FWIW my friend had an experience just like the OP’s last week, except she has an iPhone. She was at the physical therapist talking about bras that would fit a specific need for her. She hadn’t searched on these types of bras, then suddenly that night she got ads for them. I’d chalk it up to being at the PT’s office and Apple using Location Services to send her sports bra ads, but I saw the ad - they weren’t for sports bras, they were different types of bras.
Yes, that makes sense. She is emailing with people at Gore Tex from work and home. We share a kindle library and she uses my Amazon Prime account with a credit card in her name.
It was a shock, though, seeing an advert for a Gore Tex book! I assumed I’d been set up by my gf as a practical joke.