Its similar to that Reply-All story where they looked into whether the claim that FB was listening in on people's discussion, and despite there being no physical way for it to happen, people still believed it.
Its similar to that Reply-All story where they looked into whether the claim that FB was listening in on people's discussion, and despite there being no physical way for it to happen, people still believed it.
I think what's much more likely is that machine learning algorithms are so good right now, they're able to determine what people are interesting in based on browsing habits and other metadata. Either that or they're using your location data and tracking what adverts people are physically seeing on signs/billboards.
So you may think your app is listening to you because you get an ad for something you talking about, but you don't realize you talked about it because your brain subconsciously registered a billboard over and over again.
Not that it's actually happening, but the best cell network speech codecs use ~6kbits/s, and it's probably possible to go lower with reduced speaker identifiability. You also only need to record when there's speech detected, and can delay and/or batch transmissions.
About ten years ago, I discovered that one of the major "rate you shopping experience"-type survey providers had a feature where shoppers would be tagged at the beginning of their session, and if they were randomly selected, everything they did on the website was recorded (including text entered in fields, etc.). When they checked out, they'd get a pop-up asking if they wanted to complete a survey. If they answered "no", supposedly that recording was deleted, but who knows? There was certainly no indication that answering "yes" would result in it being saved indefinitely for playback by the marketing department.
The only reason I found it was that I was curious about some binary data that would be uploaded every once in awhile. It wasn't hard to take apart (zlib, I think), but the standard tools didn't catch it at all because it was a custom thing developed by that vendor. I've still never seen anyone discuss it publicly, despite it having been very common on online stores for years.
It could also be that no one is doing marketing analytics based on voice recordings, of course. I just wouldn't take it for granted that someone would have found it and disclosed it.
I and a friend were discussing this before that podcast. He was of the opinion that FB was very likely listening because he had multiple occurrences where he'd visit family, and they'd talk about something, either some product that family was interested in or had just bought, and hours to a couple days later he'd see an advertisement for it.
My stance (which looks to be the same as yours) was that they don't even need to listen to you, and doing so would probably be really bad if it was found out. Instead they know who you're around, they what that they're interested in, and they make some educated guesses and tag you as interested in stuff they are also or that they just bought, and it's the hits that stick in our mind, not the random other stuff you were shown that you never talked about with others but they had searched for or bought.
Throw enough data at he problem, you'll get some hits. There's a lot of data out there right now.
Purina Dog Chow purchases "Qualified Leads" from marketers who have lists of people interested in Dog Food.
We have no idea where they get those lists. If you play Candy Crush, and that third party app has Microphone access it could be collecting the data, and selling it to marketing agencies, which then sell the qualified leads to Purina dog chow.
You are still being listened to and your information sold. It just happens on Facebook more than it being Facebook's fault directly.