I was sitting at a cafe with some friends, and we got onto the subject of weird anime we have watched. None of us are heavy anime consumers, so it's not really a topic that comes up often. I recalled a time I was at my brother's house a couple months ago and we got WAY too high and started digging through his Crunchyroll to find something dumb to watch, and found this movie from 2014 called "Satellite Girl and Milk Cow". So I described the premise, a guy who was turned into a literal bipedal cow by some wizard happens upon a sentient satellite that falls to earth and becomes a humanoid girl... I don't remember anything else I was REALLY high lol.
Over the next couple of minutes, I ended up saying "Sattelite Girl and Milk Cow" a couple more times. When eventually one of the friends doubted my recollection, they took out their phone to look it up. They started typing, and then stopped dead, stared at the screen for a few seconds and blurted out "What the fuck?"
They turned their phone around and they had typed "Satellite" into google and THE FIRTS RECOMMENDED RESULT was "Satellite Girl and Milk Cow". A goofy relatively obscure animated movie from 2014.
Ok, what? How? That's such a specific and obscure thing to pull as a first result. There are hundreds, if not THOUSANDS of other things that would be immediately more relevant to any random person's life than that. I don't have a smartphone so couldn't try on the spot, but when I got home, neither my laptop or desktop gave any similar behavior on multiple different search engines. It gave me a few seconds of dread that even though I have consciously made the choice to exclude myself from the constant data gathering of smartphones, other people's phones are still listening to me, which is something I hadn't really thought about before.
I'm sure plenty more people have stories like this, but the specificity and obscurity of this example is just so baffling to me, like there's NO way that it wasn't picked up as an audio cue.
I can almost feel the tinfoil hat beginning to grow out of my skull.
EDIT: Well, I definitely wasn't expecting this many replies lol. I guess the strategy to get more engagement is to just be wrong on the internet. (It looks like the search recommendation might have been an insane coincidence based on some unknown spike in search activity, per the google trends data)
Smartphones are not recording conversations.
Strike 1: Battery life would be enormously impacted. "Ok google" and such keywords run on specific low power hardware that THEN wake up the rest of the phone when triggered. General recording would need the full phone to always be running == very short battery life.
Strike 2: The whole combined cybersecurity field are constantly probing mobile phones (hardware and software) for security issues. If there was either code or hardware that was always listening you would have seen lots of headlines with actual proof.
The word you're looking for is "synchronicity".
I think a more likely explanation is that Google knows who your friends are, knows your friends watched this movie, and thus thinks it could be relevant to you.
Yeah, it's probably more of a correlation between all the data points at that specific time (everyone's phones in the same location and all the data points social media has on them), and the algorithm was able to piece together that information to form a guess on auto-complete. If we're really leaning hard into phones listening, then the most plausible explanation is someone in the cafe was actively using siri (or google's equivalent) while OP mentioned the anime title and it was picked up in the background.
Phones are designed to have massive talk times as is.
Most of that power is required for audio output.
But if you're just parsing for adwords? Capture and send snippets back to a server and process them there, attach them to the google/facebook ID.
Also, consider that cell phones used to have stand by charge times in the range of a week or two if left unused. Smartphones die if left unused for half a week.
I could've added that I spent 15 years working as a developer in the mobile phone industry, but in reality that shouldn't be needed. All that's necessary for you to independently verify what I wrote regarding how voice keyword triggers function and the difference in power draw between that and the full audio pathway for recording is available through the nearest search engine.
All you need to do is independently verify the claim that I have made. chefs kiss
Isn't that what people should be doing anyway? Am I missing something? If some rando tells me something without providing relevant data then (providing I care enough) I'd look it up. Otherwise, I just consider whatever they said to be untrue and move on.
But why couldn't they have secret "wake" words that don't actually wake, but instead have another secret functionality?
Like, you say "ok google" and it wakes, but you say "Toyota," and it checks a box on your secret profile they use that says "he said Toyota on Friday at 12:45pm" and then you and your friend and a couple other people at the starbucks you've never met get ads for Tacomas for a month due to all being in the same precise location for 47 min. Or you say "bomb" and it immediately puts you on a watchlist they share with the FBI?
I'm not saying it necessarily does, but for all the "it's impossible" people, why wouldn't that work? Why couldn't they hide that traffic with their normal data collection, or update and add new codewords? It at least theoretically seems possible. Sure maybe it isn't a 24/7 hot mic, but does it need to be?
Sure. And then all the developers who have made that functionality work need to be in on the conspiracy, and all the cybersec researchers need to be also (or incompetent). And the false positive rate would be sky high due to trying to match so many different words to the waveforms.
The phones don't listen to your conversations.
NDA.
It'd be impossible to hide the data transfers with the rest of it? That would be something but I'm still skeptical.
Fair, but also, so? Who cares? Let cocaine users get ads for coke, if that means coke drinkers get it too. Besides, as I'm told by the "there's definitely 100% no listening even for keywords" people, "they don't need to," by combining those keywords with the profile they already have, they could cut down on false positives.
Sure they may not "listen to your conversations" like a hot mic, and sure they may not currently have secret-functionality words, but why couldn't they? It still seems entirely possible.
I think you're mistaken. I'm not saying that smartphones are recording full conversations, but they are definitely capable of listening for keywords without their owner noticing. Auto Shazam (indefinite automatic identification of all music that plays around your phone all the time) has been a thing since 2013. How much simpler would it be to identify words being spoken around your phone and then send a short text hash of that word to a server? Also, see the post from pelespirit on this thread with information about actual, specific ad-tech for this.
I don't think it's either. I think it's reasonable to assume google knows that
From there it can just infer a bunch of stuff (products) they might talk about.
Those wake words are indeed running on very low power consumption electronics. And it's not like they're made to run exclusively in the USA and listen only for the american english version of "OK Google". No, that ASIC is made to listen to a select set of phonemes from the most common languages. It then feeds a tokenized representation of what was spoken to whatever needs it downstream. The wake word is programmable - a series of phoneme tokens. But what about the rest? Cache them somewhere and next time the phone is woken up transmit them to google for ad processing.
"Strike 2" applies even in that scenario.
Where are these mythical security engineers that volunteer their time and expertise to delid ASICs and reverse engineer their functionality?
waving
Not sure what you mean with "mythical" though - whenever we're allowed to talk publicly about something we happily do.
(My own speciality is in IoT devices, others focus on mobiles)
Two decades ago I had a candybar dumbphone - less than two decades later I bought this tiny device (which I got when it came out two years ago) that was significantly more powerful than my full-blown computer back then, along with features nobody would've thought could be packed into something so small.
There's always newer, better, more efficient tech being developed. What were once known limitations are regularly being overcome. The advances in tech just seem to keep accelerating well past "ludicrous speed," and so to assume that such roadblocks as you describe are hurdles that will keep those determined to find a way past them at bay for a long time to come is a fool's folly, IMHO.