Also, I suspect it's not so trivial to actually do this - there's a big difference between how people intentionally speak to a phone (e.g. making calls, FaceTime) and just ambient audio from a phone in a pocket or on a table.
>there's no way they'd be able to keep it secret, NDA or not, and there'd be a massive shitstorm about it.
If a developer goes public with that information, their career is pretty much over. They get sued for millions of dollars and their life is pretty much over.
The developers working on it could be led to believe they were working on something else completely. It only takes 1 person to pull the libraries together, and that would have to be someone you trusted.
Coca-Cola protects their formula by having several vats of stuff, so to make Coke, you mix 1 part vat A, 2 parts vat B, and 1 part vat C and you're done (just a hypothetical). Some people knew what was in vat A, but not vat B, etc. The same thing can be applied to programmers.
>there's a big difference between how people intentionally speak to a phone (e.g. making calls, FaceTime) and just ambient audio from a phone in a pocket or on a table.
Siri can do it easily. You have to be in close proximity for it to respond. Say "Hey Siri" from across the room and nothing happens. Just check volume levels and that's it. Anything quieter than a few decibels and you ignore.
Have you actually worked in software engineering? There's no way this approach would be practical, especially not in a public company, with as many platforms and sophisticated tech like FB. This is a ridiculous conspiracy fantasy.
Also, having worked in ad-tech, in order to sell this to clients you have to have product explain it to sales, you need marketing to create powerpoint presentations explaining where the data comes from, with stats showing why it's better, possibly your data science team needs to do bespoke reporting, etc... it would take at least dozens of people from multiple teams to implement something like this.
Yes, 20 years. When dealing with a large application, it's broken into smaller chunks and each chunk has a team. It's called "divide and conquer" and has been a strategy as long as I can remember. You have an AI team and a UI/UX team, a data persistence team, a voice to text team, security / encryption team, multiple teams dealing with various middle tiers / micro services; all providing libraries and services. You have teams that don't even work for you that provide FOSS code.
>This is a ridiculous conspiracy fantasy.
It is but it's fun.
>in order to sell this to clients
That's the point, you don't sell it to clients, at least not directly, you just add weight based on captured conversations to existing "weighted preference" of the client. Do you explicitly explain every technical detail, line by line, algorithm by algorithm of how you acquire advertising data to your clients today? Of course not, they don't care, they just want to sell their wares.
I mean Apple has secret teams by the truck load, and most things are leaked by third party manufacturers. I don't see how you can't imagine how feasible this is to do.
There's just no way you can sneak new functionality into an app without the engineering and product owners of the app knowing. You can't add new data feeds without the data teams knowing how much, how long, who can access etc... And someone has to make that data somehow useful to the actual advertisers, which requires that that they know what it is...
And yes, ad-tech clients do want to know exactly where the data comes from, whether it's reliable (somewhat and sometimes, in my experience), and whether they're allowed to use it (pharma).
Apple's stuff doesn't usually leak until it hits the manufacturers, which are third party. Apple itself is very good at keeping secrets within it's own company. One of the main reasons Snowden had access to that data is the NSA oursourced tens of thousands of tech people. If it's just one or two people, it's much, much less likely to be leaked.
>without the engineering and product owners of the app knowing.
You can sneak it into an app with just 1 engineer knowing. There are still 2-3 people who know the complete formula for Coke. You can get a SVP come directly to the developer and circumvent the product owners, or you could make a skunkworks team of just 1-2 people. Hell, FB manages to keep it's private keys private because only a select group of people can access it, let alone know where they are stored.
>ad-tech clients do want to know exactly where the data comes from
Funny, when I take out ads on FB, it doesn't tell me the algorithms used to determine who likes what, it just gives me the option on what demographics to focus on.
I would venture to say not only is it technologically feasible and organizationally feasible, but they could do it without getting caught unless people start noticing ads popping up for things they never search on, and only talk about person to person. I think that can even be hidden by adding weight only to other weighted preferences captured through other means.