Hacker News Reader

Comment by lonelappde | Hacker News Reader

Amazon's code is not secret internally. If Alexa was targeting ads on your speech, someone from Amazon present or past would have confirmed it by now. Your conversation was probably influenced by the web ads campaign, or it was a coincidence, there are a lot of people seeing a lot of ads every 10 minutes.

View on HN

> Amazon's code is not secret internally.

That era is long gone. Internal-public is the default, but there are plenty of secret projects (private source/build artifacts)...

Amazon's code is not secret internally. If Alexa was targeting ads on your speech, someone from Amazon present or past would have confirmed it by now.

Or not? How many people have access to Alexa source code, let alone that part of Alexa source code, and are also willing to break their NDAs and get fired and/or sued?

Your conversation was probably influenced by the web ads campaign

I think this is likely correct, and a million times more sinister.

I like your point as I can relate to that opinion. However my outlook has changed in the last year or so.

There has been several occasions where my conversation somehow turned into relevant web ads. I don't think I'm making a mistake here.

A friend of mine told me of his good friend who left Amazon recently as his job was to listen to what users say to Alexa and ensure Alexa responded correctly. FYI that's both when users say Alexa first and also when they do not. More on that here [1].

I would love some more clarity into what's going on here.

[1] https://edition.cnn.com/2019/08/05/tech/alexa-amazon-human-v...

1) I would assume your entire social network is well documented. A lot of conversations include one person looking something up before or after the conversation. Most conversations don’t come up randomly. That can probably signal that you might be interested in such a topic too.

For 2) it’s worth noting that when you say you didn’t say Alexa, it means Alexa at least thought you did. It’s not trying to constantly record, but may still be a problem to you.

I've been keeping my eye out for good, solid proof that these assistants are doing ad-based targeting based on things being overheard, but I haven't found anything despite looking, and the arguments they aren't are reasonably compelling. (The people looking are quite motivated to find it. Big feather in their hat if they do and can even convincingly cast shade on Amazon, let alone prove it.)

My conclusion is that the targeted ads that people report are either A: coincidence (millions of users, only really tens of thousands of types of "things" to advertise with, sooner or later life coincidences would be expected to happen to large numbers of people) and/or priming (you notice the ad for camping gear since you were just talking about it, but don't notice the several hundred other impressions for camping gear that you just edited out entirely) or B: they didn't get it from your speech; you gave out other signals without realizing it, like a Google search that may not have been "wanna buy camping gear" but unknownst to you is known to be correlated to it, or as you say, your friend ran a search and now you're getting ads, or you're talking about camping gear in the first place because you are in a coffee shop near Camping World Outlet, and in general, the advertising machine knows that you're interested in camping gear right now because that's how well they have you under surveillance... they don't need your actual speech to know that's what you're talking about.

The first is mathematically inevitable, but poorly explains the cases where it recurs to one person. That seems to be the second. In a sort of way, I say that's even creepier than "they're listening to your literal speech and targeting you based on that". They've got you so pegged from everything else they're doing they don't even need that signal. That's creepy.

People have dozens of conversations every day that don’t match with the hundreds of ads they see every day. But then once or twice coincidentally a conversation preceded a relevant ad and the user jumps to the conclusion that their conversation was being listened to without noticing that 99.9% of the time the ads have nothing to do with any topic of conversation. Its occam’s razor or confirmation bias or some other phenomena HN likes to point out.

Alexa does not send data unless you say “Alexa”. This happens at a hardware level. This is confirmed by hardware schematics and by security researchers who have sifted through every piece of data that leaves the device.

What you are saying isn’t true. It cannot be.

How reliable is the keyword detection? Last week we had this story about Apple where Siri heard the word "Siri" out of random noise like creaking furniture.

My point exactly.

Just last night myself and my girlfriend were talking and Alexa decided to activate itself without us saying Alexa first.

Alexa rarely, but quite regularly activates from my TV in the same room.

the device can't magically detect the world Alexa - there is some lossy process of translating incoming noise into a vector that has some probability of being the hotword, and above this probability the device enables.

a more accurate thing to say would be:

Alexa does not send data unless, in the course of listening for the hotword Alexa, it interprets some noise or utterance to be close enough to Alexa to start sending data

Whether or not they are actually listening to you is a detail. The creepy thing is that they obviously know you as good as if they were listening. How that is achieved isn't really the point.

That is an odd reasoning.

If they know this based on data from Alexa, you and others can eliminate Alexa - or decide not to buy it in the first place.

Reply on news.ycombinator.com