I use Flipgrid to consolidate inbound tech and economics news; along with a few podcasts and my weekly Economist that represents the bulk of my news media intake. This time of year it’s a particular minefield, of course, with politics. But for the most part it’s my regular vegetables of tech and economics that get me what I want to know.
I was reading an article about how Amazon is launching an Alexa service for property management — e.g., the property manager pays for/owns the Alexa that lives in the residence with the renters, using it as a de-facto localized presence to control smart home things and, essentially, as an “added service/feature” of renting the place. (So much as you’d look to see if there was that extra half-bathroom or if there was a walk-in closet, you’d see if they included Alexa, too).
For the record, I read articles, because a pet peeve is when you get the poster who forwards an article that they clearly haven’t read (e.g., using the article to make a point that the article actually counterpoints). This is a case of me reading two separate articles, coming to a conclusion, and that conclusion was wrong. It’s a better case of a colleague gently educating me.
Firstly, to the other article. Granted, this NYT article is about a year old but we all remember the news that made the rounds about how Alexa is always listening. It’s true, she is: she *has* to. Obviously she can’t start your timer or add your biodegradable pet waste bags to your Amazon cart if she can’t hear you. In the NYT article, it’s about what she has done, and where that data goes, once she hears you. There is a sentence from that article, however, that did not stick in my brain from last year, so when I read the TechCrunch article, I made a comment on Twitter/Linked In.
My comment, quoted, is here:
“Two things: 1. interesting way to make IoT accessible to a broader base and 2. I would not at all be reassured the data is truly deleted (and isn’t, say, shipped off in snippets for “logs”/“troubleshooting”, for example). Also, the hand waving over who’s data it is needs to stop. Alexa has to listen to everything in the first place to trigger on her name.”
For the record, I still think #1 is true, and most of #2 is still an open question for me. I’m not at all clear on what happens to the data (yes, deleted at the end of the day, but… is it? What part of it is deleted? Is it every command, every call; or for example is there a record still in the smart thermostat (or a downstream reporting service) of all the changes I made, for example? And so forth.) Or who owns it (e.g., if something happens in the home, and the home belongs to the property manager, and the Alexa belongs to the property manager, but I’m the one renting the home, is that day’s data mine or the property managers?) However, this post is to talk about someone who reached out to address the last point: “Alexa has to listen to everything in the first place to trigger on her name.”
Now, it’s true that she does have to listen. However, a generous colleague reached out — privately, via LinkedIn messenger — to reassure me that Alexa does listen in for her name, but that listening happens only on the device… she doesn’t “trigger” until she hears her name, so no data leaves her until she does. Or put the way they put it (bold is mine):
“Wake word detection is done on device in a closed loop, that is no audio sent to Alexa (aka. the cloud). Only when the on-device model detects the wake word with a high confidence, the audio of the wake-word it sent to the cloud for additional verification (besides false-positives this handles for example “Alexa” being said in ads). No audio is ever sent to Alexa without a visual cue (the blue light).”
(Incidentally, the NYT article has this in a sentence that didn’t stick in my brain at all (bold is mine):
“…it’s true that the device can hear everything you say within range of its far-field microphones, it is listening for its wake word before it actually starts recording anything (“Alexa” is the default, but you can change it to “Echo,” “Amazon,” or “computer”). Once it hears that, everything in the following few seconds is perceived to be a command or a request, and it’s sent up to Amazon’s cloud computers…”)
I wanted to share my colleague’s message because *this is exactly how it is done, folks*. While I would’ve been just fine with them pointing this out as a comment to my LinkedIn post, they’re being polite and careful, because not everyone would be and frankly, they and I had one lunch at one time and that’s about all we know of each other.
My larger point — because I know that not everyone is in to public correction and many could find it disconcerting — is that we need to be better at private correction, at accepting new data, and at assimilating it or at least making the sincere attempt. You will read articles and they will be carefully constructed on the part of the author — either attempting to be scrupulously fair or attempting to sway you one way or another — but what you don’t get to see is what was omitted, either via editorial jurisprudence or a required word count or assumed common knowledge. What you don’t get to realize is what your brain has omitted, either via convenience, or simply the wear of time.
So thank you. I happily sit corrected :).