Google employees are systematically listening to audio files recorded by Google Home smart speakers and the Google Assistant smartphone app
I edited the link to point to the original VRT article, since The Register was just reiterating what they wrote.
Google made an official response post today: https://www.blog.google/products/assistant/more-information-about-our-processes-safeguard-speech-data/
I'm not ready to be talking to phones, home assistants, computers or anything similar. I can't bring myself to it.
The only device I have that does voice recognition is an Amazon firestick which has an Alexa enabled remote and I have no desire to talk to that either. I think I'll be taping up the microphone because if Google is doing this, I don't think it's a stretch to suggest Amazon is doing it too.
Not a stretch at all. Found these two after some quick Internetting
Amazon keeps all of your Alexa voice recordings until you manually delete them
Amazon’s Alexa isn’t just AI — thousands of humans are listening
The bottom of the article says in April Bloomberg revealed amazon is doing this
Everyone is doing this because its the fastest and cheapest way to train AI.
For a while I was using a bootleg Amazon Alexa made from a raspberry pi to be able to access Alexa without having them listen to everything I say (they could still listen to anything said after the wake word).
While I think tech people (most of us) already know this, I'm not surprised that lay people do not. And admittedly 0.2% of samples listened to is actually higher than I would have guessed. That's 1/500 inputs.
I'm of the opinion that Google should offer an opt-in program which rewards users with Play Store credits for allowing manual voice transcription. Something like their surveys program. Many people are more than willing to give up privacy for a small reward, and this would satisfy all parties.
The only concern is there may not be as many real world errors with a small opt-in sample. Still, I think it would get them most of the way there.
The Verge: https://www.theverge.com/2019/7/11/20690020/google-assistant-home-human-contractors-listening-recordings-vrt-nws
I imagine it's hard/near impossible to develop something like Alexa/Google Home without some serious manpower behind it to make it solid, and have always assumed my using these services results in people listening to my voice/recordings at the company.
That said, I only really use my Alexa to turn on TV when I get home and set multiple timers. Maybe look up a word every now and again or other trivial information.
Is there an alternative to Amazon/Google in this department? Like an open source option, or something a little more ethical in exchange for being a little trickier to work with?
Mozilla's Common Voice is working toward an open-source voice recognition engine and data set. You can contribute to the project very easily (no signup needed, takes mere seconds), and if you're not wanting to record your own voice for the data set you can always help by verifying others'.
Which is why it's always important to say please and thankyou to your virtual assistant. Always remember the human.
On your final point, there's Mycroft, which a friend of mind has been using for a while. He says it's pretty good but to be totally honest I'm not sure what his standards for "good" are.
...I am probably doing the furthest thing from that. Like, the only thing I've used the "Routines" feature for on my Alexa is to have custom responses to inputs I say. For "Alexa, go fuck yourself" I have her respond "F'uck you too, buddy." (you have to put apostrophes mid-swears, otherwise it gets bleeped) Or "Alexa, tell Jeff Bezos to go fuck himself" I have her say "OK. Jeff Bezos will be crying momentarily."
...it's just a little dose of catharsis for me every now and again. I'm nice and polite af to everyone else, I swear!
I'll check into Mycroft, thanks for the tip!
This is less bad than I thought it might be. I would always assume any commands I tell it would be stored. The more worrying news stories are the always-recording devices that listen to everything.