3 ms·
The reason a human reviewer needs that information is to troubleshoot the response. 1. Customer makes a request that requires location — “Alexa, what’s the wea
by FakeComments 7y ago
The reason a human reviewer needs that information is to troubleshoot the response.
1. Customer makes a request that requires location — “Alexa, what’s the weather?”
2. Alexa gets the response wrong — “A turkey is a flightless bird.”
3. You flag that as incorrect in the application.
4. An Amazon employee reviews that flagged response, and needs to figure out how Alexa misinterpreted — did she mishear you? ...is your location data malformed, so that interpretation was discarded? etc.
I don’t see how Amazon employees can troubleshoot the Alexa application without both reviewing the audio files and the contextual information provided.
Disclaimer: at one point, I reviewed security for Alexa teams. I no longer work there, my opinions are my own, etc.
- JohnFen 7y ago> I don’t see how Amazon employees can troubleshoot the Alexa application without both reviewing the audio files and the contextual information provided. From my point of view, the issue isn't really that Amazon does this. The issue is that Alexa users are surprised that Amazon does this. Their surprise indicates that no real consent was obtained. If Amazon had done something like, when a user flags an incorrect response, asking the user for permission to use that data, I would not have any issue with it.
- FakeComments 7y agoI completely agree that surprising users is always a problem, precisely for the reasons you state. I just hoped to give a better idea of how it came to be people who (in my biased experience) are thoughtful about user privacy, ended up bringing together a data set users were uncomfortable with: Of course the team reviewing responses needs to see the user context! Whoops.