Hey Amazon, chat transcripts are personally identifiable information. Stripping them of user ids and declaring yourself in compliance will not hold up when you have people already being identified by accidentally leaked recordings.
Hey Amazon, chat transcripts are personally identifiable information. Stripping them of user ids and declaring yourself in compliance will not hold up when you have people already being identified by accidentally leaked recordings.
That's OK if users explicitly opted in, but I doubt Amazon warned users about that.
`Get an inside peek at the Amazon-sponsored university competition to advance AI. The Alexa Prize "socialbots" are in early development and understandably rough around the edges. Help students develop more engaging AI by striking up a conversation and providing feedback. To get started, just ask: Alexa what's a socialbot? Alexa let's chat`
The only interactions students would have access to was interactions that occurred via the chatbot
I don't think you're necessarily wrong, but that seems like a stricter standard than currently exists for removing PII from transcripts.
Consider how easily people get doxed based on tiny shreds of information they post.
Including usernames that could de-anonymize.
There was a story on HN in the last few days where just one side of chats with a home assistant device was enough to identify the person, his S.O., and their friends.