Solving smart assistants’ human problem
The tech giants have taken these steps following recent news stories that contractors who check the quality of the service these assistants offer have heard sensitive things like sex, drug deals or private medical information.
Google is facing a data protection investigation about the practice of passing recordings to contractors (potentially a GDPR violation), and the likes of Amazon, and Microsoft must be considering their next steps too.
The fact is though, that you can’t have a good quality voice assistant without some human input at the assistant’s end occasionally . Tech companies need to check the quality of of their assistants’ output using real-world examples. Without that, the assistants will likely become less ‘smart’ and useful for some users over time.
How might this be resolved? Not using contractors, and instead having direct employees do the listening could solve technical data protection issues, but what about the 'creepy factor?’ The fact that humans listen at all is a dealbreaker for some users.
Apple says it will in the future give users the option to opt into the process it calls 'grading.’ That’s one solution, and one we may see replicated elsewhere. It doesn’t necessarily tells users which recordings have been listened too, though.
Another solution would be for users to actively choose to send a recording to quality assurance each time they’re not happy with a result, but not many users would remember to do that, even if they’d be okay with it in principle. The other option would be for assistant to speak up and ask the user for permission each time one of their recordings is selected for listening. But how many people would say yes?
Whatever the eventual solution, it’s likely to break the illusion that there’s a super-smart assistant sitting inside that box in your kitchen.