Whisper Thine Name
It happened the other day. My daughter came up to me and wanted to refer to Alexa without the Echo triggering so she whispered her question to me. This would have been an out-of-context trigger. It’s not quite a false trigger but it is still an unintended one.
Avoiding an AI from listening in on us is not a new concept. 51 years ago, Stanley Kubrick depicted HAL 9000 going rogue and the crew trying to hatch a plan, only to be foiled because HAL could read lips. Now, we have mute switches and lens covers but remain oblivious to the three plus mics we carry around, plus many others in Bluetooth headsets.
We’re probably going to have to take an all or nothing approach to avoiding AI assistants doing stuff around us and listening to us all the time (and streaming the audio). Maybe there will be a house-wide switch to power off all mics and potentially suppress all outbound voice traffic. When privacy mode is off, all our audio might be streaming to an edge process that not only does some verification of a local trigger but also processes for context.
The other method is to push even more context processing to the local process so that context can be done on the device. Maybe this means less ubiquity of voice but a better interaction where it does exist. There is still another way.
Open the flood gate doors. Allow for total streaming of voice data. BUT sign the data from the mic itself and use something like a blockchain for everywhere the audio passes through. This way, you know everything that’s happened to your recorded voice and have more transparency. Then, you won’t need to worry about whispering.