Sitting at a coffee shop and writing this, I’m overhearing a conversation between six men over the course of an hour. It meanders from sexual harassment, marijuana legalization, paying off a debt and gambling. It’s because I didn’t charge my noise cancelling headphones and am now stuck listening to the “ambient” environment. It’s cognitively demanding to keep up with the discussion and to figure out the social dynamics — and it’s a distraction.
It makes me think about the biggest concern around voice first products — that they’re listening in our homes. If you hired someone to listen in on a home for an entire day, it would exhaust them quickly. They’d likely quit after a short time.
The real issue with mass surveillance is the amount of noise it generates. First just think of the data and storage required for hundreds of hours per household per day. Second, the analysis of this data beyond real time speed is very power intensive. It requires voice separation, diarization, transcription, natural language understanding (to flag intents). It could potentially require multiple languages and noises… it’s a huge mess.
Target eavesdropping makes more sense and those who might be concerned that information gained through eavesdropping might be used against them should exercise caution against any device that isn’t secured by them — this includes Echos and Homes and third party implementations of Google Assistant and Alexa as well as any cell phone within hearing range. We completely forget that these devices are even easier to hack than newer technology.