Earlier quoted context omitted.
It's a 1-in-a-million edge case.... I would hope people on HN at least would understand this stuff happens in software.
1 in a million happens about 4000 times a second on a single CPU running at 4ghz... Did you really want to talk concrete numbers? Because if so, I am wondering about the probability that Amazon's voice recognition mistakes random conversation as a valid entry in someone's contact list, as well as the other terms in this equation. I don't know if you work for Amazon, and if so I don't want to single you out specifical…
1 Wake up w “Alexa.” 2 respond to “send message” 3 respond to “To whom?” 4 respond to “[contact name], right?”
As an engineer (well, an AI researcher who used to engineer), that looks to me like they were not negligent and it was hard to predict background conversation would produce this unlikely set of inputs - it would be nice to see stats but this is the first time it is covered in the media to my knowledge. And as with the laughter story, they will now change the inputs to make it likely 1-in-a-trillion this will happen.
However, it does seem like Alexa etc. will have to be better about recognizing audio from TVs/conversations and stuff directed at it - and I am sure they are working on it.
PS comparison to CPU is not great obvs since it's about number of instances (how many times does "Alexa" get woken up by background audio - not 4ghz)