ONLY A FOOL would currently entrust its life to the terminally dim Alexa, lest your shouts for help be interpreted as an urgent requirement for early Beatles on Spotify.
But researchers from the University of Washington still see potential in always-listening devices, and have managed to train a system to spot the early audible signs of cardiac arrest.
The researchers trained the system with 911 samples - the emergency line, not the 90s boyband - where it learned to listen for familiar sounds of agonal breathing that tend to accompany cardiac arrest. Its first response is to sound an alarm, asking for help from anybody nearby. If nobody responds, it then takes matters into its own hands and dials an ambulance without human involvement.
That could all get quite messy if it routinely made mistakes - although you'd imagine the mistaken subject would respond to the calls for help - except the system is unerringly accurate. In the researchers' tests, it only misidentified breathing 0.22 per cent of the time in single events, and that dipped to zero in events where it listened for events ten seconds apart.
It's proven so successful, that the scientists have founded a company - Sound Life Sciences - that hopes to sell the technology and save lives in the real world. They're likely best off looking at the health sensors market, because selling a technology that requires the device to be permanently eavesdropping without a wake word would be a tough sell for the Amazon Echo or Google Home, both of which face pretty serious questions about privacy as it is.
That's a pity, as the ability to save lives is one hell of a selling point. You don't get that with YouTube Premium. µ
Now you can watch documentaries about horribly disfigured people whenever you like
Brad to the bone
Being in a minority of one doesn't make you right
WeWork needs a rework