Worried About Privacy at Home? There’s an AI for That
WIRED, January 21, 2020
By Clive Thompson
“How edge AI will provide devices with just enough smarts to get the job done without spilling all your secrets to the mothership.”
Alexa, are you eavesdropping on me?
I passive-aggressively ask my Amazon Echo this question every so often. Because as useful as AI has become, it’s also very creepy. It’s usually cloud-based, so it’s often sending snippets of audio—or pictures from devices like “smart” doorbells—out to the internet. And this, of course, produces privacy nightmares, as when Amazon or Google subcontractors sit around listening to our audio snippets or hackers remotely spy on our kids.
The problem here is structural. It’s baked into the way today’s consumer AI is built and deployed. Big Tech firms all operate under the assumption that for AI to most effectively recognize faces and voices and the like, it requires deep-learning neural nets, which need hefty computational might. These neural nets are data-hungry, we’re told, and need to continually improve their abilities by feasting on fresh inputs. So it’s got to happen in the cloud, right?
Nope. These propositions may have been true in the early 2010s, when sophisticated consumer neural nets first emerged. Back then, you really did need the might of Google’s world-devouring servers if you wanted to auto-recognize kittens. But Moore’s law being Moore’s law, AI hardware and software have improved dramatically in recent years. Now there’s a new breed of neural net that can run entirely on cheap, low-power microprocessors. It can do all the AI tricks we need, yet never send a picture or your voice into the cloud.
About the Author:
Clive Thompson is a WIRED contributing editor.