Weeks later, out of curiosity, I requested the transcripts of everything Sensi was recording in my father’s home. Reading his personal conversations, I suddenly felt like a spy, with the device as my silent conspirator. I’d pushed for the thing in the first place, but now I felt uneasy about it. My father, meanwhile, didn’t remember being told that Sensi was eavesdropping on his conversations.
Reading his own words back to him, I braced for the worst.
“So, what do you think?” I asked.
There was a moment of silence in which I could hear the blood swirling in my ears.
“Well,” he finally said, sounding on edge. “It’s pretty weird that it hears words.” He seemed baffled that anyone would deem his conversations worthy of transcription.
“But I guess it’s worth it,” he added, before changing the subject.
This story is part of The Future of Home, a collaboration between the editors of WIRED and Architectural Digest to help you understand what “home” will look like tomorrow and beyond.
After my dad’s shrug of acceptance, I started digging into what I’d actually put in his house. Sensi, I learned, is one of a growing number of AI devices aimed at seniors: Earzz and Ally Cares surveil care home residents for coughs, falls, and atypical movements, while Cherish Serenity—which looks like a stylish, retro home speaker—uses radar to detect when someone in a room has fallen or is slumped over. (The device can be bundled with AT&T for speedy emergency response.)
Unlike Alexa, these devices don’t wait for someone to say “help.” Instead, they begin recording after specific events: sounds like thuds, coughs, screams, and movements like falling off a bed. In Sensi’s case, the device doesn’t even tell the senior it’s recording, which helps explain my dad’s confusion.
Sensi’s algorithm, allegedly based on “1,000 years” of audio data, claims to identify deviations in a person’s usual routines. If you’ve got a new cough, are always in the bathroom, or are simply plodding around the house in a novel way, Sensi can apparently tell. But when I asked the company’s cofounder and CEO, Romi Gubes, how the algorithm was built, she said only that its models are “trained on anonymized datasets” stripped of “personally identifiable information.” She did not elaborate on exactly what those datasets contain or where they are pulled from.
Steve Kamau, a calm, soft-spoken coordinator at Husky Senior Care, the agency helping my dad with shopping and other household tasks, tells me that the device sometimes works exactly as it should. In one case, a senior fell down while trying to reach the toilet when no caregiver was around. Sensi caught both the sound of the impact and the man’s cries for help. Kamau called the client (who always kept his phone on him) and confirmed he’d fallen, then dispatched 911; the man was then helped off the floor. In another instance, he says, a client’s cough was caught early enough that it may have saved her from a more serious illness. (Sensi claims a 90 percent accuracy rate, with edge cases reviewed by a “human in the loop”; Kamau tells me the system has also mistaken a dropped remote for a fallen senior.)
