画像クレジット: Image via source article. Used under fair use for news commentary. · source
So a company wants to send people into your home, clean it for free, and film everything to train robots. Sound too good to be true?
Look, I've been around long enough to know that when something sounds too good to be true, it usually is. This week a startup called Shift announced they'll clean New York apartments at no charge. The catch, as The Verge reported, is that cleaners wear cameras and the footage gets fed into AI systems. "You get a spotless apartment. We get training data. Everyone wins."
I'll be honest. I've spent enough years watching robotics companies chase the domestic automation dream to be skeptical of anyone claiming everyone wins.
Here's the thing about training robots to do household tasks: it's genuinely hard. When I was at Kuka, we had a running joke that industrial robots could weld a car chassis to micron tolerances but couldn't fold a towel to save their lives. Structured environments with predictable objects? Easy. Your cluttered kitchen with seventeen different types of containers and a cat that knocks things over? Nightmare.
The big players have been throwing money at this for years. I called my old colleague Frank who's still plugged into the automation circuit, and he confirmed what I suspected. Everyone's hungry for domestic manipulation data. Real footage of humans doing real tasks in real homes. Not lab setups, not simulations. The messy reality of someone scrubbing a burnt pan or figuring out which cleaning product goes on which surface.
関連記事
More in Consumer
A startup called Shift will scrub your apartment for free if you let them record everything. I've seen this playbook before.
Mark Kowalski · 8 hours ago · 5 min
Two massive discounts on WD Black and SanDisk drives suggest the storage market is more complicated than the headlines suggest.
James Chen · 3 days ago · 4 min
I've been tracking these sales for years, and most of the 'deals' aren't worth your time. Here's what I'd actually buy.
James Chen · 4 days ago · 5 min
The holiday sales are wrapping up, and after digging through the discounts, I found more marketing than actual savings on robotics hardware.
Shift's approach actually makes a certain kind of sense from a data acquisition standpoint. They're essentially paying for high-quality training data with free labor. The economics might even work out if the data is valuable enough. And based on what companies are reportedly paying for robot training footage these days, it probably is.
The Verge's coverage mentions cleaners wearing "awkward-looking hats" with cameras attached. I've seen similar setups at trade shows, those head-mounted rigs that capture egocentric video for imitation learning. The technology isn't new. What's new is putting it in private homes at scale.
Shift claims they have privacy protections in place. I don't doubt they've got lawyers who've written something up. But here's what I know from decades of watching this industry: data has a way of sticking around longer than the companies that collect it. Startups get acquired. Servers get breached. Policies change. That footage of your home, your possessions, your daily routines, it doesn't just disappear when you decide you're uncomfortable with it.
And we don't know yet what happens to that data long-term. The company is young. Their retention policies, their security practices, their plans if they get bought or go under, all of that remains unclear.
I'm not saying Shift is acting in bad faith. I genuinely don't know. But I've seen too many "move fast and break things" approaches in this industry to assume good intentions translate to good outcomes.
Maybe. Probably, even. The data bottleneck for domestic robotics is real, and this is a creative way to address it.
But there's a gap between collecting data and deploying useful robots that the promotional materials tend to gloss over. When I think about the manipulation challenges involved in, say, loading a dishwasher (something I still do myself, thank you), the problem isn't just "we need more video of humans doing it." It's that every dishwasher is different. Every kitchen is different. The generalization problem is, well, it's a hard problem.
Some researchers I've talked to think foundation models will crack this. Others are more skeptical. I'm somewhere in between, which is probably a sign of getting old.
What I do know is that Shift isn't the only company pursuing this strategy. The Verge reported this week that multiple tech companies are now actively seeking ways to film domestic tasks. The race for household data is on, and ordinary people's homes are the resource being extracted.
I've watched industrial automation transform factories over thirty years. The productivity gains were real. So were the job losses. So were the safety improvements. It's complicated.
Domestic robotics will be complicated too, assuming it ever actually works at scale. And companies like Shift are part of that story, whether they succeed or flame out.
For now, though, I'd think carefully before letting anyone film the inside of my home in exchange for a free cleaning. The data you give away today might power the robots of tomorrow. Or it might just sit on a server somewhere, a permanent record of your private life owned by whoever buys the assets when the startup folds.
Call me old-fashioned, but I'll keep doing my own dishes.
