Replying to Avatar Lyn Alden

One of the crazy things about AI and robotics is that in the year 2025, most people still don't use Roombas or other robotic vacuum cleaners.

They're useful in many contexts, but they're not clearly better across most metrics than a human with a vacuum cleaner yet. They've been out for a very long time, gradually improving. And that's one *very specific* task with pretty clear visualization requirements and floor mobility requirements and pretty low safety thresholds with high repetition levels, and yet that market isn't dominated by robotics yet.

That's an example of why I continue to view white collar computer-work AI as being *way* ahead of in-the-field blue collar robotic AI in terms of competing with human jobs.

The moment where it's a joke to buy a human-powered vacuum instead of a robot vacuum, rather than a debatable trade-off, is kind of the canary in the coal mine moment for consumer robotics. We can't even nail that yet, but once we do, it's kind of a floodgate moment, considering how long that task has been in the works for, and it will probably quickly expand to other areas following that moment.

That's kind of my basic test for robot hype. Yes, they're getting better and better. Yes, they do backflips now. Yes, it's a big deal. But in-the-field blue collar skilled work is a really high bar, and we haven't fully cleared the "vacuum carpeted areas of the same house floor area over and over" stage of that yet.

Everything is kind of hype until that stage is fully breached. Then it's off to the races.

What's your view of that heuristic?

The Roomba enalogy isn't exactly fair. If you compare a roomba v.s. a hired maid, they aren't that much better than the other. In both cases you have do to 90% of the work by picking up and organizing before either of them are of any use.

Reply to this note

Please Login to reply.

Discussion

That's actually part of the problem.

Imagine a roomba that can identify objects on the carpet, move them aside safely, and vacuum. We solved the "clean a blank carpet" problem decades ago. We spent the whole time struggling with "okay, now deal with objects and frictions."

I have a maid, not a roomba. The maid cleans every surface in the house. She's my vacuum, my solid floor scrubber, my counter cleaner, my stove cleaner, my bathroom cleaner, etc. She uses her intelligence to work around questions like, "should I move this, should I clean this, or leave it be?" When we travel and aren't at home, we can give her instructions on how to let herself in. She even brings the mail in for us to check when we get back. When she was hit by a truck and couldn't work for months, we paid her full wage during that time despite no work, as though she had benefits, since we valued her work so much.

Roombas have made very little progress at replicating any of that, to bring that capability from the upper-middle class and the upper class to everyone. When they start to do that (maybe not all of that, but the basics), then I'd pay attention.

This brings up an important aspect of automation vs human labor: cost. In California we always paid a service to mow and blow the lawn, but in Sweden even well off people mostly do their own lawn care. Robot lawn mowers have really taken off in Sweden because labor costs are so high (mostly taxes tbh). I wonder if reducing imported illegal labor will not create more business for automation products. Even hiring a backhoe operator vs. team of day workers is a win for automation.

True, but it seems we are on the verge of making that breakthrough. Maybe that's the same kind of delusional thinking we've had for the past 80 years, that it's just around the corner, but it feels closer than ever before.