Discussion about this post

User's avatar
Random Reader's avatar

> If your model of the future involves ‘robotics is hard, the AI won’t be able to build good robots’ then decide for yourself now what your fire alarm would be for robotics.

OK, I used to work for a robotics company, and I do think that one of the key obstacles for a hostile AI is moving atoms around. So let me propose some alarms!

1- or 2-alarm fire: Safer-than-human self-driving using primarily optical sensors under adverse conditions. Full level 5 stuff, where you don't need a human behind the wheel and you can deal with pouring rain at night, in a construction zone. How big an alarm this is depends on whether it's a painstakingly-engineered special-purpose system, or if it's a general-purpose system that just happens to be able to drive.

3-alarn fire: A "handybot" that can do a variety of tasks, including plumbing work, running new electric wires through existing walls, and hanging drywall. Especially in old housing stock where things always go wrong. These tasks are notoriously obnoxious and unpredictable.

4-alarm fire: "Lights out" robotic factories that quickly reconfigure themselves to deal with updated product designs. You know, all the stuff that Toyota could do in all the TPS case studies. This kind of adaptability is famously hard for automated factories.

End-game: Vertically-integrated chains of "lights out" factories shipping intermediate products to each other using robotic trucks.

In related areas, keep an eye on battery technology. A "handybot" that can work 12 hours without charging would be a big deal. But the Terminator would have been less terrifying if it only had 2 hours of battery life between charges.

The nice thing about robotics is that it's pretty obvious and it takes time.

Expand full comment
Daniel Donnelly's avatar

I think you have a couple of typos where you are referring to ‘George Hinton’ instead of ‘Geoffrey Hinton’

Expand full comment
43 more comments...

No posts