Part of the problem for them is that most of that work involves input that they don’t have a means to capture.
Think of manual work that becomes too difficult or impossible just by wearing gloves and dulling the touch sensation, and even then you still using how it feels as a significant input. Also output is far more complex and not instrumented, and camera footage doesn’t cut it.
This is why driving is, comparatively, easier for the machine learning approach. The vast majority of input is visual, some audio. The only outputs that matter are the turn of a wheel and actuating a couple of pedals.
They’ve been trying with remote operation (the glasses don’t do it, but remote operation ensures all input and output can be captured), but it’s just really hard to remotely do a lot of this stuff without just being able to touch and feel things through.
Open ended manual work is going to be actually trickier than “knowledge work”, and knowledge work isn’t exactly fantastic as it stands yet either.
Part of the problem for them is that most of that work involves input that they don’t have a means to capture.
Think of manual work that becomes too difficult or impossible just by wearing gloves and dulling the touch sensation, and even then you still using how it feels as a significant input. Also output is far more complex and not instrumented, and camera footage doesn’t cut it.
This is why driving is, comparatively, easier for the machine learning approach. The vast majority of input is visual, some audio. The only outputs that matter are the turn of a wheel and actuating a couple of pedals.
They’ve been trying with remote operation (the glasses don’t do it, but remote operation ensures all input and output can be captured), but it’s just really hard to remotely do a lot of this stuff without just being able to touch and feel things through.
Open ended manual work is going to be actually trickier than “knowledge work”, and knowledge work isn’t exactly fantastic as it stands yet either.