Art and poetry are just 0s and 1s. Generative AI takes 0s and 1s as input and produces 0s and 1s as output. It's a math problem. Mining ore is a physical task. Getting a robot to perform a physical task is complicated because the computer needs to track all kinds of things we do naturally.

You can close your eyes and touch the tip of your nose because you have proprioception. A robot doesn't have that. It needs some kind of way to determine where parts of it are in 3D space. Do you put GPS on those parts? LIDAR? RFID tags? The more sensors you add, the more expensive the robot becomes. The less sensors you have, the more guesswork is being done in the calculations, and the more likely it will fail at the task. Either you have a robot the can do the work autonomously, but costs more than a human labourer, or you have a cheap robot that needs the constant supervision of a human labourer.

Automation is less autonomous robot and more machine that has been purpose built to do one hyper specific thing and I think it'll be like that for a while. Sarah Davis Baker has a great video on the topic of the robot body and the difficulties of reducing reality to a stream of sensor data.