> Dancing is one that will be very hard to copy
I'm not sure I'd hang my hat on a particular current inability of AI, particularly inabilities where there's no obvious impediment to AI doing it. You're just begging for AI to eventually gain that ability, and now you're back where you started. Hooking AI up to physical actuators and getting it to work properly strikes me as a "this will 100% happen, I guarantee this is already being worked on" situation; basing your self-worth off the inability of AI to do those things seems like a ticking time bomb.
They don't require mining rare earth metals, they're more versatile, and uh... oh gosh it's only a matter of time before we're all replaced by cheap carbon based robots. It's been real <3
Similarly, people still play chess, go, and Starcraft even though AI can beat them. Human Jeopardy matches still get televised even though Watson could beat them. We still have math olympics even though Wolfram Alpha could have won all of them decades ago.
I think human dancers are safe for as long as there are humans.
For now. There's no technical reason why we can't have CNC painting machines - there's plenty of existing tech in manufacturing that has the required precision and range of motion. I think we don't see such machines because there is no business need for them - or at least, there wasn't.
To date, the kind of digital images you'd want to reproduce using a CNC painter would be drawn by actual artists, using their hands and - most likely - tools like pens and drawing tablets. The work of an artist is similar enough to actual painting that, if there's a need for the 3D texture, they can just... use real brush on real canvas. There technically is a market for cheap reproductions, but AFAIK this one is covered by "painter farms" - warehouses full of low-paid painters mass-producing paintings, in countries with cheap labor costs.
Now, if you could get the ML model produce "original" work, encoded as a time series of brush strokes (either straight from the model, or by another model turning JPEGs into painting instructions), then this just might provide an incentive for CNC painters. It might even start as a hobby thing - there's lots of people who would build such machine, the way they build 3D printers now, so they could produce their own paintings at home. Those same people will be unlikely to pay a far-away painter farm instead.
> Dancing is one that will be very hard to copy, because you need actual humans for it.
AFAIR that was already done, to a passable level, by Boston Dynamics, a few years ago. Sure, that was a one-off carefully choreographed stunt, but it demonstrated that 1) robotics can handle dancing just fine, and 2) the result is actually enjoyable to watch. The missing bit is software, but given the ML work being done in recent years, in particular all those video editing/generating models that start by parsing human movement from input videos, I think it won't be hard for someone to start generating dance moves.
Not to mention, the kid may enjoy making art, and his/her art may give others joy, even if it's not a profitable career. It can still be a fun hobby - synthesizers have been around for nearly half a century now, but people still play guitar :)