I am also having trouble separating out what I would consider to be actual AI from AI used in marketing copy. A lot of startups are now claiming to offer "AI" to solve problems when they are effectively not doing anything differently than other programmers have done in years past.
Here is a study that finds a number around half of all jobs being replaced: http://www.oxfordmartin.ox.ac.uk/downloads/academic/The_Futu...
With these advances, you could easily train multipurpose robots to perform all manner of tasks. The industries that will be decimated first are all the minimum wage industries. Stocking grocery store shelves, deliveries, moving things around the warehouse, laundry, folding clothes, all things retail, etc.
The tech necessary to replace humans in all of those jobs already exists. The only remaining question will be if we allow robots to replace those jobs or not.
Some what further in the future, more skilled work like construction, plumbing, welding will be done by robots. Hell medical care and counseling will be done by robots.
Given sufficient mobility and energy there is no limit to what can be achieved with AI. It's only a matter of time.
ML is not to the point where we can hand it a bunch of data, some features, and have it control robotic arms with near human ability. Not even close. There just isn't enough data or computational power (in a small enough package anyway); not to mention robotics are still really in their infancy, we've barely got a four legged robot jumping over a single object head on.
I think you're really underestimating energy and computational requirements; these systems would need an umbilical cord to a datacenter and power to even function. This isn't even discussing the need for them to learn, which would be a requirement for more complex jobs like folding clothes or moving through unmapped environments.
I think automation has been put forward as a boogieman to those who don't know better so UBI can be pushed forward. We aren't even close to a time when machines are able to repair themselves, or function semi-autonomously in an environment not strictly built to support it.
We're going to be in this kludge for a long long time. Sure unemployment will go up with automation, but it'll be a slow plotting course with a lot of stopgaps put in place. Eventually when that becomes burdensome enough we may revisit UBI, but we haven't reached that point yet, not even close.
I think the point that I'm not really understanding is where the commodity-level multipurpose robots come from. It's one thing to take an existing Toyota Camry and trick out it's computer systems with deep learning and advanced algorithms and sensors. It's another thing entirely to have multipurpose robots everywhere that can replicate not only some degree of low level human intelligence but also key human physical characteristics (senses, dexterity, strength, etc.)
I am sure (and really hope) that we get there soon but it seems like a 2 or 3 decades away at a bare minimum.
FoldiMate https://www.foldimate.com/
Go ahead and sign up for pre-order if you like.