Part of it comes down to what AI can do physically.

I suspect that we'll see some more displacement in the knowledge fields well before we see ones requiring physical work. The adoption of legal management systems for searching and updating contracts and other legal filings has taken a big bite out of the demand for lawyers and paralegals already.

Because if the physical jobs were amenable to automation, many more of them would have been already. In many cases, the bottleneck there hasn't been a lack of AI. Current software and computers are well up to the task already in many instances. The cost/effort required to actually manipulate a physical process effectively with automation is the big factor.

There's edge cases where (weak)AI and machine learning can make the difference, and it'll creep into physical work tasks eventually. But it's more a factor of economy of scale for the physical robots, and the entire logistics of reworking a business, factory, a farm, a mine, etc. than it's an issue of "not smart enough."

And it's difficult to predict what exactly AI will make more efficient, eliminating jobs, vs. creating entirely new ventures and industries that require more employment.

Take the internal combustion engine and vehicles. The automobile in every aspect employed orders of magnitude more people than the stable hands, hay farmers, horseshoe blacksmiths, and street sweepers it displaced. Car salesmen, mechanics, oil/gas workers, road construction, automobile factory workers, engineers, designers, the list goes on...

On the other hand, farmers went from about 30% of the US population by employment to roughly 2% today. And of that 2% today, that's "agriculture related," so some are presumably doing things like operating grain elevators, various technical things, or whatever. And aren't directly "farming." All because of the tractor, and other machines like the combine.

Those who claim that AI and automation can and will displace almost everyone eventually aren't completely wrong, but the concerns of how fast it'll happen and how the economy will react are overblown. And there's a fundamental issue that in a world with no employment, the cost-basis for goods starts approaching zero. Hyperdeflation.

The only real economic scarcity could become energy. And there are solutions to that already. At least technical/scientific ones. The barriers are largely political/social in nature. Meltdown proof, disaster-proof fission reactor designs are known. And add to it fast breeder reactors tech, mining ever more Uranium isn't an issue. Aggressive fuel reprocessing solves a great deal of the nuclear waste problem. Neutron activation/deactivation of secondary wastes solves more.

High capacity energy production from nuclear can solve many more problems. Mass desalination of seawater for agriculture and drinking in arid regions. Massive recycling of wastes that's currently too expensive. Even direct conversion of plastics and organic-carbon waste to produce carbon-neutral petrochemicals for the million-odd things we use them for besides fuel are possible.

And that's just fission, if one or more of the dozen-odd fusion methods being pursued pays off, or the trend of ever cheaper and more efficient solar panels keeps going, there's more sources of energy coming too.

First-world living standards seem to be well tied to shrinking non-replacement birthrates and population contraction.

Free or nearly free goods, ever higher technology, and a shrinking population that doesn't even need as much food, goods, energy, or raw materials... we may ultimately be looking at the wrong problems.

And I'm sure some will grumble that they feel it's inevitable. There's going to be a high-tech billionaire class and a world of serfs coming. Although it seems to me that economic oppression just for its own sake is a potential false premise. Even beyond the basic zero-sum game fallacies. Cynical enlightened self-interest alone could conceivably prevent it. A world where everyone lives in comparable comfort and luxury is a world not inclined to do the whole "pitchforks and torches"-thing.

In the meantime, we'll have to wait and see how many "Automobiles" vs. "Tractors" come out of AI and machine learning.