I think Jevon's Paradox[1] might apply to our relationship with AI. Jevon's Paradox states that the more efficiently a resource is used, the more the demand for it increases.
Thus with AI, the more efficiently human labor is used, the more demand for it increases? Thus one person is able to accomplish far more with the aid of robots, therefore the usefulness of labor increases. The out of work argument only works if there is a fixed amount of production that needs to be done and then there's no more to be done after that. However, if that were the case, then we'd have all we would need and no one would have to work because working more would just produce useless stuff that would not be used.
Jevon's Paradox states that the more efficiently a resource is used, the more the demand for it increases.
Jevon's Paradox is a historically contingent observation that held for a while for some resources and has later been unjustifiably treated as a natural law (on occasion). Just like Moore's Law was a historically contingent observation that held for a while.
There is no guarantee that Jevon's Paradox will remain paradoxical, and indeed I can already find counterexamples: electricity for lighting is used more efficiently in modern LED bulbs than ever before, and the effect is to reduce electricity consumption in facilities that switch to LED lighting. "But more efficient lighting prompts people to use lighting where they wouldn't have before!" True, some of the gains are lost that way, but not to the extent that it actually drives more total electricity demand than prior to LEDs. You need total consumption to increase following efficiency gains to get the "paradoxical" result.
"But lower electricity demand causes prices to fall, which leads to more electricity consumption elsewhere!" Also a plausible objection, but remember that you need a rebound effect strong enough to wipe out the original savings and then some for it to be an example of Jevon's Paradox. And we're not observing that.
Sorry for the terrible web site that doesn't let me post a direct link, but look at electricity consumption per capita for the United States here:
According to that, the most recent year with data (2013) actually saw slightly lower consumption per capita than 1998: 12,988.3 kWh and 13,154.8 kWh respectively. The all time peak was in 2008 at 13,663.4.
Jevon's Paradox is a simplified description of one small range of a dynamic equilibrium. It's a range that tends to appear often in practice, but it's very far from the entire history.
Everything depends on how fast automation will hit people. With time (and not total automation), society can adapt and you'll see Jevon's Paradox. But if it happens too fast, and nothing is done about it, you'll see people starving, and violent reactions, just like in the Industrial Revolution.
If it's accelerating, you can be certain it will become too fast at some point.
And then, you can't just wave the issue of total automation away - at some point we'll have to deal with no jobs being available, at least if we escape having something more pressing at the time.
I think the question of what happens if humans have no jobs left to do is extremely premature. Different stages of AI and automation are being conflated in to the same arguments.
Rather, the issue that needs to be considered right now is what happens, in the US, if there is a big disruption to a large group of jobs which have been fairly stable for decades in a very short period of time? Specifically I am thinking about accounting (imminent), lawyers (already happening), financial advisors (happening), insurance agents (happening), real estate agents (imminent), drivers (coming soon), travel agents (transition complete) and so on.
Society as a whole gains benefits, if not direct access, but perhaps at a cost of removing the socio-economic status of a large group of people. This, of course, has happened many times before.
The greater AI-employment questions should be answered after we are pretty sure humans are even going to survive 5, 10 decades from now.
Humans made it through Napoleon, the Bolsheviks, chattel slavery, WWII and various flu epidemics. I doubt we'll end ourselves in 10 years.
"Removing socio-economic status" SHOULD be a goal, to the extent that that status comes from rents ( which it usually does ) . Status seeking is a huge drain on us, and if production starts to go exponential, then the "economic" part fades in importance.
Thus with AI, the more efficiently human labor is used, the more demand for it increases? Thus one person is able to accomplish far more with the aid of robots, therefore the usefulness of labor increases. The out of work argument only works if there is a fixed amount of production that needs to be done and then there's no more to be done after that. However, if that were the case, then we'd have all we would need and no one would have to work because working more would just produce useless stuff that would not be used.
[1] https://en.wikipedia.org/wiki/Jevons's_paradox