Economists have discussed it, and just like with previous automation, there'll always be new jobs as human wants are unlimited (and by the time AI are able to do every job humans can do, they won't be willing to do it for free).
Past performance is not indicative of future results.
> human wants are unlimited
Physical resources are limited, and producing work requires the allocation of capital. If people want to work but cannot access the basic tools and resources to work then they'll languish in unemployment. Society doesn't just default to a productive society with high employment; these conditions need to be nurtured. Given the current state of economic inequality we're trending away from conditions that will lead to full employment if AI wipes out large swathes of jobs.
Based on what evidence? Do you think better ML will create no new opportunity? What does economic inequality have to do with whether new companies or services are built using new ML tools?
The automation we've been working on for decades is aimed—often explicitly—at eliminating the need for human beings to do certain kinds of jobs. Any philosophy or policy that does not then say, "...and then we will use the extra productivity to support those people in perpetuity" is morally bankrupt as well as failing to recognize social and economic realities we've known for hundreds of years.
Everyone’s deeply up their own asses about AGI, existential risks, this will solve climate change, bias, etc etc.
Nobody seems to discuss the risk that millions of white-collar jobs with little clout (paralegals, admin clerks, logistics coordinators, some portion of software developers, etc) are eliminated.
I think the largest risk in the next 10-20 years is that the tech gets good enough to shed jobs without replacing them (certainly not at a 1:1 ratio), furthering inequality, social division, etc. in the US at least, it seems that any broad policy proposal to address this is DOA.