my take
we're heading toward a post labor society. not overnight, task by task. either the way we split income adapts with it, or this century gets ugly for everyone.
how it actually happens
jobs don't disappear in one shot. a job is a bundle of tasks. automate one, it's cheaper, so it sticks. then the next one. pull out enough of them and there's nothing left to pay for. each step looks small and profitable, so nobody stops it.
the loop nobody's solved
no labor means no wages. no wages means no buying power. no buying power means no customers, and no customers means no profits.
every company automating makes sense on its own. all of them doing it at once removes the people who buy the stuff. one firm's savings is another firm's lost revenue.
so the market won't fix this by itself. same as data centers: someone gets the benefit, someone else pays. that's what policy is for.
why the loop won't save us
i don't put too much weight on that chain though. automation goes task by task, and each step pays for itself on its own. call it a return on compute. wages come back to firms as revenue, but profit comes from a lot of places.
the mass market exists because mass labor does. if you own what does the work, do you need either? profit is a means, not the end.
so companies follow the money up market, to whoever owns the machines. that's the scary version. the economy doesn't stop, it just stops being about most people.
which is why i think consumer companies end up being the strongest lobby for ubi. they need people to have money to spend. the push might come from business, not just politics.
this has happened before, kind of
- keynes, 1930. predicted we'd work 15 hours a week by now. productivity showed up. the free time didn't. we kept inventing new work.
- new jobs. about 60% of us jobs in 2018 had titles that didn't exist in 1940 (autor). every wave of automation made work we couldn't have imagined.
- but the transition hurt. early in the industrial revolution, output per worker climbed for decades while wages stayed flat. economists call it engels' pause. it ended well, but a whole generation paid for it.
the difference this time: past machines did one thing. a general model can learn the new jobs too. that's what makes "we'll just make new work" less obvious.
where the money goes
- it's already shifting. the share of us income that goes to wages has been sliding for decades, from around 65% to the high 50s. ai speeds up a trend, it doesn't start one.
- ownership is concentrated. the top 10% own roughly 90% of us stocks. "own what does the work" means very few people today.
- the tax base breaks. governments run on income and payroll taxes. fewer wages means less tax right when more people need help.
ideas already out there
- resource dividends. alaska pays every resident a yearly check from oil money. swap oil for compute.
- a public fund. norway owns a huge wealth fund for its citizens. altman pitched a version that owns a slice of every big company and pays out to everyone.
- tax the machines. gates floated a robot tax. or tax compute, same idea as my data centers take: the users of the compute pay.
- human-preferred work. care, teaching, anything where you want a person. those get relatively more expensive as everything else gets cheap, and that's where work might move.
what has to adapt
- who owns the machines. if robots do the work, whoever owns them gets the income. does that stay with a few companies, or get spread out?
- income without a job. some way to get money that isn't tied to working. ubi, dividends, something we haven't named yet.
- what school is for. people are still training for the old world. degree, then a job. learning for what, if the job isn't there?
- what people do all day. work is a lot of people's identity and structure. take it away and you need something to replace it.
when
digital work goes first, it's already moving. then structured physical work: warehouses, factories, trucks on known routes. messy physical work goes last: homes, care, trades. that's the hard part of robotics.
guesses i've heard range from post labor for digital work by 2028, to humanoids taking a decade, to not for 30 to 50 years. nobody agrees on when. almost nobody argues it isn't coming.
the pushback
people say it's two arms on a flat table in a controlled sim. that robotics needs deterministic outcomes. that robots can't do anything that isn't set up for them.
fair on timing. a sim benchmark isn't a home. but it doesn't change the direction. and a decade is nothing, it's one career stage.
what i'm wondering
- if each automated task is profitable, where does the money go when nobody's wage is in the loop?
- is "we'll invent new jobs" still true when the machine can learn the new jobs too?
- can we skip the engels' pause this time, or does a generation pay for it again?
- do we adapt before it hurts, or only after?
- what's the version of this where most people end up better off?