My fiancé works at Anthropic.
asserted
fiancé → work → Anthropic
One of my main preoccupations this year has been the extent to which artificial intelligence threatens jobs, either now or in the future.
asserted
intelligence → threaten → future
Our Platformer podcast miniseries took the question to seven experts with a variety of perspectives, and the debate ended mostly in optimism — with most guests casting doubt on the idea of mass long-term unemployment, even as they acknowledged that AI will likely cause most jobs to change dramatically.
asserted
jobs → take → unemployment
Conversations around AI and jobs can often feel slippery in part because the data we have on the subject is mediocre.
asserted
we → feel → subject
Jobs are rarely destroyed for any one reason; corporations often have an incentive to lie about why they’ve eliminated a job; and much of the data we’ve seen this year can seem contradictory.
asserted
we → destroy → data
For these reasons, I took note this morning of “We Must Act Now,” a statement signed by more than 200 economists, AI researchers, and Nobel laureates.
asserted
We → take → economists
Other signers included executives from Anthropic, Google, and OpenAI.
asserted
signers → include → Anthropic
The statement is short enough that I can reproduce it here in full:
AI may become radically more powerful over the next 10 years.
uncertain
AI → reproduce → years
This could drive an unprecedented transformation of our economy, larger than the Industrial Revolution, but unfolding over a vastly shorter time frame.
uncertain
This → drive → frame
It could bring risks, including large-scale job displacement, as well as opportunities such as major gains in living standards.
uncertain
It → bring → standards
Economists, policymakers and technology leaders must act now to understand the economics of transformative AI and to build the incentives, guardrails, and institutions needed to steer AI in a direction that complements humans and benefits society.
asserted
that → act → society
Like most statements that 200 people are willing to sign, this one can read as fairly anodyne.
asserted
one → sign → statements
But as Ben Casselman notes in the New York Times, among other things it represents a cohering view among economists that the risks of large-scale disruption are growing:
asserted
risks → note → disruption
Notably, the list of signatories includes some people who have in the past been prominent A.I. skeptics, including Daron Acemoglu and Simon Johnson, professors at the Massachusetts Institute of Technology who won the Nobel in economics in 2024.
asserted
who → include → 2024
“There’s been a notable change in the profession,” said Erik Brynjolfsson, a Stanford economist who helped organize the statement.
asserted
who → say → statement
He said his goal was to get economists and policymakers to take the disruptive potential of A.I. more seriously.
asserted
economists → say → A.I.
“I still see a big gap there, a big mismatch, and I’m kind of worried that we’re not going to be ready for the tsunami that’s coming,” he said.
asserted
he → see → tsunami
The new statement comes a month after a Wall Street Journal survey of 16 leading economists found that half said AI would lead to no net change in jobs; five said it would lead to net losses, and just three said it would lead to net growth.
asserted
it → come → growth
Notably, China last week opted not to set a numerical target for the number of urban jobs it would create in the next five years — the first time it had not set a target since the 1990s.
asserted
it → opt → 1990s
At times like this I find it helpful to slow down and ask what we know for sure, what’s disputed, and what we might do about it.
uncertain
we → find → it
Start with what we know.
asserted
we → start → what
“The occupational mix is not yet changing in ways that clearly align with the introduction of AI into the workforce” the Yale Budget Lab, a nonpartisan policy research center, reported in June.
asserted
Lab → change → June
“Measures of AI usage show no connection to changes in employment or unemployment.”
asserted
Measures → show → employment
Two, there’s growing evidence that AI is beginning to erode work — in particular, entry-level work.
asserted
AI → ’ → work
This month, Stanford’s Canaries Dashboard — another Brynjolfsson project — offered two relevant data points.
asserted
Dashboard → offer → points
(The data comes from ADP, a large provider of payroll services.)
asserted
data → come → services
The first is that jobs that are considered the most “exposed” to AI shrank 0.5 percent, while the least exposed grew 0.2 percent.
asserted
least → consider → AI
The second is that early-career jobs shrank 2.7 percent this year, while jobs for mid-career workers (ages 35-40) grew 1.6 percent.
asserted
jobs → shrink → workers
And they also help me make sense of the apparently contradictory data we sometimes see on jobs.
asserted
we → help → jobs
Take, for example, a seemingly happy story about AI and jobs from last week: US job postings related to software development are up 15 percent since the launch of Claude Code in February 2025.
asserted
postings → take → February
The implication is clear, and worrisome: employers are increasingly turning to AI systems to do the entry-level work that their junior workers once did.
asserted
workers → turn → that
For the moment, that seems to be creating jobs for more senior workers.
asserted
that → seem → workers
But what will happen should AI systems be able to do those tasks, too?
asserted
systems → happen → tasks
Job losses often have multiple causes, and economists have cited interest-rate hikes, overhiring during the pandemic, and remote work as reasons that hiring is slow in some parts of the economy.
asserted
hiring → have → economy
Economists are also mystified by how AI could be significantly contributing to job losses when we don’t yet have measurable productivity gains.
uncertain
we → mystify → gains
And so when the signatories of this new letter say we “must act now,” what do they mean?
asserted
they → say → what
The letter doesn’t include any specific ask.
asserted
letter → include → ask
Brynjolfsson told the Times that one of his highest priorities is getting better data.
asserted
one → tell → data
“The lack of reliable data has been a major obstacle for researchers in recent years,” Casselman notes, “with different measures telling conflicting stories about whether A.I. is leading to job losses and which workers will be most affected.”
asserted
workers → note → losses
The good news is that the US government has begun to think through some solutions: there is now bipartisan support for a sovereign wealth fund paid for by AI companies, for example.
asserted
government → begin → example
…and 37 more, not listed.
The Aporia
