Will we know when AI is taking our jobs?
AI is making economists get creative
We’re taking The Argument to San Francisco! On May 13, Kelsey Piper and Jerusalem Demsas are debating a question that feels unavoidable right now: Is AI actually changing how science gets done, or are we in the middle of a very expensive illusion? Jerusalem is bullish; Kelsey is skeptical.
And you won’t just be watching. You’ll get to join in on the argument, too.
Join us May 13 at The Chapel from 7 to 10 p.m. Come argue with us! RSVP here.
AI’s impact on jobs could happen really fast, but economists’ ability to tell us what’s happening in the labor market is often delayed.
You might be used to hearing about employment numbers regularly: New jobs data comes out from the Bureau of Labor Statistics on the first Friday of every month, and it appears in news stories about the unemployment rate and the number of jobs gained or lost. But if you read the news and learn that the tech sector lost 57,000 jobs in the last year, that doesn’t necessarily mean that 57,000 software developers were displaced by AI coding agents.
Monthly jobs data tells us what happened — it does not tell us why it happened. To isolate causes, economists often need to wait for more data to accumulate.
“National statistical systems are not designed to track task-level AI reallocation at monthly cadence,” economist Pedro Serodio recently wrote on X, introducing his report on AI’s labor market impacts in the U.K. “The unprecedented speed of [AI] adoption means that aggregate figures may take a long time to reflect real-world impacts.”
Even one of Anthropic’s recent economic reports opens on a note of humility: From offshoring to the China Shock to the automation of factories, the report warned, the effects of past economic disruptions are still being hotly debated years after the fact.
Analyzing this disruption as it happens is a major challenge. It is not impossible, though.
If economists could collect data that more closely reflects real-world conditions, they could tell us more about what will happen, rather than what has happened long after it’s over. That’s why Alex Imas, an economist at the University of Chicago, has called for a “Manhattan Project” for more real-time economic data, and his sense of urgency is shared by other economists I spoke to for this story.
From scraping internet data to interviewing workers regularly, economists are attempting novel methods to keep up with change in the labor market. Below are the ones that I found most promising.
What is job “exposure” really?
So far, the most common measures have focused on AI’s capabilities — what tasks it could do — and not how it’s used in the real world.
The foundational paper on how AI will affect jobs was published by OpenAI in 2023, called “GPTs are GPTs.”1 Its headline finding was that 80% of jobs could have at least one-tenth of their tasks exposed to AI, and 19% of jobs could have at least half of their tasks exposed.
People hear “exposure” and think, “This job is disappearing tomorrow!” but it’s more complicated than that.
Exposure is measured by taking a job, breaking it down into a list of tasks and then seeing whether AI can speed up completion of that task by 50% or more.
If a task can be sped up by 50-plus percent it is considered exposed. If a job has a lot of exposed tasks, well, it tends to be considered at risk.
This is not how automation works in the real world:
Keep reading with a 7-day free trial
Subscribe to The Argument to keep reading this post and get 7 days of free access to the full post archives.