AI EXPOSURE DOES NOT MEAN JOB LOSS

An economist bet $1,000 that the jobs most 'exposed' to AI will employ more people by 2030. Not fewer.

He was responding to a week when social media decided 40% of the American workforce was about to be automated away. The source? A two-hour weekend project by Andrej Karpathy that scored 342 occupations on how digital they are. Software developers scored 8–9. Roofers? 0–1.

Then Twitter did what Twitter does.

Karpathy's own .readme notes said the scores don't account for demand elasticity, latent demand, regulatory barriers or social preferences for human workers. Most people didn't read the .readme.

The panic which ensued revealed a category error that matters for every organisation. Exposure measures how much of a role involves tasks AI can touch. Displacement measures whether that role shrinks. These are different questions. Often with opposite answers.

Chicago Booth economist Alex Imas put it bluntly: AI-exposed jobs may increase hiring and attract higher wages. It depends on whether demand for the work grows when it gets cheaper to do - and historically, it usually does.

Anthropic's Peter McCrory agrees. His team introduced 'observed exposure' - an attempt to identify where theoretical exposure translates into actual displacement. One observation they made: when much of a job is automated, the remaining bottleneck tasks may increase demand for complementary human skills.

Meanwhile, the Fed chair told reporters that adjusted for overcounting, there's effectively zero net job creation in the US' private sector. Exposure maps say disruption everywhere. The macro data says stagnation. Both can be true - and that's the point.

So what's today's in-the-end-at-the-end?

An AI exposure score measures proximity to change. The mistake - and the one most workforce plans are built on - is assuming proximity to change means proximity to elimination.

It doesn't. There are more job like roofers out there than we realise.