The question ‘Is AI coming for my job?’ is about as old as ChatGPT. Anthropic just tried to answer it with data instead of gut feelings.
A New Metric: ‘Observed Exposure’
Researchers Maxim Massenkoff and Peter McCrory developed a new metric called ‘observed exposure.’ Instead of just theoretically assessing which jobs AI could do, their approach combines technical feasibility with actual real-world usage data. And it weighs automated uses more heavily than purely augmentative ones.
Sounds like a subtle difference, but it matters. Because there are plenty of jobs that AI could theoretically replace — but that are barely affected in practice.
The Findings
The headline numbers: computer programmers have the highest task coverage at 75%. Customer service reps, data entry keyers, and medical record specialists also rank among the most exposed occupations.
But here’s the crucial part — there’s no systematic increase in unemployment for highly exposed workers since late 2022. At the macro level, not much has changed.
What is showing up, though: hiring of younger workers in exposed occupations appears to be slowing down. That’s subtler than mass layoffs, but potentially more significant in the long run.
Who’s Affected?
Here’s an interesting detail: the most affected workers tend to be older, female, more educated, and higher-paid. That runs counter to the narrative that AI primarily threatens low-skill jobs.
My Take
I find this study remarkable for two reasons. First, because Anthropic is creating transparency about the societal impact of their own technology. That’s unusual for a tech company.
Second, because reality is more complicated than the headlines. Fortune went with ‘Great Recession for white-collar workers.’ The data tells a more nuanced story: the shift is happening, but slowly and not as dramatically as feared. Not yet.
Sources: