Exclusive: New AI jobs risk paper posits less doom and gloom

Workers whose jobs are most vulnerable to automation — data-entry keyers, bookkeepers and more — are already using AI for three times as many of their relevant tasks as workers in less-exposed jobs, according to a new study by OpenAI.

Why it matters: The research, first seen by Axios, shows that those workers are using AI for only a fraction of what it could theoretically do.

• The researchers posit a less doom-and-gloom outcome: Workers might not automatically be on the frontlines of a jobs bust, even as AI use expands.
• Paradoxically, it could ultimately expand demand for certain types of work.

By the numbers: OpenAI sorts the 900+ occupations that cover nearly all of U.S. employment into four buckets.

1. 18% face the highest near-term automation risk, relative to other groups (think data-entry, bookkeeping, customer service)
2. 24% of roles could see employment shrink, even as those jobs are still human-led (HR specialists)
3. 12% of jobs could see employment expand because of AI (software developers, for one)
4. 46% face the least threat of immediate change (teachers, home-health aides)

The intrigue: Signs of disruption aren't evident in unemployment data yet.

• Workers in the highest-automation-risk jobs have seen a smaller rise in unemployment than have workers in the "less immediate change" category, OpenAI finds.
• The paper cautions: "These categories are not job loss forecasts. They are a map for understanding where near-term labor market pressure may emerge first."

Zoom in: Workers in the most vulnerable categories are using AI more than those in any other bucket for the tasks most central to their work. Yet they've barely closed the gap between current usage and what AI could hypothetically do in their jobs.

• AI could theoretically handle 90% of tasks in the highest-risk occupations. But those workers are currently using it for less than a quarter of that, according to OpenAI usage data.

Yes, but: Whether AI ultimately destroys or creates jobs hinges on a critical tension — when AI makes a task easier to perform, people may simply consume more of it.

• "When coding tools first came out, people assumed maybe we would always write a fixed amount of code," OpenAI chief economist Ronnie Chatterji tells Axios.
• "Now I'm writing code, you're writing code — you produce more of something, and more people might demand it and pay for it."