One of OpenAI’s founders, Andrej Karpathy, recently published and then removed a list ranking U.S. jobs by their vulnerability to artificial intelligence. The analysis, based on data from the Bureau of Labor Statistics, assigned an “AI exposure” score (0-10) to 143 million positions. The higher the score, the greater the risk of automation or AI integration.
The AI Exposure Ranking
The data revealed a counterintuitive trend: higher-paying jobs face greater AI disruption. Roles like software developers, data scientists, and financial analysts topped the list, while positions in construction, barbering, and nursing assistance showed minimal exposure.
This finding contrasts with earlier assumptions that low-skill jobs are most at risk. Karpathy’s approach used an AI model to assess how “digital” each job is, meaning how easily its tasks could be automated. This metric doesn’t guarantee displacement but indicates potential for change.
Why the List Was Removed
Karpathy took down the data after users misinterpreted it as a definitive prediction of job losses. He explained on X (formerly Twitter) that the “exposure” score was just one factor, ignoring real-world demand and economic factors. He restored the data later, clarifying that AI impact is complex.
AI’s Impact: What Other Research Says
Karpathy’s findings align with other reports. OpenAI’s 2023 study showed similar trends, while Anthropic’s recent report found high AI exposure in well-paid professions. However, despite the risk, mass unemployment due to AI hasn’t materialized yet. Anthropic’s research suggests hiring slowdowns for younger workers in high-exposure roles, but no systematic job losses.
“We find no systematic increase in unemployment for highly exposed workers since late 2022,” the Anthropic report noted, “though we find suggestive evidence that hiring of younger workers has slowed in exposed occupations.”
This suggests AI is changing how work is done, rather than eliminating jobs entirely. The focus is shifting from wholesale replacement to integrating AI tools into existing workflows.
The debate over AI’s impact on employment is ongoing. While some jobs are clearly more vulnerable, the actual effects depend on economic forces, worker adaptation, and the pace of technological adoption.
