Research Alert: AI Agents' Testing is Too Focused on Programming, Overlooking 92% of the Real Labor Market
A joint study by Carnegie Mellon University and Stanford University found that current AI agent development suffers from 'path dependency.' The research analyzed 72,000 tasks in mainstream AI benchmarks and found that evaluations are overly concentrated in the programming field, while neglecting non-programming occupations that make up 92% of the U.S. labor market. This imbalance may lead AI technology development to deviate from practical application needs.