- The study introduces a new index called “observed exposure” to measure the risk of labor replacement by AI, combining the theoretical capabilities of LLMs with real-world usage data, prioritizing full automation and job application cases.
- The analysis is based on three data sources: the O*NET occupational database covering approximately 800 US occupations, AI usage data from the Anthropic Economic Index, and task automation assessments from 2023 research.
- The β scale evaluates AI’s ability to accelerate work: β=1 if LLMs can double the speed, β=0.5 if additional tools are needed, and β=0 if automation is impossible.
- Results show that AI is currently utilizing only a small fraction of its theoretical potential. For example, the Computer & Math occupational group has 94% of tasks supportable by LLMs, but real-world usage is only around 33%.
- Occupations with the highest exposure include computer programmers (75% of tasks automatable), customer service representatives, and data entry clerks (around 67%).
- Approximately 30% of workers are virtually unaffected by AI because their tasks do not appear in usage data, including chefs, motorcycle mechanics, lifeguards, bartenders, and dishwashers.
- Comparison with US Bureau of Labor Statistics forecasts shows that occupations with high AI exposure tend to have lower job growth projections through 2034.
- For every 10 percentage point increase in exposure, the projected job growth rate decreases by approximately 0.6 percentage points.
- Workers in highly affected occupational groups are often older, have a female ratio 16 percentage points higher, earn 47% more on average, and possess higher education levels.
- Individuals with postgraduate degrees account for 17.4% of the high-exposure group, nearly four times the low-exposure group (4.5%).
- Analysis of US unemployment data from 2016 to the present shows no evidence of AI increasing unemployment since ChatGPT’s emergence in late 2022.
- However, hiring data shows early signs: the recruitment rate of young workers (ages 22–25) into high AI-exposure occupations has decreased by about 14% compared to 2022. 📌 Conclusion: Research indicates that AI is still far from its theoretical capabilities and has not caused large-scale unemployment. However, professions such as programming, customer service, and financial analysis face high exposure. The affected labor group typically has 47% higher income and higher education. Data from 2016–2025 shows unemployment hasn’t risen, but recruitment of young workers into AI-replaceable roles has dropped by 14%, signaling early AI impact.
VIET NAM CONSULTING AND MEASUREMENT JOINT STOCK COMPANY
© 2026 Vietmetric