Assessing GPT's Potential to Replace Chinese Jobs: A Data‑Driven Study

OpenAI previously reported that about 80% of U.S. jobs could be affected by AI, with 10% of tasks potentially fully replaced. To explore the impact on China’s labor market, the authors analyzed billions of recruitment records from the past eight years.

Each occupation was broken down into specific functions and tasks, mapping Chinese jobs to the O*NET taxonomy, resulting in 19,265 distinct work tasks and 23,534 content items. These granular tasks were then evaluated for AI replaceability.

Instead of costly human labeling (≈¥10,000 for 40,000 items), the team used GPT‑4 (or GPT‑3.5‑turbo) as a scorer via a prompt that asked the model to rate how much a large language model could reduce human labor for each task on a 0‑5 scale. The cost of labeling 40,000 tasks dropped to under $5, with comparable quality to human annotators.

Aggregating task‑level scores to the occupational level produced a ranking of 50 occupations with the highest and lowest AI replacement probabilities. The most vulnerable jobs were translators, insurance underwriters, playwrights, visual designers, editors, and even computer programmers (≈75% of their work at risk). The least vulnerable were blue‑collar manufacturing roles and service jobs such as greening, cleaning, and massage therapy.

Unlike the U.S. findings where higher wages correlated with higher replaceability, the Chinese analysis showed no such relationship. Instead, a strong positive correlation emerged between a job’s “growth rate” (wage increase per year of experience) and its AI replacement probability (significance p < 0.001). Occupations with annual wage growth above 20% faced over 60% replacement risk.

The authors conclude that AI first displaces the experiential, skill‑intensive aspects of work—those acquired through prolonged practice—while innate human abilities remain harder for AI to replicate.