Data‑Driven Assessment of GPT’s Potential to Replace Chinese Occupations

OpenAI researchers warned that about 80% of U.S. jobs could be affected by AI, prompting a similar investigation for China. By mining billions of Chinese recruitment records from the past eight years, the authors evaluated which occupations are most likely to be replaced by large language models such as GPT.

The study first decomposes each job into its constituent functions and tasks. For example, a human‑resources officer’s duties are broken down into recruitment, onboarding, payroll, and channel management, and each sub‑task is further split into concrete actions like preparing training materials or scheduling sessions.

Using the O*NET taxonomy, the Chinese occupations were mapped to 19,265 tasks and 23,534 work contents. To label these items, the team employed GPT‑4 (and GPT‑3.5‑turbo) as an automated scorer, prompting the model to assign a 0‑5 rating that reflects the proportion of human labor that could be reduced.

Labeling 40,000 items with GPT‑4 cost roughly $48 (or $3 with GPT‑3.5‑turbo), compared with at least ¥10,000 and a week of human labor for the same work, while delivering comparable quality.

Aggregating the task‑level scores yields occupation‑level replaceability estimates. The top‑ranked vulnerable jobs include translators, insurance underwriters, and playwrights, each with over 90% of tasks deemed replaceable. Visual‑design roles, editors, writers, call‑center agents, and even software developers also show high susceptibility (70‑80%).

Conversely, low‑replaceability occupations are mainly blue‑collar manufacturing roles, as well as green‑keepers, cleaners, laundry workers, masseurs, nail artists, and traditional Chinese pastry chefs.

Analyzing the data reveals that in the U.S. study, higher wages correlated with higher replaceability, but the Chinese results show no such wage‑replaceability link. Instead, a strong positive correlation exists between a job’s “growth rate” (annual wage increase tied to experience) and its AI replaceability; occupations with >20% yearly wage growth have a >60% chance of being automated.

The authors argue that AI is especially adept at supplanting skills acquired through prolonged practice and experience, rather than innate human abilities. High‑growth jobs, which rely heavily on learned expertise, are therefore the most at risk.

Overall, the research demonstrates that large language models can efficiently evaluate labor‑market vulnerability, offering a cost‑effective alternative to manual annotation and highlighting which professions in China may face the greatest AI‑driven disruption.