Core Conclusion: 43% is the automation ratio of work tasks by AI, rather than the overall replacement of jobs, sourced from the Anthropic report in February 2025 (Anthropic Economic Index), based on Claude's anonymous dialogue data, contrasting with '57% tasks enhanced.'
1. Core Data and Common Confusions
- Concept Clarification: It is task-level automation, not 'disappearance of jobs'; only about **4% of occupations use AI for ≥75% of tasks, about 36%** of occupations use AI for ≥25% of tasks.
- Collaboration is Key: AI is more inclined to enhance rather than replace, with **57%** of tasks being supportive (validation, learning, iteration), and **43%** being direct execution of tasks.
- Industry distribution: Software development and writing account for nearly 50% of total AI usage; computer/mathematics (37.2%), arts/media (10.3%) lead; agriculture/fisheries/forestry only 0.1%.
- Salary correlation: High and medium-income positions have the highest AI usage rates, while low-salary and extremely high-salary positions have lower usage rates.
2. The essence of transformation and response
- Essence: It is not a 'wave of unemployment', but rather task restructuring + skill migration, giving rise to new positions in AI training/prompt engineering, AI ethics compliance, etc.
- Response: Enhance non-automatable capabilities (creativity, cross-domain collaboration, complex decision-making, humanistic care); master AI collaboration tools to become efficient executors of 'human-machine collaboration'.
3. References from other institutions
- Goldman Sachs: About 80% of the U.S. workforce has at least 10% of tasks affected by LLMs, with 19% of employees having ≥50% of tasks impacted.
- McKinsey: From 2030 to 2060, about 50% of occupations will be affected by AI, with generative AI accelerating this by a decade.