Anthropic Economic Index: Summary

文章 5 min read · 未标注

Overview

The Anthropic Economic Index is a new initiative launched on February 10, 2025, to understand AI's effects on labor markets and the economy over time. Its initial report provides data and analysis based on millions of anonymized conversations from Claude.ai, offering a detailed view of how AI is being integrated into real-world tasks. The dataset is open-sourced for researchers.

Key Findings

Based on analysis of ~1 million Claude.ai conversations:

1. AI Usage Concentration

Software development and technical writing are the dominant use cases.
37.2% of queries fall under "computer and mathematical" (e.g., software engineering).
10.3% fall under "arts, design, sports, entertainment, and media" (mainly writing/editing).
Occupations with high physical labor (e.g., farming) have minimal representation (0.1%).

2. Depth of AI Use Within Occupations

36% of occupations see AI use in at least 25% of their associated tasks.
4% of occupations use AI across at least 75% of their tasks.
No evidence of entire jobs being automated; AI is diffused across specific tasks.

3. Automation vs. Augmentation

57% of tasks involve augmentation (AI collaborates with humans).
43% of tasks involve automation (AI directly performs tasks).
Augmentation subtypes: validation, learning, task iteration.
Automation subtypes: directive, feedback loop.

4. AI Use and Salary

AI use is highest for mid-to-high wage occupations (e.g., computer programmers, data scientists).
Both lowest-paid (e.g., shampooers) and highest-paid (e.g., obstetricians) roles show low AI use.
Likely reflects current AI capabilities and practical barriers.

Methodology

Data Source: Anonymized conversations from Claude.ai Free and Pro plans.
Tool: Clio (automated analysis tool) matched conversations to occupational tasks while preserving privacy.
Task Classification: Used the U.S. Department of Labor's O*NET database (~20,000 tasks).
Process: Conversations -> Tasks -> Occupations -> Categories (e.g., "computer and mathematical").

Limitations & Caveats

Cannot confirm if AI use was for work or personal tasks.
Unclear how users utilized AI responses (e.g., copy-pasting vs. editing).
Data excludes API, Team, and Enterprise users.
Potential misclassification of tasks by Clio.
Claude cannot generate images (except via code), limiting creative use cases.
Coding may be overrepresented due to Claude's advertised strengths.

Future Directions

Longitudinal tracking: Regular updates to monitor changes in AI's labor-market impact.
Key metrics to watch: Depth of AI use within occupations, automation vs. augmentation trends.
Open data: Dataset available for download; researchers invited to provide feedback.

Call for Input

Anthropic invites economists, policy experts, and researchers to:

Provide input on the Index.
Suggest new research directions via a feedback form.

Acknowledgements

Thanks to researchers including Jonathon Hazell, Anders Humlum, Molly Kinder, Anton Korinek, Michael Kremer, John List, Ethan Mollick, and others for comments and discussion.

Anthropic is hiring for Societal Impacts Research Scientist and Research Engineer roles.

Source: Introducing the Anthropic Economic Index (Feb 10, 2025)

来源

暂无来源