Anthropic has introduced the Anthropic Economic Index, an initiative designed to monitor the effects of AI on labor markets and the broader economy. The Index's initial report provides data and analysis derived from millions of anonymized conversations on Claude.ai, offering insights into how AI is being integrated into real-world tasks.
A subsequent report from the Index covers usage data on Claude.ai following the launch of Claude 3.7 Sonnet, which includes an "extended thinking" mode. This second report analyzed data from 1 million anonymized Claude.ai Free and Pro conversations over 11 days after the model's release.
Key Points
- The initial report found that 37.2% of queries to Claude were in the "computer and mathematical" category, primarily for software engineering tasks. The second largest category was "arts, design, sports, entertainment, and media" at 10.3%.
- Approximately 4% of jobs used AI for at least 75% of their associated tasks, while roughly 36% of jobs showed some AI use for at least 25% of their tasks.
- AI use was lowest in both very low-paying and very high-paying jobs, with mid-to-high median salary occupations like computer programmers and copywriters showing the heaviest AI use.
- The initial analysis indicated a slight lean towards augmentation, with 57% of tasks involving AI collaboration and 43% involving direct automation.
- Following the launch of Claude 3.7 Sonnet, usage proportions increased modestly in categories such as coding, education, and the sciences.
- Claude 3.7 Sonnet's "extended thinking" mode was predominantly used in technical and creative problem-solving contexts, with tasks for computer and information research scientists leading at almost 10% usage.
- Anthropic has open-sourced the datasets used for these analyses, including a new bottom-up taxonomy of 630 granular usage categories.
Context
According to Anthropic, the research focuses on the ongoing impact of AI by analyzing direct usage data rather than surveys or forecasts. The methodology involves focusing on occupational tasks, which are classified using the U.S. Department of Labor's Occupational Information Network (ONET) database of approximately 20,000 specific work-related tasks. An automated analysis tool called Clio was used to organize conversations by occupational task while preserving user privacy.
Why It Matters
This initiative provides researchers and policymakers with data on how AI is currently being adopted in various occupations and tasks. The findings offer insights into the diffusion of AI across the economy, distinguishing between augmentation and automation, and highlighting which job types and salary ranges are experiencing the most AI integration. The release of datasets supports further independent research into these labor market transformations.
What To Do
- Review the initial report and the second research report from the Anthropic Economic Index for detailed findings.
- Examine the open-sourced datasets to understand the methodology and granular usage categories.
- Note the usage patterns of Claude 3.7 Sonnet's "extended thinking" mode in technical and creative problem-solving contexts.
- Consider how the distinction between AI augmentation and automation might inform future policy discussions or product development.