As AI adoption grows, many people wonder how it will affect the labor market in the long term (ChatGpt Deep Research has its own predictions). Humanity is trying to find it.
On Monday, the company released its first economic index. This investigated the types of employees using humanity’s Claude chatbots and what tasks they use. It takes a different approach from many AI and work research that wants future mapping, and humanity focuses on similarity of work tasks rather than job titles, and actual chatbot queries instead of survey responses. It was there.
Also: Crawl and walk before running with AI agents, experts recommend
“In many cases, jobs share a common skill with specific tasks. For example, visual pattern recognition is a task performed by designers, photographers, security screeners and radiologists,” the company said. explained in the report’s announcement.
Using its own Clio system for privacy, humanity analyzed a million anonymized conversations. Anthropic announced that each mapped to the Occupational Information Network (O*NET), a US Labor Bureau database of 20,000 work tasks, with Clio identifying tasks that “bestly represent the role of AI.” Chats were grouped into occupations such as arts and media, computers and mathematics, business and finance.
The software engineering task constituted the majority of the queries in the dataset. 37.2% of conversations were related to debug code, network troubleshooting, and more. The queries in the next largest category involved writing and editing in 10.3%. This was grouped as jobs such as “art, design, sports, entertainment, media” copywriting.
Also: Pitted Claude 3.5 sonnet against AI coding test ChatGpt Aced – and it failed creatively
In this study, we observed that these two occupations constitute only 3.4% and 1.4% of the US economy, respectively. For example, they use AI at a much higher rate, although much less than office managers or sales jobs. Work in science and education also showed higher AI utilization compared to saturation in the economy.
Beyond programming, top tasks within these four categories include producing, conducting research, and creating educational materials for entertainment such as film and television.
Automating and automating work
We also found that AI enhances human capabilities for 57% time and automates tasks – directly perform tasks for people – 43% time.
Also: Deepseek’s Cerebras CEO: Every time computing gets cheaper, the market gets bigger
“Light more than half of the cases, AI was not used to replace those who performed the tasks, but instead engaged in tasks such as validation (double-checking the user’s work) and learning ( For example, the report clarifies the knowledge and skills that help users to acquire new things, and iterations of tasks (for example, help users to do brainstorming or repetitive generation tasks).
However, humanity admits that users cannot confirm that they are referring to Claude for work purposes in these conversations. This is particularly relevant given that the survey does not review data from APIs, teams, or enterprise users.
Humanity also cannot confirm whether users have used Claude’s written response or code snippet as is or edited outside the application. This creates the difference between augmentation and automation.
Other survey results
Using median pay data from O*Net, this study found that AI use is more common in “moderate to high wage occupations” tasks such as data science. People in lowest and highest paying band jobs, such as salon workers and doctors, were much less likely to use Claude, as their roles emphasized manual work.
The US also supports “AI control” and puts AI safety aside
“This likely reflects both the limitations of current AI capabilities and practical barriers to using technology,” humanity added. Furthermore, this study found:
Approximately 4% of employment used AI, while at least 36% of tasks used AI in at least 25% of tasks
Take home
Humanity’s plan will regularly rerun its analysis to see if a particular role is experiencing more automation. “We will be able to monitor changes in the depth of AI use within our occupation,” the announcement said. “If AI is used only for certain tasks, and only a few jobs using AI for most tasks, the future is something that most current jobs evolve rather than disappearing. It’s possible.”
The report itself does not create policy recommendations. “The development of policy responses to address future changes in the labor market and the impacts on employment and productivity will take a variety of perspectives,” Humanity said. “To that end, we are also inviting economists, policy experts and other researchers to provide their opinions on the index.”
For those interested in the data itself (or in search of a grain of salt), humanity has opened sourced conversations for additional research efforts.
