I do wonder how much of the population is using AI.
On page 7 of the paper there's the diagram "Minimum fraction of tasks in use". On the left side about 75% of occupations use at least one tasks and on the right side the maximum is some occupation that uses slightly more than 95% of the tasks.
Cool.
Here I start to wonder how they got that graph.
At the start of section 3. Methods and analysis on page 4 it's said:
> To understand how AI systems are being used for different economic tasks, we leverage Clio [Tamkin et al., 2024], an analysis tool that uses Claude [Anthropic, 2024] to provide aggregated insights from millions of human-model conversations. We use Clio to classify conversations across occupational tasks, skills, and interaction patterns, revealing breakdowns across these different categories. All analyses draw from conversation data collected during December 2024 and January 2025.
So this means they use real people's chats to make these estimations. I don't know Clio, but perhaps they did this? They sample chats from individuals, and some individuals never chatted and some individuals delegated all their work to Claude. But I wonder how they estimated the total numbers of tasks of an individual.
I am sure these answers are found by really going deep and reading the cited sources and running some experiments yourself, but I can't be bothered, sorry.
Again, I really wonder how much of total population use AI? How much? How do parts of population differ? Can this be found out at all?
On page 7 of the paper there's the diagram "Minimum fraction of tasks in use". On the left side about 75% of occupations use at least one tasks and on the right side the maximum is some occupation that uses slightly more than 95% of the tasks.
Cool.
Here I start to wonder how they got that graph.
At the start of section 3. Methods and analysis on page 4 it's said:
> To understand how AI systems are being used for different economic tasks, we leverage Clio [Tamkin et al., 2024], an analysis tool that uses Claude [Anthropic, 2024] to provide aggregated insights from millions of human-model conversations. We use Clio to classify conversations across occupational tasks, skills, and interaction patterns, revealing breakdowns across these different categories. All analyses draw from conversation data collected during December 2024 and January 2025.
So this means they use real people's chats to make these estimations. I don't know Clio, but perhaps they did this? They sample chats from individuals, and some individuals never chatted and some individuals delegated all their work to Claude. But I wonder how they estimated the total numbers of tasks of an individual.
I am sure these answers are found by really going deep and reading the cited sources and running some experiments yourself, but I can't be bothered, sorry.
Again, I really wonder how much of total population use AI? How much? How do parts of population differ? Can this be found out at all?