An Anthropic research team has investigated to what extent current robots can perform existing jobs using AI. The research team created a database of approximately 900 occupations and about 19,000 tasks, and used the AI model Claude to evaluate whether each task can be executed by a robot.
Survey finds robots can perform 74% of manual labor, but only 0.3% is cheaper than humans
This article is a translation. Read the Japanese original
The study revealed that of the manual labor accounting for 46% of all working hours, 74% can be performed by robots under certain conditions. Based on the complexity of the task execution environment, robots are classified as follows:
- E1: Tasks that can be performed in environments built specifically for robots, such as factories (23%)
- E2: Tasks that can be performed in human work facilities with fixed structures, such as logistics warehouses (10%)
- E3: Tasks that can be performed in complex and unstructured environments, such as urban roads (1%)
On the other hand, tasks requiring high dexterity or interpersonal skills, such as dyeing hair or providing first aid, were classified as unable to be performed by current robots (E0).
Separate from technical feasibility, economic cost remains a major barrier to adoption. Anthropic estimated the cost required for robots to achieve productivity equivalent to humans. The results showed that while there are cases where robots are cheaper than humans, such as packing tasks, the vast majority of jobs—such as taxi driving or washing dishes—are tasks where execution is possible, but hiring humans remains more economical.
Tasks where robots outperform humans in terms of cost efficiency account for only 0.3% of the total. While robot prices have been falling at a rate of approximately 3% per year since the 1990s, even if this trend continues, it is calculated that it will take about 40 years for the proportion of tasks where robots are cheaper than humans to reach 10%.
Sources
- ロボットは人間の肉体労働の74％をすでに実行可能だが「人間より安く働ける仕事」に限るとわずか0.3％という調査結果 （GIGAZINE、2026-10-01）
- Anthropic