AI and jobs: how researchers study exposure, automation, augmentation
Researchers use task-level analysis, expert input, and field studies to measure how AI exposes, automates, and augments jobs. Their findings range from broad exposure estimates to modest productivity gains.
Measuring occupational exposure to AI
Researchers assess AI's impact on jobs by analyzing the tasks that make up occupations. For example, the arXiv study GPTs are GPTs developed a rubric to score tasks based on alignment with large language model capabilities, integrating both human expertise and AI classifications. The study reports that around 80% of the U.S. workforce could see at least 10% of their tasks affected.
The International Labour Organization's refined global index combines task-level data, expert input, and AI model predictions to measure occupational exposure across countries. It places workers into four exposure gradients and finds that one in four workers globally are in an occupation with some exposure. The index also notes that clerical occupations have the highest exposure levels, and that exposure varies by income level and gender.
Task-based frameworks for automation
Acemoglu and Restrepo propose a task-based framework to study automation and AI in their paper Artificial Intelligence, Automation and Work. They describe a displacement effect where machines and AI replace labor in some tasks, reducing labor demand, and a productivity effect from cost savings that increases demand in non-automated tasks. They also emphasize the creation of new labor-intensive tasks as a countervailing force.
In a later analysis, Acemoglu's Simple Macroeconomics of AI uses a task-based model to estimate that AI's contribution to total factor productivity is modest, with an upper bound of 0.66% over ten years, and possibly lower when accounting for hard-to-learn tasks. The analysis highlights that AI's macro effects depend on the fraction of tasks impacted and average task-level cost savings.
Evidence on augmentation from workplace studies
The NBER study Generative AI at Work examined a generative AI conversational assistant used by customer support agents. Access to the tool increased productivity by 14% on average, with a 34% improvement for novice and low-skilled workers but minimal impact for experienced workers.
The study also observed that customer sentiment improved, retention rose, and there were signs of worker learning. This suggests that generative AI can augment human performance rather than simply replace it, with effects that vary across skill levels.
Policy research and guidance
The OECD's AI-WIPS programme analyzes how AI affects training needs and labor markets, providing resources for governments to design responsible and human-centred AI policies. This work supports social dialogue and targeted policy responses to manage the transition.
The MIT Work of the Future Initiative conducts multidisciplinary research on how technologies including generative AI change work. It convenes a working group on generative AI and job quality, and runs an Automation Clinic to understand how organizations deploy new technologies in practice, with a focus on improving opportunities for workers.
| Approach | Example | What it measures |
|---|---|---|
| Task-level rubric | Eloundou et al. (arXiv) | Percentage of tasks exposed to LLM capabilities |
| Global index with expert and AI scoring | ILO | Occupational exposure across countries and sectors |
| Field experiment | Brynjolfsson et al. (NBER) | Productivity changes from AI assistance |
| Task-based theoretical model | Acemoglu and Restrepo (NBER) | Displacement and productivity effects of automation |
Frequently asked questions
What is occupational exposure to AI?
Researchers define exposure as the proportion of tasks in an occupation that could be affected by AI capabilities, based on task-level analysis. The ILO measures exposure using a global index that places workers into gradients of potential impact.
How do researchers distinguish automation from augmentation?
Automation refers to AI replacing human tasks, while augmentation refers to AI assisting workers to perform tasks more effectively. The NBER study on customer support agents found that AI assistance improved productivity, illustrating augmentation.
What do the studies conclude about the scale of AI's effect on jobs?
Estimates vary. The arXiv study suggests around 80% of the U.S. workforce could see at least 10% of tasks affected, while the ILO finds that one in four workers globally have some exposure. However, macroeconomic analysis by Acemoglu predicts modest productivity gains.
What policy guidance emerges from this research?
Researchers emphasize the importance of social dialogue and targeted policies to manage the transition. The OECD provides resources for governments to design responsible AI policies, and MIT's Work of the Future Initiative studies how AI can be implemented to improve job quality.