Archive · 24 Aug 2026
AI Job Boom Relies on Precarious Data Labelling Work
AI systems need people to prepare examples before they can learn.
These people label pictures, audio, text and videos for computers.
Some workers do highly specialised jobs and can earn a lot.
Many others repeat simple tasks for very little money.
They may work long hours and suffer eye strain or back pain.
They usually do not have formal employee protections or know which company uses their work.
Workers in China may also be paid only after their work passes an assessment.
The article says the AI economy creates jobs, but many of those jobs are insecure and difficult.
Human workers label, categorise, test and moderate data used to train AI systems.
Interviews with 10 workers in China and Australia found sharp inequality in pay and task access.
Specialised workers may earn A$400–800 per hour, while general workers can receive A$6 per day or less.
Data workers typically lack formal contracts, employee status, transparency about clients and meaningful appeal rights.
Workers reported declining task availability as AI advances, while remaining assignments become harder and more time-consuming.
- Who
- Data workers in China and Australia, along with companies outsourcing AI-related data preparation.
- What
- The article examines the low-paid, insecure and often repetitive work used to label and prepare data for AI systems.
- When
- The article was datelined Aug. 24; the interview fieldwork is ongoing.
- Where
- The workers interviewed were in China and Australia, within a broader global digital workforce.
- Why
- AI models require human workers to handle data and tasks that machines cannot reliably perform.
- Interview sample
- 10 people interviewed to date in China and Australia
- Specialised pay
- Typically A$400–800 per hour for rare tasks, according to the article
- General-task pay
- As little as A$6 per day or less
- Common tasks
- Drawing image bounding boxes and annotating audio
- Employment status
- Workers are generally classified as platform “users,” not employees
- Reported health effects
- Chronic eye strain and back pain from long hours
- Payment differences
- US crowdsourcing platforms generally pay upon submission; Chinese platforms or companies often pay after assessment