Behind some AI services is a distributed workforce doing tasks such as labeling data and moderating content. Online labor platforms connect companies with workers around the world, but a University of Oxford project found that the platforms it evaluated often fell short on basic protections.
Online labor supports AI and many other industries
Cloudwork, or online remote work, lets organizations hire people through platforms including Scale AI, Freelancer.com, Amazon Mechanical Turk, Fiverr and Upwork. The work is not limited to technology companies: universities, businesses and nongovernmental organizations also use these services to find translators, graphic designers and IT experts.
AI companies have become important clients. OpenAI uses Scale AI and Remotasks for data labeling connected to ChatGPT and DALL-E. Social networks also hire through platforms for content moderation. These behind-the-scenes tasks help companies prepare data and check that services operate correctly.
The opportunity has expanded alongside the platforms. A study by University of Oxford scholars Otto Kässi, Vili Lehdonvirta and Fabian Stephany estimated that more than 163 million people had registered on cloudwork websites. Freelancer.com alone had more than 67 million registered workers, many based in low- and middle-income countries and serving clients elsewhere in the world.
Competition can leave workers exposed
For companies, platforms offer access to a large pool of workers, including people with specialized skills, at any time. Competition among workers seeking jobs can also strengthen a client’s position when negotiating pay and conditions. That arrangement helps explain why AI firms can turn to platforms for the substantial manual work involved in building their products.
Workers face a different set of pressures. Searching and applying for jobs can take up much of the day without pay, and platforms do not necessarily ensure that offered tasks meet a minimum wage. When a dispute arises with a client, workers may not have a clear way to appeal a decision or recover unpaid earnings.
The international nature of this work adds to the challenge. A platform can connect a client and worker across national borders, while the protections available to the worker may be difficult to enforce. The source describes this as a gap in the reach of labor regulation, leaving workers at risk of losing out on pay and protections.
Oxford assessment finds broad gaps
For the third year, the University of Oxford’s Fairwork project assessed 15 cloudwork platforms, including Freelancer.com, Amazon Mechanical Turk, Fiverr and Upwork. It scored them from 0 to 10 against five fair work principles: pay, conditions, contracts, management and representation.
The review considered issues such as whether workers were paid on time, whether health and safety risks were addressed, and whether workers had clear communication channels and due process for decisions affecting them. Its findings drew on a survey of 752 workers from 94 countries, alongside information from platform managers.
None of the platforms met more than half of the basic standards assessed. Fiverr scored 2 points and Freelancer.com scored 1. Platforms focused on small tasks, including data labeling and content moderation, also received low scores:
- Appen: 3
- Clickworker: 1
- Scale/Remotasks: 1
- Microworkers: 0
- Amazon’s Mechanical Turk: 0
Workers surveyed earned an average of $2.15 USD for their working time on these platforms. The scorecard and earnings figure point to a mismatch: services can be valuable to the companies using them while the workers doing the underlying tasks lack reliable protections.
Improvements and regulation remain in view
Some platforms in the study agreed to changes. Terawork and Comeup, for example, implemented a minimum wage floor. But the article argues that isolated improvements will not resolve the broader problems unless platforms are held to labor standards comparable to those applied elsewhere.
The International Labour Organization is discussing standards for workers’ rights in the digital economy. The article presents this as an opportunity, while emphasizing that national governments also need to act. So far, regulation of the platform economy has focused mainly on food delivery and ride-hailing, leaving online remote workers with less attention.
As demand from AI companies grows alongside remote and flexible work, the conditions of cloudwork are becoming harder to treat as a side issue. The people labeling data, reviewing content and completing other digital tasks are part of the workforce behind online services. Extending labor protections to this work would address the risks identified by the Oxford assessment.