'They Are Purposely Hidden': Sociologist Says Millions of Data Workers Power AI in Secret

Millions of workers around the world are labelling data, moderating disturbing content and training AI systems but their labour is often deliberately hidden, according to sociologist Antonio Casilli.

Casilli, a professor of sociology at Télécom Paris, part of the Institut Polytechnique de Paris, is the author of Waiting for Robots: The Hired Hands of Automation. He made the claims in an interview with Democracy Now on Thursday and is also a co-writer of the documentary In the Belly of AI, directed by Henri Poulain.

Why Companies Hide the Workers

Casilli said data workers are essential to how AI chatbots and models function, despite the technology often being presented as increasingly automated. 'They are purposely hidden by the companies that produce AI because they are the secret ingredient of AI itself,' he told the programme.

He said these workers train AI systems, verify outputs and sometimes carry out live maintenance that can effectively substitute for the machine. Casilli said his research team has spent six years surveying, interviewing and observing data workers in more than 30 countries.

His team interviewed more than 4,000 workers in places including Venezuela, Kenya and Madagascar.

Estimated Scale of the Workforce and Pay

Casilli's own estimates put the number of data workers at around 160 million in 2021, a figure he says has since risen. He said he remains sceptical of broader industry estimates.

In a separate interview with the University of Groningen, Casilli cited World Bank estimates suggesting that between 100 million and 400 million people were engaged in online gig work. Those figures cover online gig work more broadly and should not be treated as a specific count of AI data workers.

The World Bank's 2023 report said the online gig economy could account for up to 12 per cent of the global labour market. Casilli said pay varies significantly depending on the country and platform.

In Venezuela, he said, workers were sometimes paid in dollars or cryptocurrency rather than the rapidly depreciating bolívar, making the work appealing despite rates that would be unviable in higher-income economies.

In Madagascar, Casilli said his team encountered at least one company where workers were compensated in food rather than money. 'They were paid in sugar, rice, beans and other foodstuff,' he said.

The company used a meter at its entrance to show workers how much food they would earn for different levels of performance.

The Psychological Toll

Former Kenyan data worker Faustine Makira described experiencing nightmares, anxiety, depression and symptoms of post-traumatic stress disorder after reviewing graphic content, including material involving murder and sexual abuse, as part of her work.

'I developed anxiety. I could not go out,' Makira said in the documentary, adding that she continues to attend therapy.

She said colleagues had experienced similar symptoms, including insomnia. Makira said she could not name the companies she had worked for because of a non-disclosure agreement.

'Immediately I say them and they go out there, I will be jailed for more than 10 years,' she said.

Why Workers Are Kept Silent

Casilli said non-disclosure agreements can serve several purposes for technology companies, including protecting unreleased AI models and intellectual property from competitors. He also argued that restrictions on communication can fragment the workforce and make it harder for workers to organise collectively.

'If workers speak among themselves... they do not reach the awareness that they are doing the same kind of work,' he said.

The claims highlight the human labour behind AI systems that millions of people use every day.

Many users may be unaware of the workers involved in training models, evaluating their outputs and reviewing or filtering material used by AI systems.

Casilli's account suggests that the industry's presentation of AI as increasingly autonomous can obscure a global workforce that he says often faces low pay, psychological pressures and contractual restrictions on discussing its work.

His claims also raise broader questions about labour standards in AI supply chains as governments and companies face increasing scrutiny over the industry's social and ethical impact.