Debates at One World: Debating In the Belly of AI

This March, we participated as expert discussants at three screenings of the documentary In the Belly of AI (V útrobách AI), held as part of the One World (Jeden svět) international human rights documentary film festival, one of the largest such festivals in the world, organized annually by People in Need (Člověk v tísni). This year’s edition ran from March 11 to April 24, 2026.

The film offers a critical look at the infrastructure underpinning today’s large language models. Through testimonies from experts and people directly affected by AI systems, it examines the less visible costs of the technology: low-wage data labeling work in the Global South, psychological damage suffered by content moderators, and the extractive mining of minerals that power our devices. These are framed against the ideology of longtermism and the priorities of the Silicon Valley elite.

While we appreciate the film raising these questions for a broad audience, we found its tone and some of its factual claims to be somewhat one-sided. The film mentions there are 145 to 435 million data workers, which sounds staggering, but on closer inspection, it encompasses all online gig work, including web developers, graphic designers, administrative assistants, and many others. Using such a broad category to illustrate the specific plight of AI data labelers inflates the picture considerably and muddies an otherwise legitimate concern.

A more subtle issue the film raises—perhaps without fully confronting it—is the widening divide between the Global South and the North. People in lower-income countries are currently doing significant amounts of data labeling and similar low-skill AI support work. But this work is inherently self-undermining: as AI systems improve, partly thanks to that very labor, the need for human labelers will diminish. The same dynamic is already visible in other routine service jobs (like call centers) that are being automated away. These are young people spending formative years on repetitive, unskilled tasks. What happens to them once the systems they helped build no longer need them, and when they have had little opportunity to develop more transferable skills in the meantime? The film gestures at exploitation but doesn’t fully reckon with this longer-term structural trap.

We participated in post-screening debates twice with student audiences and once with a general public audience. The student discussions were particularly engaged. The question that perhaps best captured the mood was a simple but important one: What can we actually do about this? Our answer was twofold. At the individual level, the effect is limited to being mindful of the tools we choose to use and how we use them. The more effective lever is political: electing representatives willing to put real pressure on technology companies, especially at the EU level, where regulatory frameworks with similar scope can take shape. And beyond the systemic, we also noted that some of the most direct harms, including cases where AI chatbots have allegedly contributed to users harming themselves or others, call for a more personal response: staying attentive to the people around us and not leaving people close to us navigate these technologies entirely alone.