How Aranya Sahay’s Humans in the Loop spotlights rural, invisible workers training AI

Humans in the Loop, which runs for 174 minutes, follows Nehma (played by Sonal Madhushankar), a woman from the Oraon tribal community in Jharkhand, who is in a ‘dhuku’ marriage (a form of live-in). She returns to her village with two children after separating from her husband and takes up work as a data-labeller at a small annotation centre. To sustain her family, she takes up remote work at a data-annotation hub set up by a tech contractor. Her job — tagging images and videos that train AI systems — becomes the film’s entry point into the hidden human backbone of machine learning. Written and directed by Sahay, the film features dialogues both in Hindi and Kurukh, a North Dravidian language spoken by the Kurukhs and the Oraons, which gives its world cultural and linguistic specificity.

Human choices fed into AI

A key focus of the film, which released in theatres on September 5 and is currently streaming on Netflix, is the “invisible labour” behind AI, the workforce whose repetitive, meticulous tasks rarely feature in global conversations about automation. Through Nehma’s routine, Sahay shows how lived experience is forced into rigid annotation templates designed far from the communities that must use them. This leads to a larger thematic inquiry: the tension between indigenous ecological knowledge and algorithmic taxonomies. An insect labelled a “pest” in a dataset might hold very different cultural or ecological value in Nehma’s region, which reveals how AI systems often homogenise or erase context.

The film, which owes its title to a system in which humans and AI collaborate, integrating human intelligence and expertise into the AI’s decision-making and development process, also interrogates how bias enters machine learning. Rather than treating “algorithmic bias” as a technical flaw, Sahay locates it in the human choices embedded in datasets: the assumptions, omissions, and cultural blind spots that chart how algorithms perceive the world. By tracing these decisions back to people and communities, the film turns a global discussion on AI fairness into an intimate story about representation, identity, and the politics of being seen.

Sahay builds the drama out of the friction between Nehma’s Adivasi world and the abstract demands of machine learning. Inspired by journalist Karishma Mehrotra’s long-form piece “Human Touch” about Adivasi women working with AI, the film is attentive to the texture of labour: fluorescent-lit rooms, lagging computers, supervisor chasing targets, and workers trying to make sense of labels written in a different idiom from their own lives. What emerges is a portrait of the “hidden workforce” behind AI: rural and often female data-labellers whose painstaking annotation powers systems that are marketed as automated intelligence.

Also read: Neeraj Ghaywan on Homebound: ‘We owe each other the humanity we all crave’

One of the film’s most striking narrative conceits is its use of the analogy that training an AI is like raising a child. At the data centre, this is the simplest way to explain the job: you show the machine enough examples, and it starts to “understand”; you correct its mistakes, and over time it learns. Sahay explicitly connects these two tracks in Nehma’s life: feeding patterns into the algorithm at work and trying (and often failing) to read the patterns in her daughter at home. As an explanation for audiences, the “AI as child” metaphor is an intentional simplification, but the film uses that simplicity to open up a bigger question: if AI is a child, who gets to be the parent, who writes the rulebook?

A social film about AI

From there, Humans in the Loop moves into the terrain of cultural erasure. Nehma and her colleagues are constantly nudged towards “clean”, “standardised” labels that match the client’s expectations, even when those categories clash with local knowledge. A specific, unnamed insect is tagged as a ‘pest’ by the AI system being trained, and Nehma gets in trouble for failing to label the creature as a pest, because based on her community’s ecological knowledge, she knows it is not harmful to crops. The film uses this specific scenario to highlight the biases of AI, which relies on universal categories, often ignoring valuable local and indigenous knowledge.

Also read: Tribeny Rai interview: ‘Sikkim lives in my mind as pieces of places, people, sounds’

The point is not that annotation is malicious, but that the grid itself is too coarse: when the world is forced into a handful of machine-readable boxes, nuance is the first casualty. Bias, in Sahay’s telling, is less a glitch than an accumulation of such decisions. The film shows how dataset guidelines, client briefs and workplace hierarchies encode dominant assumptions into the material Nehma is asked to produce. The film makes this politics legible without drowning the viewer in jargon, tracing how AI’s “neutrality” is undermined by who gets to name things, whose faces are over-represented and whose stories simply never enter the frame.