Blog · Review Essay · June 2026

Heteromation and the Labor Hidden Inside the Interface

Hamid R. Ekbia and Bonnie A. Nardi's Heteromation, and Other Stories of Computing and Capitalism gives a name to a labor relation that the AI era keeps trying to pass off as convenience. Instead of pushing humans out of the system, many digital systems pull humans deeper in, make their effort look like ordinary participation, and convert that effort into value for someone else.

The Book

Heteromation, and Other Stories of Computing and Capitalism was published by the MIT Press on May 4, 2017. The publisher lists it as a 280-page hardcover in the Acting with Technology series, with an ebook following on May 11, 2017. Ekbia was then a professor at Indiana University working across informatics, cognitive science, and international studies; Nardi was a professor of informatics at UC Irvine with long-running work on activity theory, human-computer interaction, gaming, sustainability, and technology-mediated practice.

The book expands an argument the authors had already developed in a 2014 First Monday article on "the invisible division of labor between humans and machines." Their claim is that computerization changed the division of labor in a way that the automation debate often misses. Automation imagines humans being displaced by machines. Heteromation names a different pattern: systems that depend on human participation while making that participation look like use, play, self-service, volunteering, communication, or ordinary life.

The MIT Press listing summarizes the range: communicative labor in social media and user-generated content; cognitive labor in microwork and self-service; creative labor in games and literary production; emotional labor hidden within paid jobs; and organizing labor in collaborative projects such as citizen science. The book's examples include Mechanical Turk, FoldIt, League of Legends, social media, reviews, self-service systems, design competitions, and other places where people perform economically useful activity through computer-mediated environments.

Automation Reversed

The strongest idea in the book is that automation and heteromation are not opposites so much as alternating moves in the same political economy. A company automates part of a process, then discovers that exceptions, classification, training, moderation, customer participation, content, repair, or affective care still require people. Instead of rebuilding a stable job around that work, it routes the work through software and makes the labor relation harder to see.

This is why the concept is still useful after the generative-AI boom. The public story says the machine writes, sees, recommends, classifies, summarizes, translates, tutors, evaluates, and decides. But the machine is surrounded by people: users creating training traces, annotators labeling data, reviewers rating outputs, workers correcting drafts, moderators absorbing harmful content, creators feeding platforms, customers entering their own information, and employees adapting messy institutions to rigid systems.

Heteromation makes the interface central. A self-checkout kiosk, social network, search box, game, app store, content platform, delivery workflow, or AI assistant does not merely receive input. It assigns work. It tells people what counts as contribution, how to format it, when to click, which choices are available, what will be remembered, and which forms of effort will vanish into a product metric.

That matters because the labor is not always coerced in the old factory sense. It can be enjoyable, expressive, necessary, career-building, socially meaningful, or hard to refuse. People write reviews because they want to help. They post because they want to belong. They tag photos, solve puzzles, rate answers, correct maps, train filters, build profiles, click through forms, and produce the cultural texture that makes a platform valuable. The system's power is that it can turn participation into surplus without naming the participant as a worker.

The User as Worker

The book belongs beside Ghost Work, but its emphasis is different. Ghost Work looks closely at workers who are visibly working to themselves but hidden from customers: microtaskers, moderators, transcribers, verifiers, and other people doing automation's last mile. Heteromation also cares about microwork, but it widens the frame to include the ordinary user whose labor is embedded in participation.

That wider frame is the book's contribution. The user is not only watched, profiled, advertised to, or personalized. The user is also productive. A profile improves targeting. A review improves a marketplace. A post improves engagement inventory. A correction improves a map. A search query trains ranking. A prompt and follow-up can become evaluation signal. A refusal, edit, thumbs-up, purchase, dwell time, or abandonment can become evidence for the next system.

This gives the book a sharper account of recursive reality than a simple exploitation story would. Platforms first organize action through interfaces. People act inside those interfaces. Their actions become data, reputation, labels, content, feedback, training material, and product improvement. The improved system then reorganizes the field of action again. The user is not outside the machine. The user is one of the machine's operating conditions.

That also means "free" services are not free in the relevant institutional sense. The payment may not be money, but value still moves. Time, attention, social knowledge, taste, emotion, culture, local expertise, identity, and language are converted into something that can be sold, optimized, ranked, modeled, or used to build a moat. The interface makes that conversion feel natural.

The AI-Age Reading

Read in 2026, Heteromation is a useful correction to the cleanest AI product story. A model demo tends to frame intelligence as an internal property of the system: the model can answer, code, reason, see, plan, and act. Ekbia and Nardi push the reader to ask what social arrangement makes those performances possible and who is being recruited into it.

Generative AI depends on heteromated surroundings. People create public text, images, code, documentation, forum answers, videos, captions, reviews, examples, prompts, corrections, votes, and preference signals. Some are paid. Many are not. Some consent meaningfully. Many do not. Some understand how their activity will be reused. Many only discover reuse after the fact. The model's apparent fluency rests on a world already formatted by platforms, search engines, documentation systems, schools, workplaces, fandoms, bureaucracies, and everyday users.

The concept also clarifies why AI assistants can intensify work while promising relief. A tool that drafts an email may require the worker to prompt, verify, edit, sanitize, document, and absorb responsibility for the output. A support bot may push more explanation work onto the customer. A hiring platform may make applicants optimize resumes for machine parsing. A medical portal may make patients act as intake clerks. A workplace agent may turn every employee into a process annotator whose corrections improve the workflow that will later supervise them.

This is not an argument that every user contribution should become a wage claim. The book is more useful than that. It asks institutions to stop pretending that computer-mediated participation is outside labor politics. If a system economically depends on human effort, then governance questions follow: compensation, consent, credit, refusal, data rights, portability, appeal, collective voice, safety, and shared control over the system built from that effort.

Where the Book Needs Care

Heteromation can feel too capacious. Once the term covers microwork, social media, gaming, self-service, reviews, emotional labor, and citizen science, readers have to keep asking what kind of labor relation is actually present in each case. A paid Mechanical Turk task, an unpaid fan contribution, a patient filling out a portal form, a Wikipedia edit, a cashierless checkout scan, and a prompt correction inside an enterprise AI tool do not have the same moral or legal structure.

Lilly Irani's 2021 review in Mind, Culture, and Activity is useful on this point. She reads the book as an analysis of industries using computing and networks to harvest value from social and cultural life, while also pressing the argument beyond its immediate digital-culture cases. The book is strongest when the term "heteromation" functions as an audit question, not as a universal answer: where has work gone, how is it being named, who captures value, and what obligations are avoided by calling the worker a user?

The book's utopian turn also deserves friction. Ekbia and Nardi want end users brought more fully into the prosperity created by computerized capitalism. That instinct is better than resignation, but the political path is hard. The firms best positioned to profit from heteromation often benefit precisely because the labor relation is ambiguous, diffuse, entertaining, legally weak, or spread across millions of small acts. Repair requires more than better interface etiquette. It requires enforceable rights and institutions capable of seeing distributed labor as labor.

Finally, the book predates the current foundation-model supply chain. It does not fully cover reinforcement learning from human feedback, model-evaluation vendors, synthetic-data pipelines, AI safety contractors, enterprise copilot telemetry, data-license fights, or the global politics of training data. That is a limitation, but not a defect. The book gives the underlying labor grammar; later books and reporting fill in the new machinery.

What This Changes

The durable lesson is that a machine-readable world is also a work-assigned world. Before a model can learn from action, the action has to be captured. Before a platform can personalize, people have to produce signals. Before an assistant can improve, someone has to correct it. Before an institution can automate, someone has to translate a messy practice into fields, prompts, labels, workflows, exceptions, and metrics.

That translation is where politics hides. A platform can call people users while treating them as trainers. A company can call a process self-service while transferring clerical work to the public. A model provider can call activity feedback while building product value from unpaid correction. A school, hospital, agency, or firm can call an interface convenient while making people responsible for making themselves legible to the machine.

Heteromation gives readers a practical diagnostic: when automation appears, look for the humans it has recruited rather than the humans it has supposedly replaced. Ask who enters the data, trains the categories, absorbs the errors, produces the content, supplies the affect, works around the brittle edge, and makes the smooth interface possible. Then ask whether those people have rights, compensation, credit, refusal, and a way to change the system that is learning from them.

That is the AI-era value of the book. It makes the user visible as part of the labor system without reducing all participation to victimhood. People do real things with digital systems: they play, teach, help, create, organize, care, learn, and speak. The danger is that institutions can build wealth and authority from those acts while leaving the people who performed them with only the language of convenience.

Sources

Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.

Buy on Amazon Browse Books

Return to Blog · Return to Books