Waiting for the Robots
- Victor Hugo Germano
- Jul 19
- 7 min read
Blog post originally in Brazilian Portuguese
"Automation changes the nature of work, but it doesn't destroy it."
There is a scene, recounted by Antonio Casilli right at the beginning of Waiting for Robots: The Hired Hands of Automation , that summarizes the entire argument of the book before it even begins: Thomas Jefferson used "dumbwaiters," small elevators that served dishes during his dinners, and there seemed to be no human intervention. Except that the elevators were operated by enslaved employees hidden in the basement .
It is this same "inconspicuous workforce" that Casilli, a sociologist at the Institut Polytechnique in Paris, identifies as behind every contemporary artificial intelligence system. There is no autonomous robot, no AI that learns on its own. What exists, however, is an army of hired human hands that train, verify, correct, and imitate systems that are then sold as technological magic.
From Indian programmers and remote drivers to Mechanical Turks and Data Labelers, the market is full of automation dreamers who are actually workers hidden from the eyes of consumers, poorly paid and facing general instability.
Casilli opens the book with the obvious, which we continue to collectively ignore: no AI exists in a vacuum. "Human contribution is an essential component of all AI tools, regardless of their sophistication."
This is the same concern that accompanied me when reading Phil Jones in *The Work Without the Worker* : behind every "automatic" facial recognition or self-driving car, there is an army of data annotators defining, down to the last cent per task, what is or is not a face, a traffic sign, a hate speech. As I wrote at the time, "the magic of machine learning is the human rush of data classification," and Casilli has been rigorously documenting this same magic with academic rigor, showing that the "P" in GPT stands precisely for "pretrained": trained by humans, before anything else.
Deepening our understanding of the changing perspectives on work allows us to see where all jobs are eventually headed, preparing us for the new reality.
"Society's focus should no longer be on the expert computer scientist, but on the hundreds of millions of hired hands that, every day, pull the strings of these puppet-automatons."
Casilli describes a phenomenon he calls Displacement: human labor in the digital economy is not eliminated, it is merely moved out of sight, repositioned at another point in the production chain, where it continues to create, maintain, and activate the tools that use this extracted human intelligence. Before any interaction with Siri or Alexa, workers had to listen to hours of recordings of banal conversations; before any sonnet generated by a ChatGPT, humans had to evaluate tons of similar texts .
While Phil Jones focuses on micro-work as a symptom of a planetary employment crisis, Antonio Casilli builds a broader sociological theory on how platforms capture value, not only from workers paid by task, but also from ordinary users who, unknowingly, produce training data simply by using an app. This book is not about technology, it's about work, its current impact, and the transformations our society has undergone in changing the meaning of how work is recognized, exploited, and remunerated: this is also a discussion about what work is today .
The numbers behind this are concrete. In Kenya, more than 140 content moderators hired by Sama to review Facebook material were diagnosed with severe post-traumatic stress disorder after years of watching videos of decapitations, child abuse, and suicides, receiving about $1.50 per hour . The lawyer representing the workers summarized the problem: "If in Dublin people can't look at harmful content for more than two hours, that should be the rule everywhere." Meta paid $52 million in 2020 to affected American moderators; the Kenyan workers, to this day, have received nothing equivalent.
Moderated by Facebook/Sama - Nairobi, Kenya
Payment ~US$ 1.50-2.20/hour;
100% of those evaluated had PTSD, anxiety, or depression.
TikTok/Majorel Moderation - Nairobi, Kenya
Target of 250-350 videos/hour; without regular psychiatric support.
ImageNet annotation (Mechanical Turk). Global, international arbitration.
Payment per task, without a formal contract, without labor rights.
The big problem with "Apocalypse of Work" predictions lies in their reductionist, task-based view, where as one group of tasks is automated, entire occupations are eliminated. These predictions largely overlook the fact that employment is determined by macroeconomic trends, as well as geopolitical instability, environmental crises, and industry repositioning.
In the consulting world, this way of thinking is very common, and the result for businesses is always bad: Jeff Sutherland explains this thinking as "The Doorman Fallacy." If you understand the doorman's role simply as a way to open the door, then of course it's possible to generate cost savings by replacing it with an automated system. But if you understand the doorman as a way to elevate the hotel's status and create a better experience for the guest (receiving packages, greeting guests, carrying luggage, security), then you realize that the doorman's complete role cannot be replaced by an automated system.
Short-term savings come at the cost of the brand as an employer that offers unparalleled employee benefits and prioritizes employee satisfaction over investor satisfaction, and the long-term damage to the ability to attract new talent and retain current employees can cost the company much more in the years to come.
Casilli is clear that the real objective is not to replace workers, but to use the threat of replacement as a tool for wage discipline: automation is the whip to discipline the workforce and the carrot to attract investors. No one organizes for better working conditions for fear of unemployment .
"While fear focuses on robots taking all jobs, reason refuses to accept that AI is, in fact, made up of outsourced digital labor."
The clicking finger and platformization.
The concentration of work on large platforms is an increasingly common trend in the workplace: international arbitrage, low or no organizational capacity, algorithmic decision-making in services: what you watch, who drives you to work, what you choose to eat—your daily work is increasingly being mediated by platforms. iFood, Uber, Netflix, Instagram, OnlyFans, YouTube, Sama, Amazon, Google Maps—these are the new textile factories.
The real value isn't in the food delivered, but in everything surrounding the transaction: the order time, the delivery route, the waiting time, the messages exchanged. And all this data is used to train algorithms. Casilli describes how, through platforms, work has been emptied in the name of community and attention: they gamify the creation of restaurant reviews, so they don't have to pay content producers, while the goal is to extract more and more data from users.
For Casilli, platforms are "walled gardens" that impose data production tasks under the guise of convenience, and their "supposed neutrality" "absolves them of social responsibility and masks them as neutral social commodities." It's the same argument I developed while reading Yuk Hui on technological sovereignty : platforms are not mere intermediaries, they are power structures disguised as free markets, which capture data as a strategic asset and return it, at best, as "engagement."
This surveillance architecture became almost comically visible in the case of the "smart" cameras in European supermarkets, as reported by Casilli: employees in Madagascar, sitting in a garage , watch camera streams and manually send theft alerts to the cashiers, while the public believes that AI automatically identifies the theft. It's the Wizard of Oz of platform capitalism: pull back the curtain, and there's always a poorly paid human doing the work that the brand sells as "artificial intelligence."
Reading Waiting for Robots makes it hard not to think back to Lewis Mumford. In The Myth of the Machine , Mumford describes entire civilizations organized as a "megamachine": pharaohs as central processors, scribes as programmers, and a mass of workers reduced to interchangeable cogs, "input/output units whose value is determined by their contribution to the system's objectives, not by their human qualities." Today's AI, I wrote in that post, is that same megamachine "built of data, algorithms, and the relentless pressure of efficiency," operating "not by the whip, but by operational convenience, omnipresent surveillance, and the promise of constant optimization."
Casilli arrives at a very similar structural conclusion, but through the lens of the sociology of work: platformization does not aim to eliminate work, but rather to empty it of its social and political value . "Despite beginning by saying that automation cannot eliminate work, platformization can completely devalue it."
It's the same Mumford machine, just updated: instead of bricks and muscles, data and clicks; instead of a pharaoh, a dynamic pricing algorithm.
Finishing Waiting for Robots doesn't leave you with a "problem solved" feeling, and Casilli is honest about that, avoiding both technological catastrophism and the naive optimism of AI evangelists. His proposed way out is something like "Data Collectivism," or digital socialism, where data is a collectivized common good, a strategic asset that could generate a basic income conceived outside the libertarian discourse of Silicon Valley. If population data is a value-generating asset, it should contribute to a Universal Digital Income through the taxation of companies that extract value from citizens' data.
"If platforms had to pay for digital work, including invisible work, the business model would completely collapse. Value capture would become unfeasible."
I consider this an essential book, in the same category as Ghost Work by Mary Gray and Siddharth Suri and The Work Without the Worker by Phil Jones: works that force us to look behind the curtain of the "miracle" of AI and recognize that, if any functional artificial intelligence exists in the world today, it has a name, a social security number (or its local equivalent), and often a salary that doesn't even cover the monthly bills.
The question remains: how much of the discourse on artificial intelligence is actually an illusion to hide labor relations that we prefer not to examine closely?
Casilli answers this question with numbers, lawsuits, and testimonies. The answer is not reassuring.
Comments