Somewhere, at some point, a person opened your Instagram profile. Not to follow you. Not because they knew you. They were working. They needed money. They looked at your photos, maybe your friends’ photos, and fed what they found into a machine that will never know your name.
This is the detail that should stop you. Not the scale of it. Not the legal questions. The human detail: a contractor, possibly a journalist, a graduate student, a librarian, sitting at a desk in a random state, pulling your life into a training dataset while being watched by software that screened their own computer in case they were slacking off.
The Guardian reported this week on Scale AI, a company 49% controlled by Meta, that runs a gig platform called Outlier. The pitch is almost seductive: “Become the expert that AI learns from.” Tens of thousands of workers, many with credentials in medicine, physics, economics, were recruited to help refine AI systems. What they actually did included scraping Instagram accounts, harvesting copyrighted work, transcribing pornographic audio, labeling photos of dead animals and dog feces. One doctoral student was shown a diagram of infant genitalia. Several workers said they had been promised no nudity in their assignments. The promise dissolved fast. Ambient Intimacy and the People You Almost Know.
The workers described it as desperation dressed up as flexibility. “A lot of us were really desperate,” one told the Guardian. “Many people really needed this job, myself included.”
What sits in that statement is the shape of the shame. Because there are two separate shame loops running in this story, and they are almost mirror images of each other.
The first belongs to the workers. An artist described “internalized shame and guilt” for contributing “directly to the automation of my hopes and dreams.” That sentence is doing a lot of heavy lifting. It holds the specific humiliation of someone who trained for a skilled profession, who built an identity around expertise or craft, now being paid in pieces to process content they were told they wouldn’t have to see, in service of a system that will likely displace them. The gig economy has always traded on this dynamic: you can survive by participating in the thing that is slowly making your previous life unviable. But this version is more intimate. You are not just working for the machine. You are teaching it. You are handing it the vocabulary it will eventually use without you.
The second shame loop belongs to the people whose data was harvested, except they don’t feel it yet because most of them don’t know. The contractor quoted in the Guardian said that users “would be surprised” at how their data was collected. Pictures of themselves. Pictures of their friends. The material you posted for an audience you understood, absorbed into a training set you never consented to.
This is where the cyberpsychology of it gets uncomfortable in a specific way. When you share something on Instagram, you are not simply broadcasting information. You are performing a version of yourself for a perceived context. This is what the psychologist Erving Goffman called impression management, the work of curating who you appear to be in a given social frame. The audience matters. It shapes what you share, how you caption it, what you leave out. You are posting for your friends, your family, a loose digital community you have some sense of.
The harvesting process collapses that context entirely. The image you posted because it captured something real about a day, or because you looked the way you wanted to look, or because it documented a moment with someone you love, that image travels somewhere else entirely. It becomes training data, stripped of context, read by a stranger under deadline pressure, classified and filed. The version of you that existed in that photo, the you that was performing for a specific audience in a specific moment, is now feeding a model that will approximate human experience for millions of interactions it will have with people you will never meet. The Version of You That Lives in Someone Else’s Screenshots.
This is a specific kind of identity violation that we do not yet have clean language for. It’s not a hack. Nothing was stolen in the conventional sense. The photos were technically public. But something was taken: the contextual frame around your self-presentation, the understanding that when you shared this image you were sharing it with people, not with a machine’s education.
The taskers, the workers, were themselves inside a similar collapse. The Outlier platform promised them expert identity, a flattering mirror: you are the kind of person whose knowledge matters, whose expertise shapes what AI understands about the world. That identity dissolved quickly into tasks that bore no relationship to their credentials. The promise of being the expert AI learns from became, in practice, transcribing audio they had been told they wouldn’t encounter, and labeling content no one with expertise in physics or economics needed to label.
There is something recursive about this. The workers were sold an identity that the platform then quietly discarded. The users’ self-presentations were harvested and stripped of the identity context that made them meaningful. Everyone in this system lost something about how they understood themselves, and the machine gained.
Glenn Danas, a lawyer representing AI gig workers in lawsuits against Scale AI and similar platforms, estimates hundreds of thousands of people worldwide now work in this labor category. That is a large number of people carrying the specific psychological weight of the artist who spoke to the Guardian, the feeling of being complicit in something that compromises something you value, under conditions where refusal is economically difficult.
What the gig economy has historically done to physical labor, this system is now doing to expertise, to craft, to the particular human value of trained judgment. And it does it while producing the training data for systems whose entire pitch is that they can replicate trained judgment at scale.
The workers monitoring their own shame. The users unknowing. The machine learning. The company growing.
When we talk about what AI costs, we tend to mean infrastructure, compute, electricity. We are less practiced at accounting for the human interior cost: the person who needed the work badly enough to transcribe something they did not want to see, who looked at a stranger’s photos in a random state, who felt the particular vertigo of contributing to their own displacement while doing it.
That is the labor that built this. Not abstracted. Not automated. A person at a desk, monitored by software, looking at your life.
Digital Alma explores technology, consciousness, and what it means to be human in a digital world.
Leave a Reply