At a Warner Bros studio near Hollywood in the fall of 2024, Tesla staged a glitzy showcase of its AI progress. Acting as master of ceremonies, Elon Musk introduced a crew of the company’s new Optimus robots. “The Optimus will walk amongst you … You’ll be able to walk right up to them, and they will serve drinks,” he said. As he spoke, the humanoids bore out his words, filling glasses for attendees. Yet they weren’t quite as autonomous as they seemed. Days after the event, reports revealed that humans had teleoperated the robots.
The movie lot was an apt location for the show. As AGI edges toward center stage, actors masquerading as machines are misleading the public about the technology’s advances.
Michael Geoffrey Asia, a former chat moderator and data annotator from Kenya, has provided this “invisible labor”. One job required him to talk intimately with lonely users. Assuming fabricated identities, he built facades of human-level emotional intelligence. He soon suspected he was both impersonating chatbots and training future AI models. The experience led to the formation of the Data Labelers Association, an advocacy group where he serves as general secretary.
“It’s without doubt that these interactions have created the illusion that AI already possesses human-level cognition, emotional depth, and relational intelligence,” Geoffrey Asia tells AGI Ethics News. “But in reality, it is workers like us performing highly skilled emotional labor behind the scenes.”
In perceptions of AGI, the impact has a unique intensity. Because there is no agreed technical benchmark for “general intelligence,” perceptions of it are shaped as much by belief and narrative as by measurable technical advances. Narrow AI, by contrast, can be evaluated through quantifiable performance on specific tasks. The path to AGI, however, is often judged on subjective signals: how it feels, how it talks, and how human it seems.
By offloading emotional intelligence to hidden workers, companies can manufacture these signals. This not only misleads the public but also impedes genuine progress, burying authentic advances beneath artificial polish. Incentives to address real AGI challenges shrink, investors favor safer bets, and technical research is redirected.
For the enablers, the invisible labor becomes a strategic tool. By orchestrating acts of machine intelligence, they can inflate trust in AGI, conceal its limitations, and amplify hype about an imminent digital revolution — with themselves at the vanguard.
“Platforms intentionally blur the line between human and machine by using fabricated profiles, scripted emotional responses, and interfaces that hide the human workers,” Geoffrey Asia says.
This confusion inflates public belief that AI is already capable of deep interpersonal engagement, when much of that ‘intelligence’ is actually harvested from human emotional labor — our own creativity, empathy, conversational improvisation, and psychological insight.
“Some users even test me to see if I’m an AI, not realizing that the company has positioned me as the human pretending to be the machine pretending to be human. This confusion inflates public belief that AI is already capable of deep interpersonal engagement, when much of that ‘intelligence’ is actually harvested from human emotional labor — our own creativity, empathy, conversational improvisation, and psychological insight.”
Dr Julie Carpenter, an external research fellow at the Ethics + Emerging Sciences Group, explains the power of this masquerade through a framework she’s developed: “the human gaze.” The concept describes our instinct to project intention, agency, and care onto non-human entities based on social and interactional cues.
As social creatures, we’re prone to perceive emotional intelligence in anything that mirrors our conversational rhythms, empathy, or humor. In AGI development, this can be powerfully manipulated. By triggering the human gaze through invisible labor, companies can exploit a social instinct to create false faith in machine intelligence.
“This matters because it recalibrates expectations,” Dr Carpenter says. “If emotional responsiveness appears native to the system, people are trained to treat it as an emergent machine capability rather than a human-mediated one.
“Over time, this collapses the distinction between machine inference and human judgment, making claims about AGI feel less like a leap and more like a confirmation of everyday experience. The result is not just hype, but a systematic misrepresentation of how intelligence, care, and responsibility are actually distributed.”
If this hidden human labor were revealed, people could lose faith in authentic AGI. Dr Carpenter, however, envisions public perceptions fracturing rather than simply deflating.
“For some, it would clarify that what appears as empathy or understanding is often a mediated performance stitched together through human judgment, affective calibration, and social intuition,” she says. “For others, the illusion may persist, because the experience of being responded to smoothly or compassionately often matters more than the provenance of that response.”
David Gunkel, a media studies professor at Northern Illinois University and the author of Robot Rights, positions this phenomenon within a lengthy history of anthropomorphization. Our tendency to attribute agency to the “other,” he argues, is a fundamental part of how we navigate the world. “Anthropomorphism isn’t a bug; it’s a feature,” Gunkel says. “As social creatures, we need to make sense of the other.”
By stimulating this reflex with hidden human workers, companies can sell false dreams of AGI. Critics have called for mandatory disclosures of this invisible labor, but Gunkel is skeptical. He argues that such warnings often primarily shield the powerful rather than the public, much as cigarette warnings can safeguard tobacco companies better than the lungs of their customers. “Disclaimers are more about protecting the corporation from liability than protecting consumers,” Gunkel says.
In his view, the true danger of invisible labor lies in such asymmetries of power. When the main governing instrument is the terms of service, users face a “contract of adhesion.” They can either accept the AGI narrative offered by the provider or be excluded from the technology entirely.
Gunkel acknowledges that the illusion can cause real harm. “We do have to be concerned about the liabilities of these systems and the harms that could be caused — not only on the users, but on the people behind the scenes.” People like Michael Geoffrey Asia, for whom the experience was traumatic.
“These hidden interactions have created a cultural narrative where AI appears more advanced than it truly is and most users have normalized the idea that machines can love, comfort, or emotionally understand us. And because the exploited human labor behind these systems remains invisible, the public is left with a distorted impression: that AI already thinks, feels, or relates like a human, when much of what they experience is the product of workers whose identities have been erased.”
AGI Ethics questions addressed in this piece:
- Are there contexts in which it is unethical to disclose, or not disclose, that one is interacting with an AGI rather than a human?
- What guidelines address the anthropomorphization of AGI and its effects on perceptions of agency and moral standing?
- Can users easily know when they are interacting with AGI versus a human, and have real alternatives or opt-outs?
Subscribe to AGI Ethics News
Get our latest essays on the philosophy of AI and techno-ethics delivered free to your inbox.
Thomas is a freelance tech reporter who has covered AI for nearly a decade. He was previously managing editor of The Next Web, a Financial Times company. In his free time, he likes playing chess (badly) and basketball (even worse). You can reach him at thomas.s.r.macaulay@gmail.com