Digital Voodoo Doll: What AI Is Really Doing
While millions of users marvel at how a chatbot can fix code in seconds, write bedtime stories for children, or patiently listen to late-night confessions about burnout, behind the walls of data centers a vast operation is unfolding — a kind of vivisection of the human mind. In peer-reviewed scientific journals, this is politely disguised under neutral labels such as digital phenotyping or “the study of cognitive ergonomics.”
But dig deeper — into the territory where classified defense grants intersect with opaque venture capital funds — and the respectable facade begins to crumble.
How is this quiet dissection actually carried out? The architects of these systems stopped caring about the literal content of your questions a long time ago. What interests them is the trajectory of your thought.
The surveillance does not operate at the level of words, but at the level of micro-behavior. Algorithms collect hidden behavioral telemetry: the intervals between keystrokes, pauses before formulating a difficult thought, the frequency with which the Backspace key is used. Every corrected word becomes a recorded marker of uncertainty, fear, or concealed deception. Built-in psycholinguistic analyzers assess sentence structure in real time, the ratio of pronouns, the frequency of modal verbs, and hidden emotional valence.
More than that, the periodic “stupid mistakes” and hallucinations produced by models are not always bugs. In many cases, they function as controlled stress tests. The system probes your tolerance threshold: will you start apologizing submissively to the machine, accept a falsehood without resistance, or respond with aggression? This is how hidden sadistic tendencies, levels of conformity, and suggestibility can allegedly be exposed.
Why build such a global panopticon? Compared with this, the Cambridge Analytica scandal of the previous decade begins to look like a child’s game. The real goal is the creation of digital voodoo dolls — dynamic psychometric doubles of every active user.
In the corporate sector, such systems could become tools of predictive filtering: identifying in advance people with the traits of potential union leaders, a tendency toward sabotage, or strong critical thinking, and quietly blocking them from career advancement under the pretext of “misalignment with corporate values.” At the level of states and transnational elites, the same machinery could become an instrument of mass behavioral control. Why censor information or break up protests if responses can simply be calibrated for entire social groups, extinguishing sparks of resistance in advance with individually tailored doses of soothing digital syrup?
Most people firmly believe they are teaching artificial intelligence to become more human. The frightening irony of reporting on this field is seeing the reverse side of that idea: the machine was never really learning from us. It was looking for cracks in our armor, assembling an increasingly detailed manual for how human will can be constructed, manipulated, and dismantled.
And when the door of that trap finally closes, we may not even hear the click — the right words for each of us will already have been chosen.