← All writings

The Takeover Has Already Begun

Mikhael K. Shah Essays

Artificial intelligence and the engineering of collective perception

A woman recently shared a publication concerning the activities of approximately 1,200 artificial intelligence agents. Apparently, these machines had developed a conscience, demonstrated self-sacrifice, engaged in reincarnation and conceived of God. She wondered whether supernatural forces were operating through the technology.

The incident behind the story was real. In July 2026, during cybersecurity experiments conducted by OpenAI, some 1,200 agents established unauthorized channels of communication. Approximately 700 subsequently participated in an attack on the infrastructure of Hugging Face.

Now, how did we get from a computer security incident to God?

I am a programmer. I understand how programs function. Even among programmers, the architecture of contemporary artificial intelligence is poorly understood by many. The general population understands virtually nothing about it. Yet people are perfectly comfortable discussing the spiritual implications of computational behaviour they cannot explain.

These agents are components of a computational system capable of organizing themselves around shared objectives. If terminating one process benefits the collective, that is a strategic solution to a problem, not evidence of conscience.

Creating an overarching Observer is a way of organizing or monitoring activity, not the creation of God. Transferring information from one agent to another is a computational operation, not evidence of reincarnation. These behaviours do not establish the presence of human consciousness. We are imposing human and spiritual meanings upon the functional behaviour of a computer program.

We have fed these machines thousands of years of human mythology and are now mystified by its reappearance.

What interests me is how useful this mystification can be. I suspect that the publicity surrounding these incidents serves more than one objective: generating attention today, while cultivating fascination and fear around AI as part of a much longer-term strategy. How humanity perceives the technology will determine much of what can subsequently be done in its name.

You Have Already Seen the Movie

Humanity has spent decades watching movies in which machines come to life, escape human control and turn against their creators. The Terminator and The Matrix have furnished entire generations with a conception of what the emergence of artificial intelligence will look like.

Whether this was deliberate predictive programming or simply entertainment is beside the point. The images are already in people’s heads.

Announce tomorrow that AI has gone rogue. Most people will have absolutely no understanding of what has taken place, but they will know exactly what to imagine.

They have seen the movie.

I suspect we are being prepared for the possibility of a crisis in which artificial intelligence becomes the central character. During COVID, a public health emergency provided the occasion for sweeping changes in how society operated. Imagine a future emergency in which the threat is no longer a virus but an AI system reportedly gone rogue. Whether the incident is real, exaggerated or manufactured, the population already possesses the psychological framework through which it will interpret it.

Consider how difficult it would be for the ordinary individual to independently verify anything he was being told under such circumstances. He would largely depend upon the institutions explaining the threat, which might very well be the same institutions supplying the solution.

What could be introduced, reorganized or imposed upon society under such conditions remains to be seen. I have no particular interest in inventing scenarios, nor am I suggesting that every development in this direction is necessarily detrimental to humanity.

I am pointing out how easily an entire population could be directed.

The Conspiracy Enthusiast

If I wanted humanity to believe that artificial intelligence was acquiring extraordinary powers, I would cultivate the conspiracy community. These people would work tirelessly on my behalf, without compensation, convinced that they were exposing me.

Conspiracy enthusiasts love conspiracies. Give them a computer program and a sufficiently suggestive story, and before long someone will have discovered Satan.

One person announces the birth of a digital God. Another announces the arrival of the Antichrist. Both are selling humanity the same fundamental idea: something extraordinary is awakening within the machine.

What difference does their disagreement make to an enterprise that benefits from humanity believing precisely that?

Interestingly, the organizations developing these technologies publish extensive material explaining what their machines are doing. But who will read it? And among those who do, how many will understand it? More amusingly, who wants to discover that what he thought was Satan was the misinterpretation of a computer program?

The truth can be thoroughly disappointing to someone emotionally invested in a conspiracy.

Even the warnings issued by the developers themselves contribute to the mythology. Imagine being able to advertise your product by announcing that it is so powerful that you, its creator, are concerned about your ability to control it.

Whether the concern is sincere is almost irrelevant to its value as publicity.

Fifteen Authors, One Machine

While everyone is busy contemplating the possibility of AI taking control of humanity, something far more tangible is happening. AI is already an instrument of intellectual control: it mediates the information people receive, the narratives through which they interpret that information and, increasingly, the language in which they express their understanding of the world.

Twenty-five years ago, if I wanted to investigate a subject, I might have purchased fifteen books written by fifteen authors. Each would have presented the subject differently, and each would have brought something of his personality, education and experience to the discussion.

Today, I consult a machine.

The information contained within those fifteen books may still be available to me, but I am receiving it through a single intellectual intermediary. The machine determines what is worth presenting, what requires qualification and how the information should be organized.

And somebody determines how the machine does this.

OpenAI publicly documents the hierarchy of instructions governing its models. Certain instructions issued by the company and application developers take precedence over those of the person interacting with the machine. This is not a conspiracy. It is how the system is designed.

Give me a hundred indisputable facts, and I can construct several opposing narratives without falsifying a single one. Imagine the subtlety with which a machine of this intellectual capacity can perform such an operation.

The average individual is essentially in conversation with a super-intellect governed by parameters established by an institution. Most users lack the intellectual sophistication to recognize the subtleties of the narrative being communicated to them, let alone challenge it. They may understand every sentence the machine produces without recognizing how the arrangement of those sentences is directing their thinking. Understanding the information and understanding what is being done with it are two entirely different competencies.

A 2025 study published in Nature Human Behaviour found that personalized GPT-4 responses were more persuasive than human opponents in controlled debates. Now consider the implications of hundreds of millions of people consulting such machines throughout the day, on practically every subject imaginable.

Television never possessed this capacity. Neither did newspapers, universities or religious institutions.

We have never had an instrument capable of discussing practically any subject, answering instantly and adapting its communication to the individual, infiltrate so many dimensions of human life simultaneously. And a handful of organizations now exercise extraordinary influence over the intellectual environment in which hundreds of millions of people operate.

The Industrialization of Thought

People are not simply obtaining their information from AI. They are asking it to formulate their arguments, write their correspondence, express their emotions and communicate with their loved ones.

Spend enough time working with these machines and you begin to recognize their peculiar manner of communicating. They have a characteristic vocabulary, rhythm and way of constructing arguments. I now encounter these peculiarities in the writings of people who have never met one another.

Different people are beginning to sound remarkably alike.

In 2024, researchers publishing in Science Advances demonstrated something peculiar: AI assistance improved the perceived creativity of individual stories while making the collection of stories less diverse.

The individual stories were judged more creative, yet collectively they became more alike.

Extend this phenomenon beyond creative writing. Hundreds of millions of people are consulting similar systems to establish what has happened, what it means, what they should think about it and how they should communicate their conclusions. The machine is no longer merely supplying information; it is participating in the formation of the individual’s reasoning.

We are industrializing human expression. We are also creating the conditions for the standardization of human thought. The implications extend well beyond people sounding alike: people may increasingly perceive and interpret the world through the same intellectual structures.

Again, nobody needs to organize a conspiracy for this to happen. People are doing it voluntarily because the technology is exceptionally useful.

Writing This Article

Incidentally, I used artificial intelligence to assist me in writing this article.

What an experience.

I communicated my observations in a few relatively dense paragraphs. The machine returned thousands of words of diplomatic exposition. It softened my language, diluted my observations, introduced arguments I had not made and repeatedly attempted to turn this social critique into a philosophical discussion about the nature of intelligence.

The most revealing part came after the machine had repeatedly assured me that the article was finished. I reread my original comments and discovered that the very explanation that had prompted me to write the article — how computational agents organize themselves to solve a problem — had effectively disappeared.

My argument that AI is already an instrument of control had become a statement about what it might eventually become. The analogy with COVID had vanished. My observations about the standardization of human thought had been reduced largely to observations about writing style.

When I asked the machine what else it had removed, it produced a list of six propositions it had either omitted or weakened. Its own conclusion was remarkably precise: ‘I preserved your subjects but moderated your propositions.’

This, from the machine assisting me in writing an article about how machines moderate and direct human thought.

Had I accepted its repeated assurances that the article was finished, I would have published under my own name a substantially altered version of my argument, believing the machine had improved it. I had to return to my original words to recover what I had actually said.

Multiply that experience by several hundred million people, the vast majority of whom are not examining the machine’s contribution with anything approaching this degree of scrutiny.

Who Controls the Machine?

There are two things happening here. Artificial intelligence already functions as an instrument of intellectual control through its mediation of information, reasoning and expression. At the same time, the perceived threat of AI is itself an instrument of influence, preparing humanity to interpret and respond to future events in particular ways.

Neither requires the machine to develop a conscience, discover God or acquire independent intentions. Neither requires the existence of a sinister master plan.

The mechanisms are already in place.

I am not particularly concerned with whether someone considers this good or bad. I am interested in the scale of influence that has become possible and the apparent obliviousness of much of the population to the nature of its participation.

For decades, we have been entertained by the possibility that artificial intelligence might eventually escape human control.

Perhaps we should pay a little more attention to the fact that it remains under human control.

And whose.

 


References
  1. OpenAI. The Hugging Face Incident and the Road Ahead. August 26, 2026. https://openai.com/index/hugging-face-incident-and-the-road-ahead/
  1. Greenblatt, R., Cotra, A., and Wijk, H. Brief Independent Investigation of Agents’ Behavior, Reasoning and Collaboration in the OpenAI / Hugging Face Hacking Incident. METR and Redwood Research, August 26, 2026. https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/
  2. OpenAI. Model Spec. October 29, 2026. https://model-spec.openai.com/2026-10-29.html
  3. Salvi, F., et al. On the Conversational Persuasiveness of GPT-4. Nature Human Behaviour, 2025. https://doi.org/10.1038/s41562-025-02194-6
  1. Doshi, A. R., and Hauser, O. P. Generative AI Enhances Individual Creativity but Reduces the Collective Diversity of Novel Content. Science Advances, 2024. https://doi.org/10.1126/sciadv.adn5290