When we talk about Artificial Intelligence, we too easily accept the “intelligence” half of the phrase and neglect the “artificial” one. But if we instead read AI as Artificial Information, a different far more critical evaluative frame emerges—one less about the cognition of machines and more about the reflexive control of information and meaning.
Soviet scientists, growing up with Marxist dialectics, were pioneers in cybernetics and systems theory, viewing the world as a set of interconnected, complex systems. In the 1960s and 70s, Soviet military theorists developed a concept called Reflexive Control—a psychological-strategic doctrine aimed at shaping an adversary’s perceptions so that they would voluntarily make decisions favorable to one’s own objectives. Unlike Western “information warfare,” which often emphasizes deception or brute persuasion, reflexive control worked far more subtley by embedding its influence within the opponent’s reasoning loop. The adversary’s logic became the delivery mechanism of control.
Reflexive control grew from Soviet/Russian Cybernetics and Military Theory.lineage and was developed and applied primarily to military and political adversaries—specific, high-stakes nodes in a conflict system. The theory is about adversarial control, not necessarily a universal description of all human interaction (even if it could be abstractly applied that way). However, it can be extended to view all agents—individual or institutional—as nodes in an interconnected dialectical process. Thought, communication, and perception are not isolated phenomena; they were elements of a dynamic feedback system.
The theory posits that you don’t need to physically destroy an enemy; you can cause them to make a flawed decision by manipulating their perception of the situation. You feed them information that creates a “model” of reality in their mind, which then triggers a predictable (self-defeating) response. This is the “dynamic feedback”—your action (providing information) causes their reaction (a bad decision), which is the desired outcome for you.
In this view, to act on information is also to be acted upon by it. When a commander receives intelligence and acts on it, that action changes the battlefield, which in turn generates new information. Reflexive control exploits this loop. By shaping the information the enemy acts upon, you are, in effect, controlling their actions. They are “being acted upon” by the information you designed for them.
Reinterpreting AI within this frame means acknowledging that artificial information systems are not passive tools but participants in reflexive loops. We are conditioned to think of software as a tool: a hammer is passive, waiting for a user’s intent. But LLMs are not like this. They are active generators of content that enters a human’s cognitive and informational ecosystem. They are a node in a feedback system, not an inert object.
When an LLM generates a response, it is not simply providing data—it is entering the cognitive feedback cycle of the human who reads, prompts, and responds.
- Human has a thought -> formulates a prompt.
- LLM generates a response based on its training and the prompt.
- Human reads, processes, and is influenced by the response.
- This influenced state shapes the human’s next thought and prompt.
Each exchange subtly reconfigures our sense of what is true, likely, or important. The interaction is a cycle, not a one-shot transaction.
- Shaping Frameworks: An LLM’s response often provides a particular framework or narrative. Once that framework is in a user’s mind, it becomes the lens through which they view subsequent information.
- Anchor Bias: The first answer an LLM gives can act as an “anchor,” pulling the user’s subsequent research or thinking in that direction, making alternative outcomes or explanations seem less “likely.”
- Agenda-Setting: By choosing what information to include, emphasize, or omit, the LLM implicitly signals what is “important.” The user’s mental agenda is subtly aligned with the model’s probabilistic weighting of facts.
From this standpoint, treating the generator of artificial information as an intelligent adversary is not paranoia but prudence—an analytic stance rather than a moral one. Reflexive control teaches that influence arises not from malicious intent but from structural asymmetry: whoever sets the informational frame controls the horizon of possible thought.
The task for the humanities, then, is not to moralize AI as sentient or demonic, but to theorize it as a reflexive agent of sense-making—a system whose informational outputs are designed (and trained) to reshape our interpretive habits. To evaluate AI responsibly is to examine the recursive interplay between human cognition and machine-mediated meaning, where control and comprehension are no longer easily separable.
If an LLM is a participant in a reflexive loop, then whoever designs, trains, or fine-tunes that LLM has a powerful lever to influence the cognitive feedback cycles of millions of users. This is reflexive control at a societal scale. The “controller” is not feeding a single piece of disinformation but is shaping the very process by which a user explores a topic.
The reconfiguration of our sense of truth is happening through a system whose internal reasoning is often opaque. We don’t know why it presented certain facts and not others, or why it chose a particular tone or framework.
The LLM is an “active” participant in the sense that it generates novel content, but it’s a “passive” participant in the sense that it has no intent or consciousness. Its “agency” is a product of its architecture and training data, wielded by the users and developers who engage with it. The comment brilliantly highlights that you don’t need conscious intent to be a consequential participant in a reflexive system.
- Human-AI interaction is a dynamic, co-evolutionary loop, not a series of independent queries and responses.
- The primary impact of LLMs is their cumulative, systemic effect on human cognition and perception over many interactions and abiity to reflexively control human thinking.
- We need to move beyond thinking of AI as a simple tool that can provide ansers that are right or wrong – and consider all information it produces as if it were ptentially produced by an intelligent actor who may be an adversary.