From Propaganda to Recommendation Engines: How Fiction Imagines Manufactured Desire

Old propaganda imagined a loudspeaker. Modern recommendation systems suggest a whisper: something chosen for you, timed for you, and delivered in a form you are likely to enjoy.

Science fiction about social media becomes most interesting when it connects those two models. The question is no longer only who controls the message. It is who controls the selection system that decides which message each person sees.

MAYA: the recommendation system becomes a civilization

The MAYA Narrative Universe takes this structure and moves it into biology. On Neh, the living Maya network grows through trees. People tether for stories, games, social connection, identity, memory, and everyday participation. The system is not a separate app sitting on top of society. It is part of the social fabric.

That makes MAYA useful for thinking about science fiction about manipulation. The Divyas have access to vast information and predictive simulations. They can model possible futures and make interventions in the present. Sometimes the intervention is a policy change. Sometimes it is a story. Sometimes it is an alteration to the environment in which a decision will be made.

The political advantage is subtle. If you know enough about a person’s preferences, you may not need to command them. You can select the input most likely to move them toward the outcome you want.

This is manufactured consent with individualized feedback.

One reason the premise feels contemporary is that it connects data collection to narrative distribution. The same infrastructure can learn from a population and speak back to it. In that loop, culture as inherited steering becomes culture as adaptive steering.

Variety’s interview with Tumbbad director Anand Gandhi and Zain Memon, syndicated by Yahoo, describes MAYA in precisely systemic terms: invisible control, algorithmic nudges, attention, data ownership, and the shaping of preferences are part of the project’s central political inquiry. The creators argue that resistance begins with noticing the architecture through which minds are shaped.

That is also why “the system is the villain” is a more useful frame than searching for one evil operator. Individual Divyas can have different motives. The underlying capability remains: a platform that knows people well enough can influence the distribution of possible futures.

1984 begins with centralized narrative power

In 1984, information control is visibly political. The Party edits records, restricts language, stages rituals, and forces public agreement.

It is a useful baseline because the mechanism is obvious. One authority wants a population to inhabit one official reality.

Contemporary media systems are more complicated. Different people can receive different realities. That can make manipulation harder to recognize because there is no single poster on the wall to point at.

Feed turns advertising into the atmosphere

M. T. Anderson’s Feed imagines an information network implanted in people’s brains. The feed is social connection, entertainment, shopping channel, and data pipeline at once.

The book’s sharpest idea is that recommendation does not merely respond to desire. It participates in building the categories through which desire is experienced.

If every taste becomes a signal and every signal becomes an opportunity to predict the next taste, the border between customer and product starts to blur.

The Circle and the moral language of total connection

Dave Eggers’s The Circle builds its pressure through a technology company that turns visibility into virtue. Sharing becomes honesty. Privacy starts to look suspicious. Participation becomes a moral expectation.

That shift matters because it shows how systems thinking can reveal a form of power that does not operate through one bad feature. Each feature connects to a norm, each norm creates new data, and the new data justifies more features.

Structural violence in this kind of world can be gentle on the surface. The person who refuses participation loses access, status, convenience, or trust rather than being arrested.

QualityLand makes the recommendation engine absurdly confident

In Marc-Uwe Kling’s QualityLand, automated systems predict what people want so aggressively that they start delivering products without being asked.

The joke works because it exaggerates a real tension. Recommendation engines are sold as tools that reduce friction. Less searching. Fewer irrelevant choices. More personalization.

But friction sometimes protects autonomy. The moment before choosing is where a person can reconsider, compare, or decide that the system’s categories do not fit.

Remove every pause and power that works before choice becomes easier to hide.

The Every: participation as social inevitability

Eggers pushes the logic further in The Every, where a dominant platform absorbs more domains of everyday life and people increasingly accept technological mediation as the normal route to safety, self-improvement, and social belonging.

The satirical exaggeration matters less than the mechanism. A system becomes hard to escape when opting out carries practical and reputational costs.

Control that feels like freedom can be built from a thousand conveniences.

Recommendation is not the same as coercion

It is important not to collapse every recommendation into propaganda. People actively want filtering. A library catalog filters. A friend recommends a movie. A music service helps people discover artists they would never have found alone.

The political questions appear when recommendation becomes opaque, concentrated, behaviorally optimized, and difficult to exit. Systems fiction is useful here because political systems can emerge from feedback loops rather than one command center.

Who sets the objective function? Time spent? Purchases? Political stability? Health? Happiness? Profit? Social peace?

Two systems can use the same data and produce very different worlds because they optimize for different things.

Personalized realities create a new propaganda problem

Traditional propaganda often required repetition across a mass audience. Personalized systems can vary the message.

That changes accountability. A politician can be evaluated against a public speech. A platform that shows different persuasive material to millions of people creates a much harder object to inspect.

The audience may not even know which version of the message other people received.

That is why political science fiction increasingly has to care about information architecture. Public life depends on some shared basis for disagreement. If every person receives a highly optimized narrative environment, the society may lose the common surface on which arguments can occur.

Desire is a political resource

Power is often described as the ability to make someone do something. That definition misses the upstream version: the ability to influence what the person experiences as worth doing.

Once desire becomes a target, propaganda, marketing, social recommendation, and algorithmic governance begin to overlap.

The best science fiction about social media follows that overlap all the way through. It asks not just what a platform shows us, but what kind of people repeated exposure trains us to become.

The recommender can become part of the self

Personalization gets especially strange when people begin using the system’s recommendations as evidence about themselves. If a platform repeatedly tells you that you are a certain kind of reader, voter, shopper, or person, the category can become self-reinforcing. You click because it seems like you. The click becomes new evidence that the system was right.

Fiction can push that loop much further. A character might receive different education, friends, work, entertainment, or political information because a model has assigned them to a type. At that point the model is no longer simply observing a stable identity. It is participating in the production of identity.

That is where systems thinking matters. The harm cannot be located in one recommendation. It appears in the feedback loop among classification, exposure, behavior, and reclassification.

A good story about recommendation engines should therefore ask more than whether the algorithm is accurate. Accuracy can make the problem sharper. If a system predicts a vulnerability correctly and then exploits it, better prediction increases power. If it predicts a preference correctly and then narrows future exposure around that preference, personalization can reduce discovery.

The central political question becomes who has the right to close the loop around another person’s attention.

A system can optimize for a goal nobody chose democratically

Recommendation engines always optimize something, even if the objective is hidden behind the word relevance. A commercial platform may value engagement, purchases, retention, or advertising yield. A state may value stability. A school may value completion.

Fiction becomes politically sharp when characters discover that the system’s objective has quietly become a social objective. If maximizing time-on-platform rewards outrage, the public sphere may become angrier without anybody explicitly deciding that anger was the goal. If a model rewards predictable citizens, eccentricity can become an operational problem.

Manufactured desire is therefore not always a conspiracy. It can be an emergent consequence of optimization.

A recommender is therefore never just a mirror. At scale, it is also part of the room in which preferences develop.