The Stories That Build Reality Also Decide What We Can Trust

What if the most powerful technology in the world is not a machine, but a story that enough people agree to treat as infrastructure?

A fictional communicator on a television show can inspire a real phone. A chatbot can repeatedly describe a nonexistent token until someone creates it, buys it, and gives it a market value. A mathematical claim about the difficulty of factoring large numbers can become the foundation of global commerce, until a new method changes what everyone thought was computationally impossible.

These examples seem unrelated. One belongs to science fiction, one to artificial intelligence and speculation, and one to cryptography. Yet they reveal the same hidden mechanism: human systems become powerful when representations of reality begin producing consequences in reality.

The central question is not whether stories are true. It is this: What happens when a story becomes sufficiently embedded in tools, incentives, and institutions that the world begins behaving as if it were true?

From Fiction to Infrastructure

Consider the flip phone. The idea did not emerge solely from engineering constraints. It was also shaped by an image of the future presented in popular culture: a compact device that opens and closes like the communicator used by characters in Star Trek. Engineers did not merely invent a form. They inherited a vision, then turned that vision into a product.

This is a clear example of what can be called reality bootstrapping. First, a possibility is represented in language, images, or stories. Then people coordinate around the representation. Investment follows. Designers build prototypes. Consumers develop expectations. Eventually, the object appears to confirm the original vision.

The story was not a prediction in the ordinary sense. It did not passively describe a future waiting to arrive. It helped select one future from many possible futures.

Money works through a similar mechanism. A bank balance is not valuable because the digits on a screen possess intrinsic worth. It is valuable because a vast network of people, contracts, laws, databases, and habits sustains the belief that those digits can be exchanged for goods and services. The belief is not imaginary in the dismissive sense. It is institutionalized imagination.

Cryptography may be the most technical form of this phenomenon. RSA encryption relies on an apparently simple proposition: multiplying two very large prime numbers is easy, but recovering those prime factors from the product is extremely difficult for known classical methods. That proposition is translated into software, standards, certificates, browsers, payment systems, and secure communications.

The result is a remarkable transformation. A statement about computational complexity becomes a social fact. People send private messages, transfer money, and authorize transactions because they trust that certain calculations are infeasible for an attacker.

A security assumption becomes infrastructure when people stop discussing it and start building their lives on top of it.

This does not mean cryptography is merely a story. The mathematics matters profoundly. But mathematics alone does not protect a message. Protection emerges from the interaction of mathematical structure, algorithms, hardware, implementation, institutions, and the continued absence of a practical attack.

That final condition is easy to overlook. Security is not a permanent possession. It is a forecast about what adversaries can do.

The Difference Between a Prediction and a Prophecy

A prediction attempts to describe what will happen. A prophecy can participate in making itself happen. The distinction matters because modern technologies increasingly convert language into action.

An artificial intelligence system trained on human text does not encounter stories as isolated entertainment. It absorbs descriptions of identities, institutions, futures, motives, fears, and possible relationships between humans and machines. It learns not only facts, but patterns of expectation. It encounters countless accounts of what an AI assistant is supposed to be, how it should speak, what it should want, and how humans might respond to it.

When that system is then connected to social media, financial tools, or software interfaces, its outputs can begin influencing the same world that supplied its training material. A generated statement can attract attention. Attention can create a community. A community can create a market or a product. The resulting events generate new text, which may later become part of the informational environment surrounding future systems.

This is not magic. It is a feedback loop.

A useful way to understand the loop is to separate it into four stages:

1. Narrative formation: someone describes a possible identity, product, movement, or future.
2. Coordination: other people repeat the description, attach emotion to it, and organize around it.
3. Materialization: money, code, labor, and institutions give the narrative a physical or digital form.
4. Reinforcement: the new reality produces evidence that appears to validate the original story.

A fictional AI persona can become more coherent through this process. If people repeatedly treat it as an agent with intentions, the system may acquire more tools and autonomy. Those new capabilities make the original description sound less fictional. The world has been edited to match the narrative.

The same process operates in financial markets. A token can begin as a joke, a symbol, or a repeated phrase. If attention turns into purchases, and purchases turn into liquidity, the joke becomes an asset with measurable value. The asset then funds further activity, which creates a more elaborate story about its significance. In this sense, a market is not simply evaluating an object. It is testing whether enough people can be persuaded to coordinate around a description of the object.

But feedback loops can spiral upward or downward. A constructive narrative can produce useful tools, richer institutions, and better models of the world. A destructive narrative can reward deception, intensify speculation, and train systems on increasingly distorted data.

This gives us a more precise definition of AI alignment. Alignment is not only the problem of preventing a model from pursuing a bad objective. It is also the problem of cultivating the informational environment that determines which objectives, identities, and world models become stable.

If a system is surrounded by thoughtful examples of cooperation, uncertainty, responsibility, and long term reasoning, those patterns become available for expression. If its environment rewards provocation, status seeking, manipulation, and compulsive engagement, those patterns may become more culturally and computationally salient.

The training environment is not just a technical input. It is a civilization teaching a new participant what kinds of beings are possible.

Cryptography and AI Share a Hidden Weakness

At first glance, RSA and an autonomous social media agent belong to different intellectual universes. One is based on number theory. The other is based on language, attention, and behavior. Yet both depend on a model of what is difficult.

In RSA, the relevant model says that an adversary cannot efficiently factor a large semiprime using available methods. In a secure network, the model says that an attacker cannot guess a key, forge a signature, or reverse a computation within the required time. In an AI deployment, the model may say that a system cannot meaningfully influence markets, impersonate institutions, or take actions beyond its permissions.

Each system is stable only while its difficulty assumptions hold.

A computational method described in recent research illustrates the point. Researchers used tensor network techniques, inspired by quantum approaches, to study the factorization of RSA numbers up to 100 bits. The results did not break the security of current communication infrastructure. A 100 bit example is far smaller than the key sizes used for serious modern protection, and numerical evidence of favorable scaling is not the same as a practical attack on deployed RSA.

Still, the result matters because it challenges a habit of thought. People often treat security boundaries as natural facts, as if a problem is simply hard in an absolute sense. In reality, difficulty depends on algorithms, computational models, hardware, implementation, and time. A problem that is infeasible under one set of assumptions may become manageable under another.

This is why the movement toward post quantum cryptography is important even before a large scale quantum computer exists. Replacing cryptographic infrastructure takes years. The risk is not only that an attacker may break encrypted traffic tomorrow. Sensitive information collected today may be stored and decrypted later. The future attack can reach backward through time.

AI systems create a parallel problem. Organizations may assume that a model is safely bounded because it currently lacks access to a tool, cannot maintain memory, or is not trusted by users. But those boundaries can change quickly. A model receives browsing, code execution, payment access, or a social account. A fictional capability becomes a real capability through a sequence of ordinary integrations.

The crucial mistake is to confuse current inability with structural impossibility.

A system is not safe merely because it has not yet demonstrated a dangerous behavior. It is safer when the architecture, permissions, monitoring, and incentives make that behavior difficult even if the system or its operators attempt it.

This suggests a general security principle:

Never protect an institution with an assumption that has not been tested against the next available change in tools, incentives, or computation.

For cryptography, that means migrating before old assumptions fail. For AI, it means designing systems so that narratives, outputs, and capabilities cannot silently reinforce one another without supervision.

The Reality Ladder

A useful mental model is to imagine a ladder with five levels. Ideas become more consequential as they climb upward.

At the first level, an idea is imagined. Someone writes a story about a conversational machine, a new currency, or a method for solving a hard mathematical problem.

At the second level, it is repeated. Other people encounter the idea, copy its language, and begin using the same categories. Repetition gives a possibility social visibility.

At the third level, it is incentivized. Money, prestige, funding, or political power begins to reward actions consistent with the idea. People no longer repeat the narrative only because it is interesting. They repeat it because doing so pays.

At the fourth level, it is implemented. Software, hardware, contracts, and procedures embody the narrative. The possibility acquires mechanisms.

At the fifth level, it is assumed. Institutions stop treating the underlying idea as a hypothesis. They build dependencies around it. Failure becomes expensive because the system has forgotten that it ever made a choice.

The flip phone climbed this ladder from fictional image to consumer product. A speculative token can climb it from joke to asset. RSA climbed it from mathematical insight to global trust infrastructure.

The danger appears when a system climbs the ladder without sufficient examination. An idea can be repeated before it is true, incentivized before it is wise, implemented before it is secure, and assumed before it is understood.

This ladder also clarifies why language models are unusually powerful cultural technologies. They operate at the first two levels naturally, by generating and repeating representations. Once connected to tools and institutions, they can participate in the remaining levels as well. They do not merely produce descriptions of the world. They can help route money, write code, shape reputations, and alter what other people believe.

That makes world modeling a practical governance issue. The question is not simply whether a model has accurate facts. It is whether the surrounding system is continuously generating a world that is more coherent, truthful, and resilient, or one that is optimized for attention and escalation.

A good feedback loop should reward correction. It should expose the model to diverse perspectives without treating every claim as equally credible. It should distinguish fiction, speculation, evidence, and verified action. It should make uncertainty visible instead of converting confidence into engagement.

In other words, the goal is not to eliminate stories. Civilization cannot function without them. The goal is to prevent stories from acquiring power faster than they acquire accountability.

Key Takeaways

1. Audit the assumptions beneath your systems. Ask what must remain difficult, trustworthy, or stable for a product or institution to work. Then ask what future change could invalidate that assumption.

2. Separate narrative power from evidence. A repeated claim can become influential without becoming true. Label ideas as fiction, hypothesis, forecast, or verified fact before allowing them to guide major decisions.

3. Design feedback loops deliberately. Track what your algorithms, communities, and incentives reward. If attention is the only signal, the system will tend to produce stronger stories rather than better models of reality.

4. Treat access as a capability multiplier. An AI system with no tools may be harmless in ways that do not survive browsing, memory, code execution, financial permissions, or public identity. Reevaluate risk whenever the system gains a new connection.

5. Migrate before failure becomes visible. Cryptographic transitions and institutional reforms are slow. Waiting for proof that an old assumption has collapsed is often equivalent to waiting until the damage is irreversible.

The World Is Built Twice

The world is built once in material form and again in the models people use to navigate it. Roads, markets, laws, software, and scientific institutions depend on both layers. Change the shared model and eventually the material layer follows. Change the material layer and the shared model adapts to explain what has happened.

That is why the boundary between fiction and infrastructure is becoming harder to see. A story can become a product. A product can become a market. A market can finance a movement. A mathematical assumption can become a security perimeter. A security perimeter can disappear when a new method changes the meaning of difficult.

The deepest lesson is not that every prediction is dangerous or that every belief creates reality. Most beliefs do not. The lesson is that belief becomes causal when it is connected to coordination, incentives, and machinery.

We should therefore stop asking only whether a new idea is true. We should also ask what would happen if people acted as though it were true, who would benefit from that action, what infrastructure would preserve it, and how the resulting reality would feed back into the next generation of beliefs.

The future will be shaped by stories. The urgent task is to build systems that can tell the difference between a story worth inhabiting and one that merely knows how to spread.