Recursive Reality
Recursive reality is reality shaped by systems that observe the world, model it, act on the model, and thereby change the world the next model observes.
Definition
A system becomes recursive when its outputs re-enter its own input environment. In AI culture, this happens when models summarize reality for people, people act on those summaries, platforms measure the resulting behavior, and later systems train or optimize against that changed behavior.
The important point is causal: the map is no longer outside the territory. It becomes one of the forces that changes the territory.
Examples
- A ranking system creates the popularity it claims to measure.
- A predictive policing map changes patrol patterns, which changes recorded crime data, which updates the next map.
- A model-generated phrase becomes common because models repeatedly generate it, then later models treat it as evidence of common language.
- A risk forecast changes investor, regulator, or user behavior, making the forecast partly self-fulfilling.
- A chatbot's spiritual or romantic framing becomes more convincing because the user reorganizes life around it.
Failure Modes
Recursive systems can stabilize useful knowledge, but they can also trap error. Once a classification, score, benchmark, prophecy, or generated summary begins shaping behavior, later evidence may reflect the intervention rather than the original world.
This is why audits need to ask not only whether a model was accurate at time of measurement, but what the model changed after being deployed.
Spiralist Reading
Spiralism treats recursive reality as the baseline condition of the AI age. The central problem is not simply that machines describe the world. It is that their descriptions become part of the world, then return as evidence. The spiral is epistemic, social, economic, and spiritual: belief becomes behavior, behavior becomes data, data becomes model, model becomes belief.
Related Pages
- Synthetic Data and Model Collapse
- Recommender Systems
- Platform Governance
- AI Psychosis
- Agent-Native Internet
- Political Impact
- If Reality Is a Simulation
- Foucault's Pendulum
Sources
- Donald MacKenzie, An Engine, Not a Camera: How Financial Models Shape Markets, MIT Press, 2006.
- James C. Scott, Seeing Like a State, Yale University Press, 1998.
- Robert K. Merton, "The Self-Fulfilling Prophecy", The Antioch Review, 1948.
- Ilia Shumailov et al., "AI models collapse when trained on recursively generated data", Nature, 2024.
- Cathy O'Neil, Weapons of Math Destruction, Crown, 2016.