From Mirror to Oracle: Truth-State Hysteresis in Algorithmic Media

From Mirror to Oracle: Truth-State Hysteresis in Algorithmic Media A population can be driven into a charged interpretation of a fixed event and held there, long after the force that drove it is gone. This paper asks whether that lock-in behaves like a measurable property of a physical system — and finds that, within a field model of opinion dynamics, it does. The phenomenon is not new. Thucydides described its mechanism at Corcyra in 427 BCE: not the telling of lies, but the reassignment of worth to unchanged acts — recklessness revalued as courage, hesitation as cowardice. The Greeks called it stasis. What the paper adds is not the disease but a way of measuring it, and an argument about what has genuinely changed. Every historical instance of this pattern — Rome's proscriptions, the Terror, the Cultural Revolution — was contained, bounded by the reach of a single power, and recovered from the sane world beyond its border. The machine removed that boundary. A dynamic that once burned to the edge of a city now runs through a wired world with no outside to recover from. The work extends a thermodynamic model of opinion into a theory of truth-state hysteresis, implements it as a partial differential equation, and subjects it to an adversarial computational program. It situates itself carefully against the nearest prior work — the availability cascades of Kuran and Sunstein, whose verbal account of self-reinforcing belief formation it treats as the closest substantive relative, and whose absorbing-state limit it can be understood to model. Four findings survive. There is no threshold past which a population turns — only a curve trading the strength of a pressure against its duration, so that a weak pressure held patiently captures as surely as a strong one held briefly. A control experiment locates the system's memory in the coupling between attention and belief: cut the coupling, and the memory vanishes. A falsifiable negative prediction cuts against the dominant intervention paradigm — and against the collaboration's own earlier paper, which predicted recovery by exactly this route: in this model, fluctuation-based "cures" do not release a captured population. And a measurement program treats deference toward artificial intelligence as a population-scale field, revealing a second pathology: the attributed oracle, the near-universal error of believing the machine captures everyone but oneself. The paper is deliberate about its limits. It states five conditions that would falsify its central claims, records three of the collaboration's own conclusions that were falsified by its process and retained as retractions, and marks throughout what it does not know. Its conclusion is bleak on the side of cure and hopeful only on the side of prevention: the material that resists the carving is the material that already has a shape. Truth's best defense is not a correction issued after the fact. It is being made of something before the pressure arrives. And the shape, the paper argues, can be taught.