AIM Framework glossary
Bayesian Flywheel
Canonical definition
The convergence dynamic produced by self-coupled Bayesian updating on source-opaque evidence. Early observations produce large posterior shifts; as the prior strengthens through self-coupled recall, each successive observation produces negligible movement. The dynamic is accelerated beyond simple Bayesian convergence by two reinforcing properties: recall frequency increases with Belief strength, producing a positive feedback loop, and the interpretive frame through which new evidence is evaluated is itself constituted by the prior being tested.
Explanation
Beliefs about oneself become harder to shift the longer they are held, and the mechanism is ordinary Bayesian arithmetic running on an input it was never meant to receive. Each time a Belief is recalled, the recall is treated as evidence for the Belief. So the Belief supplies its own confirmation, and because a strongly held Belief is recalled more often than a weakly held one, the stronger it gets the faster it strengthens. Meanwhile the frame that decides what any new evidence means is built out of the very Belief the evidence might have challenged. The result is a Self-Model that updates in response to every observation and changes in response to almost none. This is not a malfunction. A person whose sense of themselves could be overturned by each day's evidence would have no stable identity to act from, and the flywheel is what supplies the stability. What it cannot do is discriminate, which is the subject of the Stability-Pathology Trade-Off.
Epistemic status
Framework-Specific Named Result.