Immersion as inference: An active inference account of induced realities and its relation to the two-competitor model

Immersion is formalized in active inference as the belief that an induced reality best explains current sensory signals and supports the winning model, even after Bayesian model switching.

• Why it matters: Immersive systems need inputs that match how the brain samples perception, but a unified computational account is missing for how attention, affect, and prediction errors jointly sustain immersion.
• What they did: The framework treats induced realities (virtual, augmented, mixed, or even book/daydream) as competing generative models, with immersion determined by model selection under high-precision prediction error; it recovers the two-competitor attentional focus vector and angle.
• The result: A minimal simulation reproduces the rewarding and fatiguing scenarios interrupted by a slamming door, linking presence, embodiment, cybersickness, and narrative absorption to a single quantity and enabling psychophysiological prediction and closed-loop control.

AI summary of the abstract · may contain errors · not medical advice · read the paper.