Mathematical Model of Emotional Habituation to Novelty: Modeling with Bayesian Update and Information Theory
Pith reviewed 2026-05-25 10:48 UTC · model grok-4.3
The pith
Habituation to novelty is a drop in Bayesian information gain over repeated exposures of a stimulus.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
We formalized the habituation as a decrement in information gain from a novel event through Bayesian update. We derived the information gained from the repeated exposure of a novel stimulus as a function of three parameters: initial prediction error, initial uncertainty, and noise of sensory stimulus. With the proposed model, we discovered an interaction effect of the initial prediction error and initial uncertainty on habituation. Furthermore, we demonstrate that a range of positive emotions on prediction errors shift toward becoming more novel by repeated exposure.
What carries the argument
Bayesian update that reduces information gain on each repeated exposure of the same stimulus.
If this is right
- Habituation speed is governed by an interaction between initial prediction error and initial uncertainty.
- Positive emotional responses can extend to stimuli that were initially less surprising after sufficient repeated exposures.
- The decay of information gain follows an explicit functional form set by the three parameters.
Where Pith is reading between the lines
- Designs with high initial prediction error may lose appeal faster when uncertainty is also high.
- The model could be used to schedule exposures that keep information gain above a target threshold for longer product engagement.
Load-bearing premise
Emotional valence remains a fixed function of arousal even after repeated exposures, with no fatigue or changing goals entering the dynamics.
What would settle it
An experiment in which measured habituation rates fail to show the predicted interaction between initial prediction error and initial uncertainty, or in which valence decouples from information gain independently of the three parameters.
read the original abstract
Novelty is an important factor of creativity in product design. Acceptance of novelty, however, depends on one's emotions. Yanagisawa, the last author, and his colleagues previously developed a mathematical model of emotional dimensions associated with novelty such as arousal (surprise) and valence. The model formalized arousal as Bayesian information gain and valence as a function of arousal based on Berlyne's arousal potential theory. One becomes accustomed to novelty by repeated exposure. This so-called habituation to novelty is important in the design of long-term product experience. We herein propose a mathematical model of habituation to novelty based on the emotional dimension model. We formalized the habituation as a decrement in information gain from a novel event through Bayesian update. We derived the information gained from the repeated exposure of a novel stimulus as a function of three parameters: initial prediction error, initial uncertainty, and noise of sensory stimulus. With the proposed model, we discovered an interaction effect of the initial prediction error and initial uncertainty on habituation. Furthermore, we demonstrate that a range of positive emotions on prediction errors shift toward becoming more novel by repeated exposure.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a mathematical model of habituation to novelty by formalizing it as a decrement in Bayesian information gain from repeated exposure to a novel stimulus. It claims to derive the information gained after repeated exposures explicitly as a function of three parameters (initial prediction error, initial uncertainty, and sensory noise), reports an interaction effect between the first two parameters on habituation, and demonstrates that the range of positive emotions on prediction errors shifts toward novelty, relying on valence as a fixed function of arousal per Berlyne's theory.
Significance. If the derivation holds and receives empirical support, the work could supply a quantitative information-theoretic account of emotional adaptation to novelty with relevance to product design. The approach extends the authors' prior modeling but currently offers only an unvalidated modeling claim without shown steps or checks.
major comments (3)
- [Abstract] Abstract: the claim that an explicit function of the three parameters was derived is stated, but the manuscript supplies neither the derivation steps nor the resulting closed-form expression, leaving the central mathematical result unverified.
- [Model] Model: the assumption that valence remains a fixed function of arousal (information gain) per Berlyne's theory, with no independent dynamics such as fatigue entering after Bayesian updates, is load-bearing for the claimed emotion-range shift but receives no justification or alternative analysis.
- [Results] Results: the reported interaction effect between initial prediction error and initial uncertainty is conditional on chosen values of the three free parameters, with no sensitivity analysis or external benchmarks supplied to support robustness of the effect.
minor comments (1)
- [Notation] Notation for the three parameters should be defined more explicitly at first use, including their mathematical roles as inputs to the derived function.
Simulated Author's Rebuttal
We thank the referee for the constructive comments, which highlight opportunities to improve the clarity and robustness of the mathematical derivations and analyses. We address each point below and will incorporate revisions accordingly.
read point-by-point responses
-
Referee: [Abstract] Abstract: the claim that an explicit function of the three parameters was derived is stated, but the manuscript supplies neither the derivation steps nor the resulting closed-form expression, leaving the central mathematical result unverified.
Authors: We agree that the derivation steps and closed-form expression for the information gain after repeated Bayesian updates (as a function of initial prediction error, initial uncertainty, and sensory noise) were not presented explicitly enough in the manuscript. In the revised version, we will add a dedicated subsection detailing the step-by-step derivation from the Bayesian update rule through to the explicit functional form, including all intermediate expressions. revision: yes
-
Referee: [Model] Model: the assumption that valence remains a fixed function of arousal (information gain) per Berlyne's theory, with no independent dynamics such as fatigue entering after Bayesian updates, is load-bearing for the claimed emotion-range shift but receives no justification or alternative analysis.
Authors: The model extends our prior work by adopting Berlyne's arousal-potential framework in which valence is a fixed function of arousal (information gain). We will add an explicit justification paragraph in the Model section explaining this modeling choice as an initial simplification focused on the information-theoretic component, while acknowledging the absence of fatigue or other dynamics as a limitation and briefly discussing how such extensions could be incorporated in future work. revision: yes
-
Referee: [Results] Results: the reported interaction effect between initial prediction error and initial uncertainty is conditional on chosen values of the three free parameters, with no sensitivity analysis or external benchmarks supplied to support robustness of the effect.
Authors: We acknowledge that the interaction effect is demonstrated for specific parameter values and that robustness checks are needed. In the revised manuscript, we will include a sensitivity analysis section that varies the three parameters over plausible ranges (supported by references to empirical literature on prediction error and uncertainty) and shows that the interaction on habituation persists under those conditions. revision: yes
Circularity Check
No circularity; derivation applies standard Bayesian information gain independently
full rationale
The paper derives the information-gain expression for repeated exposures directly from Bayesian updating rules and the definition of information gain, expressing it as an explicit function of the three stated parameters. The interaction effect follows from algebraic analysis of that closed-form expression rather than from any fit or redefinition. The valence-arousal mapping is an explicit modeling assumption imported from Berlyne's external theory (with prior formalization cited), not a self-referential step that forces the habituation result. No equation reduces to its own inputs by construction, no prediction is statistically forced by parameter fitting, and the self-citation for the base emotional model is not load-bearing for the new Bayesian habituation chain.
Axiom & Free-Parameter Ledger
free parameters (3)
- initial prediction error
- initial uncertainty
- noise of sensory stimulus
axioms (3)
- standard math Belief updating follows standard Bayesian inference
- domain assumption Arousal equals Bayesian information gain
- domain assumption Valence is a function of arousal per Berlyne's theory
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.