Theory of Sub-entropic Attention: Quantifying the Predictability of Stimuli

In this study, we introduce the theoretical concepts and test the empirical predictions of the Theory of Sub-entropic Attention (TSA), a novel framework for modelling subjective entropy (Sub-en) in human predictive behaviour. TSA posits that humans implicitly employ Higher Order Markov Models (HOMMs) of individualised orders to predict the next symbol in binary sequences. The internal beliefs of such models enable the emergence of subjectively perceived structure, even in randomly generated sequences, when Sub-en falls below an entropic threshold. We conducted a single-participant pilot-study to investigate two related hypotheses: i) Participants achieve higher accuracy when predicting the next symbol in sub-entropic compared to super-entropic sequences. ii) Motor-evoked potentials (MEPs), elicited by transcranial magnetic stimulation (TMS) and used as an outcome measurement, reflect greater prediction errors in sub-entropic conditions.Behavioural results supported the first hypothesis. Neural data showed significantly larger MEP amplitudes for incorrect predictions in sub-entropic compared to super-entropic conditions, supporting the second hypothesis and suggesting a domain-general neural response to prediction errors that consequently decreases with greater internal baseline uncertainty. These findings provide initial empirical support for TSA and establish Sub-en as an objective metric for quantifying subjective predictability.