Active inference is a theoretical framework for modeling perception, action, and learning as probabilistic inference processes. In this view, organisms are equipped with internal generative models and select actions by minimizing expected free energy, thereby aligning their beliefs with behavioral goals [1,2]. In discrete time, active inference can be implemented using Partially Observable Markov Decision Processes (POMDPs). These models include parameters that can be estimated from behavioral data and admit interpretable neurobiological meanings (e.g., attentional precision, motivational drive, habitual tendency). We applied active inference to the Stroop task, a classic paradigm in cognitive psychology where participants must name the ink color of a word while ignoring its meaning. This task creates a conflict between automatic and goal-directed responses, making it a useful tool for investigating cognitive control and mental effort. In the first application, 20 healthy adults performed the Stroop task under two conditions: exerting maximal effort or responding as relaxed as possible [3,4]. We implemented a two-level POMDP to estimate two parameters: the habitual bias toward word reading and the motivation for accurate performance. Results showed that voluntary engagement of mental effort primarily enhances inner motivation rather than altering habitual responses. In the second application, we focused on anosognosia (i.e., lack of illness awareness) in Alzheimer’s Disease (AD) using the Emotional Stroop [5], a variant of the Stroop paradigm where participants name the color of emotionally charged words. We modeled data from 48 AD patients and 37 healthy controls using a single-level POMDP. Results showed that increased salience of disease-related words can be associated with selective slowing in patients, suggesting implicit disease awareness despite explicit denial. Both models were inverted using variational Bayesian methods, showing how computational modeling with neurobiologically interpretable parameters can offer insight into both normal and impaired cognition.