Murayama and Jach (M&J) include subjective experiences as input and output of mental computational mechanisms in their sophisticated model of motivation (their Figs. 1 and 2). However, the role of subjective experiences remains underspecified. Our commentary highlights the central role of metacognitive feelings in mental computational processes. We argue that subjective feelings are not just input and output but an integral part of mental computational processes, and they influence the outcome of computations. Motivation could be seen in terms of mental computational processes continuously monitored and regulated by metacognitive feelings. Metacognitive feelings are subjective experiences that inform an individual about cognitive processes and include affect, subjective certainty, and fluency, which is the ease with which a mental process is executed (Efklides, Reference Efklides2006; Schwarz & Clore, Reference Schwarz, Clore, Kruglanski and Higgins2007). Such feelings provide continuous information about an individual's interaction with the environment. Positive affect, high fluency, and high certainty indicate that the interaction with the environment proceeds smoothly; thus, individuals do not need to change their behavior. By contrast negative affect, low fluency, and low certainty indicate a problem whose solution needs new information or behavior change.

The following evidence supports the notion that mental computational processes are interwoven with metacognitive feelings. First, affect is inherent in perceptual processes. Brief exposure to a coherent but non-recognizable outline yields more positive affect than a non-coherent outline, as measured by EMG activity of Zygomaticus Major, the “smiling muscle” (Erle, Reber, & Topolinski, Reference Erle, Reber and Topolinski2017; Flavell, Tipper, & Over, Reference Flavell, Tipper and Over2018; Topolinski, Erle, & Reber, Reference Topolinski, Erle and Reber2015). Cognitive involvement in these evaluations was minimal, which means that affect may occur before the mental computational processes. Similarly, success of later perceptual processes, such as identification of objects or solving mental rotation tasks, yields positive affect (Lindell, Zickfeld, & Reber, Reference Lindell, Zickfeld and Reber2022). These studies show that there never is “no affect,” in contrast to models where affect only serves as input or output. The interesting question will be how the ongoing dynamics of affect in perception that guide attention influence mental computational processes.

Second, feelings of knowing and judgments of learning guide study decisions (Brooks, Yang, & Köhler, Reference Brooks, Yang and Köhler2021; Hanczakowski, Zawadzka, & Cockcroft-McKay, Reference Hanczakowski, Zawadzka and Cockcroft-McKay2014; Metcalfe & Finn, Reference Metcalfe and Finn2008). One underlying experience is fluency, either at retrieval or encoding. For example, the easier it is to retrieve fragments of the study materials, the more learners feel they know (Koriat, Reference Koriat1993); the easier it is to encode materials, the higher learners will judge the learning outcomes (Benjamin, Bjork, & Schwartz, Reference Benjamin, Bjork and Schwartz1998; Koriat & Ma'ayan, Reference Koriat and Ma'ayan2005; Rawson & Dunlosky, Reference Rawson and Dunlosky2002). Interestingly, judgments of learning may contradict learning outcomes because items that are easy to generate (e.g., answers to general knowledge questions) are often more difficult to retrieve later than items that are difficult to generate (Benjamin et al., Reference Benjamin, Bjork and Schwartz1998). Indeed, in a training program to learn typing, a schedule that made learning difficult but yielded superior learning outcomes was liked less than a schedule that made learning easy but yielded inferior learning outcomes. Not surprisingly, the former group wanted to change to the easier schedule because they erroneously believed that they would learn faster (Baddeley & Longman, Reference Baddeley and Longman1978). As we understand it, M&J's model cannot explain motivational effects where metacognitive feelings are not in line with the outcome of mental computational processes. Any theory of motivation needs to explain results where metacognitive feelings seem to play an independent role in learning decisions.

Third, research on intuitive problem solving suggests that perceived solution progress deviates from actual progress, which M&J's model again has difficulties to explain. Studies combining a stepwise problem-solving paradigm (Bowers, Regehr, Balthazard, & Parker, Reference Bowers, Regehr, Balthazard and Parker1990) with “feelings of warmth” indicating closeness to the solution (Metcalfe & Wiebe, Reference Metcalfe and Wiebe1987) showed that participants felt far from the solution until right before they solved the task, even though their actual progress was closer to the solution than they were aware of (Bowers, Farvolden, & Mermigis, Reference Bowers, Farvolden, Mermigis, Smith, Ward and Finke1995; Reber, Ruch-Monachon, & Perrig, Reference Reber, Ruch-Monachon and Perrig2007). Evidence about the role of metacognitive judgments in learning decisions (e.g., Metcalfe & Finn, Reference Metcalfe and Finn2008) suggests that problem solvers are more likely to leave a task unsolved if it feels difficult, even if the underlying computational processes linearly progress toward the solution.

Fourth, in an aha-experience, mental processes that lead to sudden insight are accompanied by metacognitive feelings (e.g., Skaar & Reber, Reference Skaar and Reber2020; Webb, Little, & Cropper, Reference Webb, Little and Cropper2018; for a review, see Wiley & Danek, Reference Wiley and Danek2024). As cognitive insight and metacognitive feelings appear simultaneously, an aha-experience is a unified construct, and metacognitive feelings do not just serve as input or output.

Finally, affect plays a major role in early phases of interest development (see Hidi & Renninger, Reference Hidi and Renninger2006). Although individuals may not be consciously aware of their interest especially in the earliest phases of interest development (Hidi & Renninger, Reference Hidi and Renninger2006; Renninger, Talian, & Kern, Reference Renninger, Talian, Kern and Fisher2022), affective experiences may have a crucial role in influencing preferences and behavior during engagement with an object (Krapp, Reference Krapp2007). Again, affect is not just input and output, but an integral part of engagement with a subject.

These findings connect to recent theories that combine computational models, such as reinforcement learning (Brielmann & Dayan, Reference Brielmann and Dayan2022) and predictive coding (Brouillet & Friston, Reference Brouillet and Friston2023; Fernández Velasco & Loev, Reference Fernández Velasco and Loev2024; Yoo, Jasko, & Winkielman, Reference Yoo, Jasko and Winkielman2024) with metacognitive feelings, mainly fluency. Fluency could be seen as a parameter in computational processes, for example, short-term value in reinforcement learning or prediction precision in predictive coding. Moreover, feelings may help determine the course of action, as Fernández Velasco and Loev (Reference Fernández Velasco and Loev2024) propose. According to this hypothesis, mental computational processes and metacognitive feelings take different roles in knowledge acquisition; the former compute the predictive dynamics whereas feelings guide action, akin to action tendencies inherent in emotions (e.g., Frijda, Reference Frijda1988).

Including algorithms of predictive coding and reinforcement learning would be a promising avenue to develop M&J's proposed model. Performance predictions based on metacognitive feelings play a major role in learning decisions. Thus, predictive coding accounts may refine the proposed model. Reinforcement learning (Sutton & Barto, Reference Sutton and Barto2018) seems promising because it builds on similar assumptions of recursive processes, including reward, as the proposed model.