Learning from informative losses boosts the sense of agency

Abstract

Sense of agency, the feeling of having control over one’s actions, is modulated by whether one’s choices lead to desired or undesired outcomes. Learning similarly depends on outcome values from previous experience. In the current study, we evaluate a possible link between the sense of agency and learning, by investigating how intentional binding, an implicit measure of agency, changes during a probabilistic learning task. In two experiments, we show increased intentional binding in trials that follow losses, compared with trials that follow wins. Experiment 1 demonstrated that this post-error agency boost (PEAB) effect is rule-specific, as it did not occur if the trial following an error involved different action–outcome contingencies. Furthermore, PEAB was not modulated by the type of outcome presentation (monetary vs. affective). Experiment 2 showed that the PEAB effect can also occur when the current action involves a forced (as opposed to free) choice, but only when the previous, loss-provoking action was chosen freely. Thus, PEAB occurs when current actions are informed by outcomes of one’s own previous action choices. Electroencephalography (EEG) data linked these effects to two event-related potential components, namely, the Feedback Related Negativity and the P300. Taken together, these results support the notion that PEAB reflects an adaptive property of human sense of agency, facilitating effective learning about the action–outcome structure of a specific task, to optimise future performance. By clarifying the conditions for enhancing the sense of agency through learning, this work adds to our understanding of human learning and agency.

Sense of agency decreases during learning when to act: an intentional binding study

“…These results are thus in line with the studies showing that sense of agency is associated with difficulty, feedback-guided adaptation, and the process of updating action values (Di Costa et al, 2017;Majchrowicz et al, 2020).…”

“…A second tradition investigated the perceptual shift of an action towards a subsequent tonean effect known as intentional binding (Haggard et al, 2002). In reversal learning tasks, this binding was increased after errors (Di Costa et al, 2017;Majchrowicz et al, 2020). Since an error triggers increased cognitive control (Botvinick & Cohen, 2014), those results suggest the sense of agency is associated with difficulty, feedback-guided adaptation, and the process of updating action values.…”

Consciousness, Exascale Computational Power, Probabilistic Outcomes, and Energetic Efficiency

“…The concept that probabilistic coding in cortical neural networks might be intrinsically connected to the emergent psychological features of cortical neural networks is not well‐established in the field of cognitive sciences. However, recent work by multiple groups has converged on the importance of this perspective, demonstrating that probabilistic approaches enhance deep learning (Gutman & Hyvärinen, 2013), lead to the generation of predictive models (Jara‐Ettinger, Schulz, & Tenenbaum, 2020), yield context‐dependent psychological states (Bruza et al., 2023; Chang, Biehl, Yu, & Kanai, 2020), and contribute to a sense of agency when navigating uncertain situations (Majchrowicz, Kulakova, DiCosta, & Haggard, 2020). Of course, the question of whether humans can actually exert agency, and whether such a capability might be compatible with causal determinism, is still hotly debated (Dennett 2014; Khalighinejada, Schurger, Desantisa, Zmigrod, & Haggard, 2018; O'Connor & Franklin, 2020).…”

The Interplay Between Affective Processing and Sense of Agency During Action Regulation: A Review

“…Some experiments found that the valence of action effects might not only influence temporal binding between the action and the ensuing effect itself but could also have an impact on subsequent actions in the same task. For reinforcement learning tasks, it was found that negative compared to positive performance feedback on a trial increased intentional binding for actions on the subsequent trial (Di Costa et al, 2018 ; Majchrowicz et al, 2020 ). Errors are known to evoke increased top-down control of one's behavior to improve performance on subsequent trials (Ullsperger et al, 2014 ).…”

“…Errors are known to evoke increased top-down control of one's behavior to improve performance on subsequent trials (Ullsperger et al, 2014 ). Therefore, stronger binding after errors could indicate that engaging in top-down control is related to an enhanced sense of agency (Majchrowicz et al, 2020 ).…”