In our everyday life, we physically interact with the environment by changing the course of external events. Under normal conditions, we are consciously aware of being the agents of our own actions by assuming responsibility for their effects. This phenomenon is called “Sense of Agency” (SoA; Haggard and Chambon, 2012; Haggard, 2017).

In recent years, several studies have characterized the features of SoA and endorsed the hypothesis that it consists of both prospective and retrospective cognitive evaluations of executed actions and their consequences (e.g., Gallagher, 2012; Chambon et al., 2014; Haggard, 2017). Specifically, before performing an action, the agent would first select which action to perform from a variety of possibilities and, in a second step, a prediction of its future output would be envisaged (the prospective component). As soon as the action has been performed, the feedback from its outcome is available to the agent, allowing comparison between the predicted and actual signals (the retrospective component). The final output of this comparison, which is called the “prediction error,” is essential for defining the strength of SoA: people will feel more in control of their actions if a prediction error does not occur (or it is negligible), compared to situations where there are substantial incongruencies between the predicted and actual outcomes (Fletcher and Frith, 2009; Carruthers, 2012; Seth et al., 2012; Haggard, 2017). Thus, temporality plays a pivotal role in this “comparator model” of SoA: events occurring too early or too late in time compared to the performed action are not considered to be linked outcomes (Frith et al., 2000; Ruess et al., 2018).

Unsurprisingly, temporal cues have been usefully employed to experimentally study SoA, as in the case of the Intentional Binding paradigm. Intentional Binding takes advantage of a temporal delay elapsing between an action and its sensory consequence (Engbert et al., 2007), and it stands out as the most widely used paradigm for investigating implicit SoA in the existing literature (for a critical view on its validity, see Suzuki et al., 2019). After a voluntary, intentional action (e.g., a key press) which generates an outcome after a short delay (e.g., a sound), participants report the time of the action or its outcome. Results typically show a reduced temporal estimate of the interval between the action and its outcome for voluntary actions compared with involuntary movements (Haggard et al., 2002). This compressed estimate is explained by delayed perception of the action execution as well as anticipated perception of outcome onset. On the contrary, when participants perform an action involuntarily – by having body parts passively moved – this temporal compression is not observed (Haggard et al., 2002; Engbert et al., 2008). Authors proposed that this discrepancy likely reveals the cognitive mechanisms related to SoA; in particular, the more people feel in control of an action, the stronger the temporal link between the action and its outcome (Poonian and Cunnington, 2013; Imaizumi and Tanno, 2019; Suzuki et al., 2019; Caspar et al., 2020; Ciardo et al., 2020).

Interestingly, studies have shown that pressing a button eliciting a tone and watching another person performing the same action led to similar perceived shortenings of the action-outcome interval, compared with a closely matched control stimulus (Poonian and Cunnington, 2013). These results suggest that similar processes are implemented to make causal attributions of sensory outcomes to our own actions and to others’ actions. The link between our own actions and others’ actions recalls the “motor resonance” mechanism, which refers to the activation of motor neurons when watching actions performed by other people (Avenanti et al., 2007; Uithol et al., 2011; Spaccasassi et al., 2022), supporting action understanding and imitation (di Pellegrino et al., 1992; Rizzolatti et al., 2006).

This self-other overlap goes beyond the motor domain; for instance, understanding others’ thoughts relies on the ability to understand our own thoughts (Carruthers, 1996; Waytz and Mitchell, 2011). This higher level “simulation” mechanism is well-documented in the moral cognition field: the greater the development of moral reasoning capabilities, the stronger the ability to understand the intentionality of others’ actions (Buon et al., 2013). Despite the growing interest in the scientific investigation of moral cognition (for a review, see Greene, 2015), as well as SoA (Moore et al., 2009; Moore and Haggard, 2010; Beck et al., 2017; Lush et al., 2017; Caspar et al., 2020; Ciardo et al., 2020), to date, evidence of their relation is meager. To fill this gap, here, we combined two behavioral tasks addressing moral cognition and SoA. We asked participants to perform an online revised version of the Empathy for Pain task (Decety et al., 2011) by rating, along different axes, harmful and neutral scenarios where two characters interact. In the Intentional Harm scenario, an offender deliberately inflicts pain on a victim. In the Accidental scenario, the agent accidentally provokes harm to the victim. In the Neutral scenario, the two characters’ interaction is emotionally neutral and no harm occurs. We also asked participants to perform the Intentional Binding task online to assess their SoA (Imaizumi and Tanno, 2019). If people with a stronger sense of responsibility for their own actions tend to ascribe more responsibility to the agents of observed actions, we hypothesized that they would attribute more intentionality to others, leading to harsher moral evaluations (and, possibly, a stronger desire to punish the agent) when the outcomes of the observed actions are negative, as in the case of scenarios depicting harmful interactions.

In the present study, we sought to investigate whether there is a correspondence between the Sense of Agency (SoA) and the way we morally evaluate others’ actions. SoA was measured through the Intentional Binding task (Imaizumi and Tanno, 2019), which allowed us to measure, both implicitly and explicitly, the subjective feeling of being in control of one’s own actions by means of temporal estimation and agency ratings, respectively. In the Moral Cognition task, participants evaluated, along four different axes, the intentionally harmful, accidentally harmful and neutral actions performed by others. Our main results demonstrate a consistent relation between SoA and morality judgments, since we found two significant correlations between the way we judge the moral permissibility of Accidental Harm scenarios and SoA (both its explicit and implicit components). Specifically, the greater the SoA, the stricter the moral judgements of agents in these scenarios. In addition, and only for explicit SoA, we observed a correlation with Malicious Intentionality ratings of the same scenarios: the stronger the SoA, the higher the attribution of malicious intentionality to the agent in Accidental scenarios. This pattern was specific to Accidental Harm as it was not found with the other two scenarios.

Concerning Intentional Binding, we replicated the well-established underestimation effect (Poonian and Cunnington, 2013; Imaizumi and Tanno, 2019), wherein people tend to show greater underestimation of the temporal interval between their action and its associated consequence, compared to control conditions. We confirmed this result by showing greater temporal compression of the action-outcome interval in the condition where participant’s own actions produced the outcomes, compared to the passing viewing condition. Our data suggest that Intentional Binding is comparable across temporal delays; we did not find a significant interaction between the Time Interval and Condition factors. This result is perfectly in line with previous studies adopting the same experimental paradigm (Imaizumi and Tanno, 2019; for different results see Zapparoli et al., 2020).

Regarding explicit SoA, we observed, instead, that it decreased at increasing time intervals: the shorter the delay between the action and the outcome, the stronger the explicit SoA (in line with Poonian and Cunnington, 2013; Wen et al., 2015; Imaizumi and Tanno, 2019). Importantly, the intentional binding index also correlated with the averaged explicit agency ratings, with shorter temporal estimations associated with higher explicit SoA (as reported also in Imaizumi and Tanno, 2019; Galang et al., 2021; for different results see Moore et al., 2012; Dewey and Knoblich, 2014; Saito et al., 2015; Wen et al., 2015; Ma et al., 2021). In their multifactorial two-step model, Synofzik et al. (2008) argued that bottom-up mechanisms account for the feeling of agency (implicit SoA), which is further processed through top-down mechanisms to form an explicit attribution of agency (explicit SoA). Following this line of thought, it seems that, in our case, the bottom-up processing of perceptual output and the top-down influence of cognitive factors showed similar patterns of effects.

Concerning the Moral Cognition task, we found that participants relied on both agent intention and action outcome when judging the morality of the scenarios (Young and Saxe, 2009; Baez et al., 2014; Gan et al., 2016). Indeed, intentional pain scenarios were judged more severely than accidental pain scenarios; moreover, despite sharing the same neutral intention, accidental harm scenarios were considered less morally permissible than neutral scenarios. These effects likely emerged due to the different outcomes of the two types of actions: victim’s suffering in the former, but not in the latter. Dual-system approaches explain the fundamental properties of human moral judgments by contrasting “emotional” and “rational/cognitive” processes (Cushman, 2013). Emotional processes are involved in the affective reaction to the injury experienced by the victim (action outcome), while the cognitive processes are involved in inferring the agent’s mental state (agent intention). Because Accidental Harm and Neutral actions share the same neutral intention (cognitive processes), the moral permissibility of the scenarios mostly depends on emotional processes, which are recruited when watching someone’s suffering (Accidental Harm) compared to a situation where there is no harm for anybody (Neutral scenarios).

Crucially, when comparing the results of the Intentional Binding and Moral Cognition tasks, we found two significant correlations – also strongly supported by Bayesian analyses – between implicit and explicit SoA indices and moral judgments of Accidental Harm scenarios. Specifically, the stronger the SoA – meaning a greater temporal underestimation effect for Action trials on the implicit intentional binding measure – the stricter the moral judgment. Moreover, for the explicit measure of agency only, and in line with our previous finding, we also observed a significant relation with the intentionality rating of the same scenarios: the stronger the agency, the more malicious the intentionality was evaluated to be. This self-other overlap is reminiscent of the “simulation theory” proposals (Carruthers, 1996; Waytz and Mitchell, 2011), which posit that perceivers can use knowledge about themselves to infer the mental states of others. In this vein, feeling more responsible for our own actions would predict a similar tendency when judging the actions of others.

It is not surprising that all these correlations emerged only with Accidental Harm scenarios. Indeed, as recently proposed by Hirschfeld-Kroen et al. (2021), Accidental Harm scenarios are an important context for broadening our understanding of the relation between agency and moral judgments thanks to the incongruence between the agent’s intention, which is neutral, and the action’s outcome, which is instead negative (Young and Saxe, 2009). Indeed, for judging these scenarios, people must inhibit a salient and prepotent representation (i.e., the observed outcome for the victim) and assign more weight to a less immediate alternative (i.e., the intention of the agent; Young and Saxe, 2009; Margoni and Surian, 2016, 2021; Baez et al., 2018; Schwartz et al., 2021). Thus, the pattern of correlations that we observed may suggest that people with stronger SoA fail to inhibit outcome/victim-based emotional processes. Future research is needed to understand whether stronger SoA is associated with a more general reduced ability to inhibit irrelevant information. Moreover, future studies over larger samples are needed to replicate the results of the present exploratory study.

Our results appear to agree with another study investigating the relation between SoA and morality: Moretto et al. (2011) found enhanced SoA in moral compared to non-moral contexts. Indeed, participants exhibited stronger picture-effects binding when the picture represented a moral dilemma compared with a purely economic dilemma. This suggests that agency is perceived differently in morally salient events relative to other events. Furthermore, our data are also in line with those reported by Lepron et al. (2015), where being held responsible for third-person pain caused an enhanced empathic response, suggesting that SoA may play a role in regulating empathy and, consequently, moral conduct. Importantly, the results of our research - the more you master your own actions, the more rigorous your moral judgment is - add to the previous literature suggesting that our sense of responsibility for our own actions goes hand in hand with our moral rigor for others’ actions. We could conclude by saying that a “responsible brain” is also a “moral brain.”

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Local ethical committee of the University of Bologna (Protocol Number 284744). Written informed consent was not provided because of the online data collection procedure where inform consent was inherent to the participation to the experiment.

CS was responsible for the conceptualization, experimental design, data analyses, and manuscript writing for this project. KC was responsible for the experimental design, data collection and analysis, and manuscript writing. SP was responsible for the experimental design, data collection, and manuscript revision. AA was responsible for the conceptualization, experimental design, resource acquisition, and manuscript revision. All authors contributed to the article and approved the submitted version.

This work was financially supported by research grants from the Bial Foundation (347/18 and 304/2022), Fondazione del Monte di Bologna e Ravenna (1402bis/2021), and the Ministero dell’Istruzione, dell’Università e della Ricerca (2017N7WCLP) awarded to AA.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.