Sense of Agency and Other Mental Functions

In document Prospective Sense of Agency: Cognitive and Neural Mechanisms (Page 36-41)

Chapter 1. General Introduction

1.3. Sense of Agency and Other Mental Functions

In the above literature review on SoA we saw that much has been learned about how we link actions to outcomes, and self-attribute agency over those outcomes. However, the role of executive functions such as planning, decision making and cognitive control have been largely neglected, even though these are typically considered critical for the exercise of our agentic capacities (Schiffer, Waszak, & Yeung, 2015). The field of metacognition, in particular in relation to decision making, has been concerned with understanding how we monitor our cognitive processes in order to flexibly adapt our cognition and behaviour (Yeung & Summerfield, 2012). Recent work on prospective SoA highlights the need to bridge these two fields for a more complete understanding of human agency. Furthermore, the literature discussed above has also largely neglected how social context may influence SoA, and behaviour. The present section will briefly review these two topics, which are implicated in the research conducted for this thesis.

1.3.1. Metacognition

Metacognition can be simply defined as involving a meta-level cognitive process which is about an object-level cognitive process (Nelson & Narrens, 1990). It has been suggested that there are two interrelated levels of metacognition: a) a lower- level involved in the monitoring and control of cognitive processes; and b) a higher- level meta-representational level that interprets behaviour based on beliefs and theories (Arango-Muñoz, 2010; Koriat, 2000). “Epistemic feelings” may arise from monitoring processes and can be used to adjust behaviour online (Proust, 2008).

However, while both levels may be associated with explicit, conscious metacognition, monitoring and control processes may remain implicit (Fleming & Dolan, 2014). Conflict Monitoring

Monitoring action selection allows the recruitment of cognitive control processes when needed, in order to adjust subsequent behaviour. For example, when a target stimulus appears surrounded by incongruent distractors, which activate a competing response alternatives, this response conflict leads to a slowing of RTs (e.g. Eriksen & Eriksen, 1974). Yet, this congruency effect is reduced in trials following response conflict, relative to following easy, congruent trials (Gratton, Coles, & Donchin, 1992; see Egner, 2007 for an overview). That is, detection of response conflict triggered behavioural adaptation, in order to improve performance. Although debate is on- going, some studies have shown that conflict adaptation can also occur for unconsciously triggered response conflict (Atas, Desender, Gevers, & Cleeremans, 2015; van Gaal, Lamme, & Ridderinkhof, 2010, but see Desender, Van Lierde, & Van den Bussche, 2013). Interestingly, a recent study, using unconsciously triggered conflict, suggested that a conscious experience of conflict, or difficulty in action selection, was necessary for conflict adaptation (Desender, Opstal, & Bussche, 2014).

Neural markers of conflict.

At a neural level, the anterior cingulate cortex (ACC) is thought to be involved in conflict monitoring, and to trigger cognitive control functions associated with more frontal regions, such as the dlPFC (Botvinick & Braver, 2015; Holroyd & Yeung, 2012). Moreover, ACC-mediated conflict monitoring can also be identified in event related potentials (ERPs; for a review, see Larson, Clayson, & Clawson, 2014). In ERPs locked to target onset, response conflict leads to a large N2 component, a negative potentials peaking around 250-300ms after the target, at fronto-central sites (Kopp, Rist, & Mattler, 1996). This N2 component is thought to index conflict detection and resolution (Larson et al., 2014).

Error monitoring.

Conflict monitoring theory has also been used to explain error monitoring and detection (Yeung, Botvinick, & Cohen, 2004). As accumulation of evidence for the correct response may continue after selecting a given action, post-decisional processing can reveal that the chosen response was incorrect, resulting in a conflict between the correct and the executed response (Yeung & Summerfield, 2012). In ERPs locked to the action, an error-related negativity (ERN) component emerges immediately after error commission (0-100ms), in comparison to correct responses (see Larson et al., 2014 for a review). In correct trials, a correct-related negativity (CRN) has been associated with task difficulty/uncertainty (Pailing & Segalowitz, 2004; Scheffers & Coles, 2000). Errors and the ERN have also been associated with ACC activity (Carter et al., 1998; Charles, Van Opstal, Marti, & Dehaene, 2013). Therefore, it has been suggested that the target-locked N2 and the action-locked ERN components reflect pre- and post-decisional conflict monitoring, linked to the ACC (Larson et al., 2014; Yeung & Summerfield, 2012). Confidence

In contrast to research on error monitoring, models of confidence judgements have typically emphasised a role for pre-decisional processing (Yeung & Summerfield, 2012). Confidence is related to the speed of accumulation of evidence, as well as to the balance of evidence between response alternatives at the time of action (Kiani, Corthell, & Shadlen, 2014). However, post-decisional processing also influences confidence (Boldt & Yeung, 2015; Resulaj, Kiani, Wolpert, & Shadlen, 2009; Scheffers & Coles, 2000), showing that confidence judgements and error monitoring can both draw on the same post-decisional processing. For example, a study (Boldt & Yeung, 2015) varied the difficulty of a perceptual discrimination task, and obtained judgements on a six-point scale, ranging from certainly correct, to maybe correct, to certainly wrong. Results showed that increasing confidence in having made an error was associated with more negative CRN/ERN amplitude, as well a more positive Pe amplitude (another error-related ERP component found around 300ms after action).

Moreover, models of confidence judgements have often neglected the role of action and the motor system, where evidence accumulation can occur in parallel to perceptual processing (Cisek, 2007). A recent study showed that TMS stimulation of the premotor cortex associated with the unchosen response disrupted metacognitive accuracy in perceptual confidence judgements (Fleming et al., 2014). Moreover, these effects were found for stimulation both before and after the action. Therefore, confidence judgements rely on late-stage metacognitive processes, which are influenced by action-specific signals. These findings are particularly relevant to the present thesis, as they are consistent with the aforementioned influence of fluency in action selection to the sense of agency, which has also been linked with post- decisional processing (Chambon, Moore, et al., 2014). Fluency

Finally, the research considered above is also relevant to the widely studied effects of fluency on a variety of judgements, such as confidence, liking or familiarity (Alter & Oppenheimer, 2009). These effects have been associated with fluency at many levels of processing, such as perceptual, cognitive, linguistic, or memory-based. The experience, of feeling, of fluency is thought to be based on a qualitative signal about information processing, and can be broadly defined as a continuum, ranging from fluent or effortless, to dysfluent or effortful processing. Note that response conflict can lead to a conscious experience of dysfluency (Desender et al., 2014; Morsella et al., 2009). These experiences, or feelings are often vague, especially about their sources (Winkielman, Ziembowicz, & Nowak, 2015). Therefore, fluency/dysfluency experiences can “leak” into judgements, even if they may not be a relevant cue. It has also been argued that conflict (or dysfluency) may serve as an aversive signal (Botvinick, 2007). Consistently response conflict can lead to more negative affective judgements of subsequent neutral stimuli (Fritz & Dreisbach, 2013). Importantly, when fluency experiences can be attributed to an irrelevant source, they no longer influence judgements (cf. Alter & Oppenheimer, 2009). Therefore, while fluency may often be a useful heuristic cue (Whittlesea & Leboe, 2003), it will not be taken into account if considered uninformative.

The work on metacognition reviewed in this section offers some insight into how action selection fluency influences the SoA. The work on conflict monitoring suggests that neural signals associated with response conflict detection and resolution could form the basis of the cue to SoA. Relatedly, the work on confidence suggests that similar, post-decisional, metacognitive processes may inform both confidence and agency judgements. Moreover, fluency effects on SoA could be linked to a general heuristic that would enhance SoA, or due to the affective consequences of response conflict.

1.3.2. Social Aspects of Agency

The experience of agency can be especially relevant in social contexts, as the presence of other agents can increase ambiguity in agency attribution. Moreover, concepts of personal and moral responsibility become more salient. These contextual effects may thus have important consequences to our SoA, as well as to our behaviour. In fact, the influence of social context on behaviour is well documented in the social psychology literature. For example, in emergency scenarios, the likelihood of someone helping decreases with the number of bystanders (Darley & Latane, 1968); when working in a group, people put in less effort than if they were working alone (Karau & Williams, 1993). These effects are thought to result from a diffusion of responsibility, in which individuals feel the responsibility for action lies with others (Bandura, 1991). Yet, this could merely reflect a post-hoc justification, related to self-serving biases, such as maintaining a positive self-image, rather than involving online changes in SoA.

Interestingly, a recent study has shown that being coerced into giving electric shocks to others leads to a reduction in intentional binding, and an attenuation of outcome processing, relative to a free choice condition (Caspar, Christensen, Cleeremans, & Haggard, 2016). This shows that social context can influence low-level, implicit aspects of SoA, in a condition in which social desirability and self-serving biases were thought to underlie a reported reduction in responsibility. Under coercion, responsibility is displaced onto another agent (Bandura, 1991), thus agency

attribution may become more ambiguous. This is clearly also the case in a group work scenario. However, the help of one person may suffice in emergency situations, therefore, changes in decision making processes may be more relevant than attribution ambiguity. Social situations are associated with more complex decision making, and uncertainty, since the potential behaviour of other agents needs to be considered. Consequently, in line with the fluency effects mentioned above, it could be hypothesised that an increased difficulty in decision making could result in a reduction in SoA, and in the motivation to act.

In document Prospective Sense of Agency: Cognitive and Neural Mechanisms (Page 36-41)