Sense of agency in joint action: a critical review of we-agency.

The sense of agency refers to the experience of control over voluntary actions and their effects. There is growing interest in the notion of we-agency, whereby individual sense of agency is supplanted by a collective agentic experience. The existence of this unique agentic state would have profound implications for human responsibility, and, as such, warrants further scrutiny. In this paper, we review the concept of we-agency and examine whether evidence supports it. We argue that this concept entails multiplying hypothetical agentic states associated with joint action, thus ending up with an entangled phenomenology that appears somewhat speculative when weighted against the available evidence. In light of this, we suggest that the concept of we-agency should be abandoned in favor of a more parsimonious framework for the sense of agency in joint action.

The reliability of assistance systems modulates the sense of control and acceptability of human operators.

Individuals are increasingly required to interact with complex and autonomous technologies, which often has a significant impact on the control they experience over their actions and choices. A better characterization of the factors responsible for modulating the control experience of human operators is therefore a major challenge to improve the quality of human-system interactions. Using a decision-making task performed in interaction with an automated system, we investigated the influence of two key properties of automated systems, their reliability and explicability, on participants' sense of agency (SoA), as well as the perceived acceptability of system choices. The results show an increase in SoA associated with the most explicable system. Importantly, the increase in system explicability influenced participants' ability to regulate the control resources they engaged in the current decision. In particular, we observed that participants' SoA varied with system reliability in the "explained" condition, whereas no variation was observed in the "non-explained" condition. Finally, we found that system reliability had a direct impact on system acceptability, such that the most reliable systems were also considered the most acceptable systems. These results highlight the importance of studying agency in human-computer interaction in order to define more acceptable automation technologies.

Active self-motion control and the role of agency under ambiguity.

Purpose: Self-motion perception is a key factor in daily behaviours such as driving a car or piloting an aircraft. It is mainly based on visuo-vestibular integration, whose weighting mechanisms are modulated by the reliability properties of sensory inputs. Recently, it has been shown that the internal state of the operator can also modulate multisensory integration and may sharpen the representation of relevant inputs. In line with the concept of agency, it thus appears relevant to evaluate the impact of being in control of our own action on self-motion perception. Methodology: Here, we tested two conditions of motion control (active/manual trigger versus passive/ observer condition), asking participants to discriminate between two consecutive longitudinal movements by identifying the larger displacement (displacement of higher intensity). We also tested motion discrimination under two levels of ambiguity by applying acceleration ratios that differed from our two "standard" displacements (i.e., 3 s; 0.012 m.s-2 and 0.030 m.s-2). Results: We found an effect of control condition, but not of the level of ambiguity on the way participants perceived the standard displacement, i.e., perceptual bias (Point of Subjective Equality; PSE). Also, we found a significant effect of interaction between the active condition and the level of ambiguity on the ability to discriminate between displacements, i.e., sensitivity (Just Noticeable Difference; JND). Originality: Being in control of our own motion through a manual intentional trigger of self-displacement maintains overall motion sensitivity when ambiguity increases.

Being active over one's own motion: Considering predictive mechanisms in self-motion perception.

Self-motion perception is a key element guiding pilots' behavior. Its importance is mostly revealed when impaired, leading in most cases to spatial disorientation which is still today a major factor of accidents occurrence. Self-motion perception is known as mainly based on visuo-vestibular integration and can be modulated by the physical properties of the environment with which humans interact. For instance, several studies have shown that the respective weight of visual and vestibular information depends on their reliability. More recently, it has been suggested that the internal state of an operator can also modulate multisensory integration. Interestingly, the systems' automation can interfere with this internal state through the loss of the intentional nature of movements (i.e., loss of agency) and the modulation of associated predictive mechanisms. In this context, one of the new challenges is to better understand the relationship between automation and self-motion perception. The present review explains how linking the concepts of agency and self-motion is a first approach to address this issue.

Modulations of one's sense of agency during human-machine interactions: A behavioural study using a full humanoid robot.

Although previous investigations reported a reduced sense of agency when individuals act with traditional machines, little is known about the mechanisms underpinning interactions with human-like automata. The aim of this study was twofold: (1) to investigate the effect of the machine's physical appearance on the individuals' sense of agency and (2) to explore the cognitive mechanisms underlying the individuals' sense of agency when they are engaged in a joint task. Twenty-eight participants performed a joint Simon task together with another human or an automated artificial system as a co-agent. The physical appearance of the automated artificial system was manipulated so that participants could cooperate either with a servomotor or a full humanoid robot during the joint task. Both participants' response times and temporal estimations of action-output delays (i.e., an implicit measure of agency) were collected. Results showed that participants' sense of agency for self- and other-generated actions sharply declined during interactions with the servomotor compared with the human-human interactions. Interestingly, participants' sense of agency for self- and other-generated actions was reinforced when participants interacted with the humanoid robot compared with the servomotor. These results are discussed further.

Brain mechanisms of automated conflict avoidance simulator supervision.

Supervision of automated systems is an ubiquitous aspect of most of our everyday life activities which is even more necessary in high risk industries (aeronautics, power plants, etc.). Performance monitoring related to our own error making has been widely studied. Here we propose to assess the neurofunctional correlates of system error detection. We used an aviation-based conflict avoidance simulator with a 40% error-rate and recorded the electroencephalographic activity of participants while they were supervising it. Neural dynamics related to the supervision of system's correct and erroneous responses were assessed in the time and time-frequency domains to address the dynamics of the error detection process in this environment. Two levels of perceptual difficulty were introduced to assess their effect on system's error detection-related evoked activity. Using a robust cluster-based permutation test, we observed a lower widespread evoked activity in the time domain for errors compared to correct responses detection, as well as a higher theta-band activity in the time-frequency domain dissociating the detection of erroneous from that of correct system responses. We also showed a significant effect of difficulty on time-domain evoked activity, and of the phase of the experiment on spectral activity: a decrease in early theta and alpha at the end of the experiment, as well as interaction effects in theta and alpha frequency bands. These results improve our understanding of the brain dynamics of performance monitoring activity in closer-to-real-life settings and are a promising avenue for the detection of error-related components in ecological and dynamic tasks.

What is new with Artificial Intelligence? Human-agent interactions through the lens of social agency.

In this article, we suggest that the study of social interactions and the development of a "sense of agency" in joint action can help determine the content of relevant explanations to be implemented in artificial systems to make them "explainable." The introduction of automated systems, and more broadly of Artificial Intelligence (AI), into many domains has profoundly changed the nature of human activity, as well as the subjective experience that agents have of their own actions and their consequences - an experience that is commonly referred to as sense of agency. We propose to examine the empirical evidence supporting this impact of automation on individuals' sense of agency, and hence on measures as diverse as operator performance, system explicability and acceptability. Because of some of its key characteristics, AI occupies a special status in the artificial systems landscape. We suggest that this status prompts us to reconsider human-AI interactions in the light of human-human relations. We approach the study of joint actions in human social interactions to deduce what key features are necessary for the development of a reliable sense of agency in a social context and suggest that such framework can help define what constitutes a good explanation. Finally, we propose possible directions to improve human-AI interactions and, in particular, to restore the sense of agency of human operators, improve their confidence in the decisions made by artificial agents, and increase the acceptability of such agents.

Leveraging human agency to improve confidence and acceptability in human-machine interactions.

Repeated interactions with automated systems are known to affect how agents experience their own actions and choices. The present study explores the possibility of partially restoring sense of agency in operators interacting with automated systems by providing additional information about the system's decision, i.e. its confidence. To do so, we implemented an obstacle avoidance task with different levels of automation and explicability. Levels of automation were varied by implementing conditions in which the participant was free or not free to choose which direction to take, whereas levels of explicability were varied by providing or not providing the participant with the system's confidence in the direction to take. We first assessed how automation and explicability interacted with participants' sense of agency, and then tested whether increased self-agency was systematically associated with greater confidence in the decision and improved system acceptability. The results showed an overall positive effect of system assistance. Providing additional information about the system's decision (explicability effect) and reducing the cognitive load associated with the decision itself (automation effect) was associated with stronger sense of agency, greater confidence in the decision, and better performance. In addition to the positive effects of system assistance, acceptability scores revealed that participants perceived "explicable" systems more favorably. These results highlight the potential value of studying self-agency in human-machine interaction as a guideline for making automation technologies more acceptable and, ultimately, improving the usefulness of these technologies.

Multitasking costs on metacognition in a triple-task paradigm.

Multitasking situations, such as using one's phone while driving, are increasingly common in everyday life. Experimental psychology has long documented the costs of multitasking on task performance; however, little is known of the effects it has on the metacognitive processes that monitor such performance. The present study is a step toward filling this void by combining psychophysical procedures with complex multitasking. We devised a multimodal paradigm in which participants performed a sensorimotor tracking task, a visual discrimination task, and an auditory 2-back working memory task, either separately or concurrently, while also evaluating their task performance every ~15 s. Our main finding is that multitasking decreased participants' awareness of their performance (metacognitive sensitivity) for all three tasks. Importantly, this result was independent of the multitasking cost on task performance, and could not be attributed to confidence leak, psychological refractory period, or recency effects on self-evaluations. We discuss the implications of this finding for both metacognition and multitasking research.

Mind Wandering Influences EEG Signal in Complex Multimodal Environments.

The phenomenon of mind wandering (MW), as a family of experiences related to internally directed cognition, heavily influences vigilance evolution. In particular, humans in teleoperations monitoring partially automated fleet before assuming manual control whenever necessary may see their attention drift due to internal sources; as such, it could play an important role in the emergence of out-of-the-loop (OOTL) situations and associated performance problems. To follow, quantify, and mitigate this phenomenon, electroencephalogram (EEG) systems already demonstrated robust results. As MW creates an attentional decoupling, both ERPs and brain oscillations are impacted. However, the factors influencing these markers in complex environments are still not fully understood. In this paper, we specifically addressed the possibility of gradual emergence of attentional decoupling and the differences created by the sensory modality used to convey targets. Eighteen participants were asked to (1) supervise an automated drone performing an obstacle avoidance task (visual task) and (2) respond to infrequent beeps as fast as possible (auditory task). We measured event-related potentials and alpha waves through EEG. We also added a 40-Hz amplitude modulated brown noise to evoke steady-state auditory response (ASSR). Reported MW episodes were categorized between task-related and task-unrelated episodes. We found that N1 ERP component elicited by beeps had lower amplitude during task-unrelated MW, whereas P3 component had higher amplitude during task-related MW, compared with other attentional states. Focusing on parieto-occipital regions, alpha-wave activity was higher during task-unrelated MW compared with others. These results support the decoupling hypothesis for task-unrelated MW but not task-related MW, highlighting possible variations in the "depth" of decoupling depending on MW episodes. Finally, we found no influence of attentional states on ASSR amplitude. We discuss possible reasons explaining why. Results underline both the ability of EEG to track and study MW in laboratory tasks mimicking ecological environments, as well as the complex influence of perceptual decoupling on operators' behavior and, in particular, EEG measures.

Resilience of perceptual metacognition in a dual-task paradigm.

When people do multiple tasks at the same time, it is often found that their performance is worse relative to when they do those same tasks in isolation. However, one aspect that has received little empirical attention is whether the ability to monitor and evaluate one's task performance is also affected by multitasking. How does dual-tasking affect metacognition and its relation to performance? We investigated this question through the use of a visual dual-task paradigm with confidence judgments. Participants categorized both the color and the motion direction of moving dots, and then rated their confidence in both responses. Across four experiments, participants (N = 87) exhibited a clear dual-task cost at the perceptual level, but no cost at the metacognitive level. We discuss this resilience of metacognition to multitasking costs, and examine how our results fit onto current models of perceptual metacognition.

An Electromyographic Analysis of the Effects of Cognitive Fatigue on Online and Anticipatory Action Control.

Cognitive fatigue is a problem for the safety of critical systems (e.g., aircraft) as it can lead to accidents, especially during unexpected events. In order to determine the extent to which it disrupts adaptive capabilities, we evaluated its effect on online and anticipatory control. Despite numerous studies conducted to determine its effects, the exact mechanism(s) affected by fatigue remains to be clarified. In this study, we used distribution and electromyographic analysis to assess whether cognitive fatigue increases the capture of the incorrect automatic response or if it impairs its suppression (online control), and whether the conflict adaptation effect is reduced (anticipatory control). To this end, we evaluated the evolution of the performance over time during the Simon task, a classic conflict task that elicits incorrect automatic responses. To accentuate the presence of fatigue during the Simon task, two groups previously performed a dual-task with two different cognitive load levels to create two different levels of fatigue. The results revealed that time on task impaired online control by disrupting the capacity to suppress the incorrect response but leaving unaffected the expression of the automatic response. Furthermore, participants emphasized speed rather than accuracy with time on task, with in addition more fast guesses, suggesting that they opted for a less effortful response strategy. As the implementation of the suppression mechanism requires cognitive effort, the conjunction of these results suggests that the deficits observed may be due to disengagement of effort over time rather than reflecting an incapacity to make an effort.

Brain-Computer Interface-Based Adaptive Automation to Prevent Out-Of-The-Loop Phenomenon in Air Traffic Controllers Dealing With Highly Automated Systems.

Increasing the level of automation in air traffic management is seen as a measure to increase the performance of the service to satisfy the predicted future demand. This is expected to result in new roles for the human operator: he will mainly monitor highly automated systems and seldom intervene. Therefore, air traffic controllers (ATCos) would often work in a supervisory or control mode rather than in a direct operating mode. However, it has been demonstrated how human operators in such a role are affected by human performance issues, known as Out-Of-The-Loop (OOTL) phenomenon, consisting in lack of attention, loss of situational awareness and de-skilling. A countermeasure to this phenomenon has been identified in the adaptive automation (AA), i.e., a system able to allocate the operative tasks to the machine or to the operator depending on their needs. In this context, psychophysiological measures have been highlighted as powerful tool to provide a reliable, unobtrusive and real-time assessment of the ATCo's mental state to be used as control logic for AA-based systems. In this paper, it is presented the so-called "Vigilance and Attention Controller", a system based on electroencephalography (EEG) and eye-tracking (ET) techniques, aimed to assess in real time the vigilance level of an ATCo dealing with a highly automated human-machine interface and to use this measure to adapt the level of automation of the interface itself. The system has been tested on 14 professional ATCos performing two highly realistic scenarios, one with the system disabled and one with the system enabled. The results confirmed that (i) long high automated tasks induce vigilance decreasing and OOTL-related phenomena; (ii) EEG measures are sensitive to these kinds of mental impairments; and (iii) AA was able to counteract this negative effect by keeping the ATCo more involved within the operative task. The results were confirmed by EEG and ET measures as well as by performance and subjective ones, providing a clear example of potential applications and related benefits of AA.

The sense of agency in human-human vs human-robot joint action.

Kinesthesis pertains to the perception of moving body parts, while the sense of agency refers to the experience of controlling one's action-effects. Based on previous work, we hypothesized that the sense of agency would decrease in joint action with a robot compared to a human partner. Pairs of participants were jointly manipulating two interconnected haptic devices enabling them to feel each other's forces. Unbeknown to participants, their partner was sometimes replaced by a robot. The sense of agency was assessed using intentional binding, which refers to a contraction of perceived time between an action and its effect for intentional actions, and participants' judgment of their contribution to joint action. Participants judged their contribution as higher when they were initiating action and when they were paired with the robot. By contrast, intentional binding occurred only with a human partner. This outcome supports the hypothesis that human-robot joint action hinders intentional binding.

Evaluation of performance monitoring ERPs through difficulty manipulation in a response-feedback paradigm.

Performance monitoring is an amply studied function, since it is of major importance in carrying out actions in our everyday life. No consensus has been reached on the functional role and the relationship between each event-related potential (ERP) characterizing this function. In this study, we used a modified version of the flanker task, measuring the impact of task difficulty on the amplitudes of response-locked and feedback-locked performance monitoring ERPs in a single trial. We observed a functional differentiation between fronto-central (ERN/CRN and FRN) and centro-parietal (Pe/Pc and P300) components: the former seem to be only sensitive to accuracy, whereas the latter seem to be mainly modulated by task difficulty. The use of a surface Laplacian transformation, estimating current source density, on our data also supported an effect of difficulty on centro-parietal response-locked and feedback-locked ERPs. This technique allowed the spatial resolution to be improved and provided clarity, associated with the difficulty manipulation, on the activity of response-locked and feedback-locked performance monitoring ERPs.

Human or not human? Performance monitoring ERPs during human agent and machine supervision.

Performance monitoring is a critical process which allows us to both learn from our own errors, and also interact with other human beings. However, our increasingly automated world requires us to interact more and more with automated systems, especially in risky environments. The present EEG study aimed at investigating and comparing the neuro-functional correlates associated with performance monitoring of an automated system and a human agent using a vertically-oriented arrowhead version of the flanker task. Given the influence of task difficulty on performance monitoring, two levels of difficulty were considered in order to assess their impact on supervision activity. A large N2-P3 complex in fronto-central regions was observed for both human agent error detection and system error detection during supervision. Using a cluster-based permutation analysis, a significantly decreased P3-like component was found for system compared to human agent error detection. This variation is in line with various psychosocial behavioral studies showing a difference between human-human and human-machine interactions, even though it was not clearly anticipated. Finally, the activity observed during error detection was significantly reduced in the difficult condition compared to the easy one, for both system and human agent supervision. Overall, this study is a first step towards the characterization of the neurophysiological correlates underlying system supervision, and a better understanding of their evolution in more complex environments. To go further, these results need to be replicated in other experiments with various paradigms to assess the robustness of the pattern and decrease during system supervision.

Action co-representation and the sense of agency during a joint Simon task: Comparing human and machine co-agents.

Recent studies have suggested that individuals are not able to develop a sense of joint agency during joint actions with artificial systems. We sought to examine whether this lack of joint agency is linked to individuals' inability to co-represent the machine-generated actions. Fifteen participants observed or performed a Simon response time task either individually, or jointly with another human or a computer. Participants reported the time interval between their response (or the co-actor response) and a subsequent auditory stimulus, which served as an implicit measure of participants' sense of agency. Participants' reaction times showed a classical Simon effect when they were partnered with another human, but not when they collaborated with a computer. Furthermore, participants showed a vicarious sense of agency when co-acting with another human agent but not with a computer. This absence of vicarious sense of agency during human-computer interactions and the relation with action co-representation are discussed.

Influence of automation on mind wandering frequency in sustained attention.

There is accumulating evidence which shows that mind wandering may be increased within automated environments. This is particularly concerning when considering the negative effect of mind wandering on short-term performance. Seventeen participants performed an obstacle avoidance task under two conditions, manual and automated, each lasting 40 min. Subjects perceived the manual condition as more demanding than the automated one. We noted a significant increase of mind wandering frequency after only approximately 20 min under the automated condition. While learning and workload alone cannot explain these results, more automation-related phenomena, such as complacency or loss of agency, could play a role. Pupil diameter decreased during mind wandering compared to focus periods, revealing a decoupling from the task. The decrease remained stable in amplitude across different times and conditions. Research on mind wandering could be used to characterize an operator's state of mind regarding issues related to system interactions.

Out of the Loop, in Your Bubble: Mind Wandering Is Independent From Automation Reliability, but Influences Task Engagement.

This study examined the influence of automation reliability on task-unrelated mind wandering (MW) frequency and the impact of MW on task engagement. Automated environment features make it particularly prone to increase MW frequency. Through mechanisms like complacency or agency, automating a task could increase MW frequency for the operator. For safety-critical industries, the lower perception and degraded stimuli processing associated with MW, summarized by the term "decoupling hypothesis," are particularly concerning. Sixteen participants supervised an autopilot avoiding obstacles with two levels of reliability. Each condition lasted 45 min. We recorded thoughts as either pertaining to being focused, task-related MW or task-unrelated MW. We also recorded perceived mental demand, trust regarding the autopilot and oculometric measures. Based on questionnaire results, our protocol succeeded in inducing more mental demand and lower trust when the automation was unreliable. Attentional states were not correlated, nor did it influence trust in the system reliability. On the contrary, mental demand ratings and pupil diameter were lower during both task-related and task-unrelated MW, compared to those during the focus attentional state. This shows that perceptual decoupling also affects the engagement of operators in automated environments, which may dramatically lower their ability to supervise automation efficiently. This research informs human-automation designers to consider operator engagement when creating automated systems.

Agency modulates interactions with automation technologies.

The increasing presence of automation between operators and automated systems tends to disrupt operators from action outcomes, leading them to leave the control loop. The theoretical framework of agency suggests that priming the operator about the system's upcoming behaviour could help restore an appropriate sense of control and increase user acceptance of what the system is doing. In a series of two experiments, we test whether providing information about what the system is about to do next leads to an increase in the level of user acceptance, concomitant with an increase in control and performance. Using an aircraft supervision task, we demonstrated the benefit of prime messages regarding system acceptance and performance. Taken together, our results indicate that the principles proposed by this framework could be used to improve human-machine interaction and maintain a high level of sense of control in supervisory tasks. Practitioner Summary: The out-of-the-loop performance problem is a major potential consequence of automation, leaving operators helpless to takeover automation in case of failure. Using an aircraft supervision task, the following article illustrates how the psychological approach of agency can help improving human-system interactions by designing more acceptable and more controllable automated interfaces.