Social constraints from an observer’s perspective: Coordinated actions make an agent’s position more predictable
Introduction
The ability to predict and anticipate the actions of others is crucial for planning appropriate behaviors before engaging or intervening in observed actions (Csibra and Gergely, 2007, Hauser and Wood, 2010). Evidently, people are able to generate predictions even with limited information about agents’ actions (Csibra et al., 2003, Saunier et al., 2013). This sophisticated ability is thought to be sensitive to an individual’s goals (or just the associative contingency between actions and outcomes) (Gergely and Csibra, 2003, Verschoor et al., 2013). However, people’s actions are not always framed as pursuing their own individual goals. Instead, they are often embedded in coordinated interactions to achieve a collective/shared goal, which are referred to coordinated or joint actions (Knoblich and Sebanz, 2008, Sebanz et al., 2006). Less is known about whether this interpersonal coordination information, beyond individual goals, could influence action prediction.
Recently, some researchers have started to explore how the information conveyed by coordinated actions affects action processing/understanding. Observing coordinated interactions has been consistently found to have a tangible benefit for extracting information from actions or at least in detecting the actors. For instance, Neri et al. (2006) found the visual discrimination of a human agent is influenced by the second agent when their actions involving physical contact could be interpreted as meaningful coordination (i.e., fighting or dancing); Manera et al. (2011) confirmed this conclusion, showing that communicative gestures, even without contact, can increase the likelihood of perceiving a second agent. The above efforts notwithstanding, no direct evidence has illustrated the role of observed coordinated interaction in action prediction.
Interpersonal coordination is not only a mere summation of individual actions, but most importantly, it is also more than the individual elements, as its behaviors are interdependent and may sometimes be mutually complementary (Sebanz et al., 2006). Moreover, coordinated actions are thought to be constrained by each other within a coordinative structure (Shockley, Richardson, & Dale, 2009). For instance, Shockley, Santana, and Fowler (2003) found that mutual interpersonal postural constraints (i.e., sharing more locations in phase space) are involved during conversation in a coordinated manner. Although it is still under debate whether the emergence of this social constraint is supported by acting together with shared representation across persons or is due to spontaneous organization (Sebanz et al., 2006, Shockley et al., 2009), in any case, such a structured constraint ensures that the adjustment of one’s actions could result in predictively aligned changes of other people’ actions when the people interact in a coordinated fashion. Therefore, from the observer’s perspective, the actions of one agent could serve as efficient predictors for the actions of other agents in a coordinated interaction. Thus, when observing a coordinated interaction, even when an agent temporarily disappears, the vision could use the characteristic of mutual constraints in coordinated actions to reduce possible hypothesis spaces when inferring or predicting the actions of the invisible agent. In this case, an observer should generate much better predictions for the temporarily invisible actions that are involved in interpersonal coordination, in comparison with those that are not involved in interpersonal coordination. Namely, the observed coordination information should influence, and even enhance, the predictive accuracy for the expected actions.
To test our hypothesis, the manipulation of the coordination information used two types of dynamic chase scenes, in which two agents acted as chasers running in a coordinated or individual manner, toward a common prey.1 The different chase scenes were modelled after displays used by Heider and Simmel (1944) that presented geometric figures only in a chasing motion. One of the advantages of this method is that motion is the only action information that contributes to the understanding of semantic social meaning; thus, if we are interested in social information (e.g., chasing relation, coordination information), only the physical features of motion need to be controlled. Previous research has extensively used this type of chasing motion, but with only one chaser, to explore the perception of animacy, intention, and interaction (Dittrich and Lea, 1994, Gao et al., 2009). All of these studies documented that the motion sequences should not simply be treated as physical movement; they should be thought of meaningful actions with goals, which influences our other processes accordingly, such as visual searches and interactive behaviors (Gao et al., 2010, Meyerhoff et al., 2014). Usually, the chasing motion with one chaser and one target was generated with specific steering rules by referring to AI algorithms. With two chasers and one target of multi-agent chasing, the principles of movement have not been determined (Rawal, Rajagopalan, & Miikkulainen, 2010). Thus, it is better to rely on man-made trajectories than to use AI algorithms to describe multi-agent chasing. Indeed, our previous research successfully used the recorded motion of real people as they controlled their own avatars (chasers) in a coordinated or individual chase toward the same target (Yin et al., 2013). As well, the current research has adopted methods of using recordings of human motion to display coordinated and individual chases.
To examine the role of coordination information on action prediction, the recorded motions from both coordinated and individual chases were shown in a forward sequence to subjects who were required to predict the expected position (i.e., action) for one chaser after it became momentarily invisible. To further isolate the effect of socially coordinated information from possible low-level physical properties, we also established some intense paired controls for each type of chase, such as backward replay (Experiment 1), making the chasing target invisible (Experiment 2), and a direct manipulation of the goal-directedness of one chaser’s movements to disrupt coordination information (Experiment 3). If coordination information enhanced the prediction of actions, we should observe fewer prediction errors in coordinated chases compared with the controls, but not in individual chases.
Section snippets
Experiment 1a
Both the coordinated and individual chases were presented by forward replaying of recorded trajectories and were compared with their own controls, which consisted of backward replays of the same trajectories. In such settings, the physical features were the same in the two types of replay sequences, whereas in the backward replay, the chasers’ intended actions became inverted and ambiguous, disturbing the processing of social meaning behind them (though it never disappeared completely).
Experiment 1b
Although we kept the whole segments identical between forward and backward replay in Experiment 1a, the trajectories of the predicted agent were different when the agent’s motion was invisible. Therefore, the effects observed may still be explained by physical differences. In this experiment, the invisible trajectories of predicted agents were matched between backward and forward replays to determine whether the same results found in Experiment 1a would be observed.
Experiment 2
To ensure that the effects in Experiment 1 were indeed caused by the coordination information and not by the low-level motion features, in this experiment, only two chasers for both forward and backward replay were presented, and the prey was invisible. Because the chasers’ behaviors respond continuously to those of the prey, if the prey becomes invisible, it is difficult for the observer to determine the intent of the chasers, even though they are still moving in exactly the same way as when
Experiment 3
The above experiments clearly indicate that action prediction is modulated by coordination information, even though agents in both coordinated and individual chases are directed toward the target. Such modulation could be caused either by disruption from backward replay as the control condition or by enhancement from coordination information transmitted by the forward-replayed trajectories. To examine these possibilities, we introduced a direct intervention for coordination information in which
General discussion
The primary purpose of this study was to explore the role of coordination information beyond individual goals on action prediction. We found that the prediction error for an agent’s position decreased when coordinated chasers were presented. The nature of this discovered influence on action prediction should stem from the experienced social coordination behind the actions instead of from physical factors or individual actions in multi-agents chases, exhibiting an enhancement effect.
Author contributions
J. Yin, H. Xu, and M. Shen conceived and designed the experiments. J. Yin, H. Xu, X. Ding, J. Liang, and R. Shui performed the experiments and analyzed the data. J. Yin and M. Shen wrote the manuscript.
Acknowledgement
This research was supported by the National Natural Science Foundation of China (31571119, 31170975).
References (29)
- et al.
One-year-old infants use teleological representations of actions productively
Cognitive Science
(2003) - et al.
“Obsessed with goals”: Functions and mechanisms of teleological interpretation of actions in humans
Acta Psychologica
(2007) - et al.
The psychophysics of chasing: A case study in the perception of animacy
Cognitive Psychology
(2009) - et al.
Teleological reasoning in infancy: The naïve theory of rational action
Trends in Cognitive Sciences
(2003) Putting predators back into behavioral predator–prey interactions
Trends in Ecology & Evolution
(2002)- et al.
Electrophysiological correlates of biological motion permanence in humans
Behavioural Brain Research
(2013) - et al.
Joint action: Bodies and minds moving together
Trends in Cognitive Sciences
(2006) - et al.
Social grouping: Perceptual grouping of objects by cooperative but not competitive relationships in dynamic chase
Cognition
(2013) The psychophysics toolbox
Spatial Vision
(1997)- et al.
Visual perception of intentional motion
Perception
(1994)
The wolfpack effect: Perception of animacy irresistibly influences interactive behavior
Psychological Science
Evolving the capacity to understand actions, intentions, and goals
Annual Review of Psychology
An experimental study of apparent behavior
The American Journal of Psychology
The mirror-neuron system: A Bayesian perspective
NeuroReport
Cited by (10)
How you act matters: The impact of coordination on 4-year-old children's reasoning about diverse desires
2018, Journal of Experimental Child PsychologyCitation Excerpt :The inconsistency between these studies gives rise to the question of what decides the sometimes differential impacts of these two types of social interaction on children’s performance on mental state reasoning tasks. An implication for answering this question comes from studies on coordination between individuals (Michael, Sebanz, & Knoblich, 2016; Tomasello, Carpenter, Call, Behne, & Moll, 2005; Yin et al., 2016). Coordination refers to a condition where an individual’s goal-directed action plan is interdependent with and constrained by others’ action plans.
The influence of intention and outcome on evaluations of social interaction
2018, Acta PsychologicaCitation Excerpt :Specifically, observers generated much more accurate predictions for temporarily invisible actions that were part of interpersonal social interactions than for those that were not. Concerning these phenomena, it was suggested that information about individuals within an interaction is constructed as a unified representation, which constrains the hypothesis space when making inferences about involved individuals; thus, it leads to the enhanced detection and prediction of various elements of that social interaction (Yin et al., 2016). However, these studies focused on how a social interaction modulates other processes, rather than how social interactions are processed.
Cooperation turns preschoolers into flexible perspective takers
2019, Cognitive DevelopmentCitation Excerpt :Consistent with these findings (Jin et al., 2017), it could be inferred that such cooperation might also help children to understand others’ perspectives. However, it should be noted that there are many different forms of cooperation in real life (e.g., direct cooperation between partners in pas de deux, and indirect cooperation between workers on an assembly line), but the most typical cooperative interaction is characterized by joint goal and action coordination (Bratman, 1992; Tomasello, Carpenter, Call, Behne, & Moll, 2005), which is commonly conducted in some cooperation research (Jin et al., 2017; Meyer, Bekkering, Paulus, & Hunnius, 2010; Yin et al., 2013, 2016). In contrast, even though people in some competitive situations do have the same goal and coordinative actions, they behave more individually and less interdependently (Jin et al., 2017; Jin, Li, He, & Shen, 2018; Yin et al., 2013).
Exploring an Imagined “We” in Human Collective Hunting: Joint Commitment within Shared Intentionality
2022, Proceedings of the 44th Annual Meeting of the Cognitive Science Society: Cognitive Diversity, CogSci 2022Level 2 visual perspective-taking at age 3 and the corresponding effect of cooperation
2019, Acta Psychologica Sinica