Socially shared realities develop rapidly within and across groups in modern societies where social media connect many people in near-real time. Examples include the formation of collective prejudice against foreigners during the COVID-19 pandemic and the broadening consensus for decarbonized transport and energy, not to mention the moral and political divides between social sectors. These socially shared realities ― including broad values, specific beliefs, and even basic perceptions ― affect what people see or attend to in the focal situation, how and to whom what they have seen is communicated, and when and how people edit their specific behaviors. Accordingly, these realities fundamentally affect both the processes and outcomes of collective decision-making at various levels, ranging from pairs to groups to organizations to societies. Whether such shared realities are beneficial or harmful to our societies or even relevant to our existing values, it is noteworthy that shared realities are not mere duplications of exogenous and extant rules but emerge endogenously from people’s interactions, and thus they may be hard to change once established.

 Despite these emergent properties, however, most previous studies on shared realities and norms have addressed how people conform to or deviate from socially fixed and extant situations. Although some studies have focused on interactive settings, even they have not been able to address cognitive processes and temporal dynamics during the interaction. Therefore, our understanding of the dynamic emergence of shared realities through interaction remains elusive. With neuro-cognitive techniques, this dissertation aims to shed light on computational mechanisms underlying the formation of socially shared realities in perceptual decision-making.

 As a general introduction, Chapter 1 first organizes the concept of shared reality, describing its relationship to social norms. I then review theories on norm formation and claim that shared realities can emerge through an interactive process similar to that of norms. To set the stage for the argument of this dissertation, this chapter next outlines two representative experimental paradigms to study shared reality and points out two limitations of previous studies: (i) exclusively focusing on how people obey fixed reality; (ii) not modeling changes in peoples’ internal processes through social interaction. In particular, it is crucial to examine the internal model underlying shared reality because the generative feature of such a model enables endogenous agreement among people on new targets beyond the initially experienced target, thus producing shared reality. The last section of this chapter gives a brief overview of the dissertation and experiments.

 Chapter 2 addresses two research questions: How do people develop generalized shared realities through interaction? Also, how persistently and reliably do the generalized shared realities modulate individual behavior after the interaction? The former concerns the emergence of shared reality, and the latter relates to the maintenance or stabilization of shared reality. Based on the results of Sherif’s (1936) seminal study, I hypothesized that: (H1) Dyadic interaction causes convergence of not only participants’ overt behaviors but also their covert psychophysical functions within a pair; (H2) Dyadic interaction stabilizes the psychophysical function within each individual after the interaction.

 To test the hypotheses, I ran a laboratory behavioral experiment and had pairs of participants perform a dot-estimation task. Participants were asked to perform a dot-estimation task throughout the experiment. In this task, participants observed random dots for 0.8 seconds and estimated the number of dots. No feedback was given to participants about their accuracy. There were three phases in the experiment. Participants performed the task individually in Phases 1 and 3 (hereafter solo phases). In Phase 2, participants were randomly assigned to the individual or pair condition. Participants in the individual condition performed the task individually again. In the pair condition, two participants observed the same dots, answered independently, and then shared their estimates. Pairs were instructed that they did not need to give similar estimates and that rewards would depend on individual estimation accuracy. Participants remained anonymous from each other and were allowed no verbal communication.

 Dyadic interaction resulted in paired convergence of not only participants' overt behaviors but also their covert psychophysical functions (H1). As a side consequence, the paired interaction helped to stabilize each individual's psychophysical function (H2). These results suggest that the convergence and stabilization of covert psychophysical functions (i.e., shared reality) occurred not exogenously but instead due to interaction. This is in line with my view that value-free shared realities emerge within a group from the same interactive processes as value-based ones such as norms. This study also supports the theory that shared reality can be generalized to other targets by examining psychophysical functions, i.e., the covert rules of behavior.

 Chapter 3 focuses on the interaction mechanics, the neuro-cognitive bases for emergence of shared reality, and the temporal processes during interaction. Given the critical roles of reciprocity and perspective-taking in interaction and human sociality, I hypothesized that: (H3) Reciprocal concession during interaction stabilizes the covert psychophysical function within each individual after the interaction; (H4) Activity of the mentalizing network during bilateral interaction modulates stabilization of the covert psychophysical function within each individual after the interaction.

 To address these points, I conducted an fMRI experiment. The experiment consisted of one pre-interaction phase, two interaction phases, and two post-interaction phases. In the pre-interaction phase, participants performed the dot-estimation task individually. Participants were paired with the Sherif- or Asch-type computer partner (described as another participant) in the interaction phases. Participants and the partner estimated the number of dots independently, and then their estimates were shared. Although both computer partners had a stronger underestimation bias initially than the average participant of the laboratory behavioral experiment, they were programmed to differ in their reciprocity to participants’ estimates during interaction. The Asch-type retained its underestimation bias throughout trials irrespective of the participant’s estimates. In contrast, the Sherif-type adjusted its weight over trials toward the participant’s estimates. The Sherif-type was designed to approximate the pattern of an average participant’s estimation-adjustment during interaction in the laboratory behavioral experiment. In the post-interaction phases, participants performed the task individually again.

 As hypothesized in H3, within-individual stability of the covert psychophysical function improved more after interacting with the Sherif-type than with the Asch-type partner. Furthermore, the time-series analysis confirmed that participants adjusted their estimation weights reciprocally during interaction only with the Sherif-type partner. At the neural level, the mentalizing network contributed to the dynamic formation and stabilization of shared realities. Consistent with the behavioral results, RTPJ activity tracked temporal changes in estimation similarity during the interaction when paired with the reciprocating Sherif-type but not with the one-sided Asch-type partner. Such RTPJ activity, which parametrically modulated the estimation similarity with the Sherif-type partner on a trial-by-trial basis, also contributed to the subsequent stabilization of participants’ covert psychophysical functions (H4). The functional connectivity between the RTPJ and the DMPFC modulated the stabilization of participants’ covert psychophysical functions after interacting with the reciprocating Sherif-type but not with the unresponsive Asch-type partner (H4). These results indicate that participants spontaneously had the reciprocating partner in mind when they made their estimations during and after the interaction.

 In Chapter 4, I conducted a pre-registered experiment to replicate the fMRI experiment’s results and examine whether similar behavior may be observed when the computer partner had an over-estimation bias in the dot-estimation task. The experiment had pre-interaction, interaction, and post-interaction phases. In the pre- and post-interaction phases, participants performed the task individually. Participants and the computer partner performed the task in the interaction phase as in the fMRI experiment. For the computer partner, I had a 2 × 2 between-participant design, with factors of the partner type (Sherif-type vs. Asch-type) and the built-in estimation bias (underestimation vs. overestimation).

 Participants decreased their estimation weights after interacting with the underestimators. This pattern replicated the results of the fMRI experiment. When interacting with the overestimators, participants increased their estimation weights for the Asch-type, but not for the Sherif-type. A 2 × 2 (partner type × estimation bias) ANOVA yielded a significant interaction and a main effect of the estimation bias. These patterns indicate that participants changed their estimation weights toward the computer partners, except for the overestimating Sherif-type that approached participants’ estimates rapidly during interaction according to the algorithm. I next examined whether the results on increased estimation stability are replicable. A 2 × 2 (partner type × estimation bias) ANOVA on the pre-post differences yielded significant main effects for both factors. This indicates that σ decreased more after interacting with the Sherif-type than with the Asch-type (H3) and decreased more after interacting with the underestimator than with the overestimator. The interaction effect was also significant, suggesting that the increased stability due to the reciprocal Sherif-type was more pronounced in the overestimator condition. These results replicate and extend the behavioral results of the fMRI experiment. Reciprocal concession stabilized participants’ covert psychophysical functions in both estimation-bias conditions.

 Chapter 5 summarizes the results and discusses their implications for shared reality. Firstly, previous studies just claimed that shared reality is about a target referent such as impressions about a newcomer or a specific TV program, about a particular politician, or political or religious matters. The results in this dissertation add that shared reality also requires the condition that the target referent is uncertain or unverifiable. Secondly, social verification and perspective-taking can form a positive loop in increasing the stability of shared reality. Being verified by your partner leads you to think that the partner is reasonable enough to deserve perspective-taking and to approach the partner. This also supports the theory that social norms emerge through the loop of positive feedback of behaviors and beliefs, as reviewed in Chapter 1. Such a common process may underpin the emergence and maintenance of social norms and shared reality. Finally, the results of the studies demonstrate the importance of defining an internal rule of behaviors when considering shared reality. Previous studies argued that shared reality is not a mere duplication or catching of another person’s response but requires that one’s inner state about some target referent converges with the other’s internal state regarding the same target. My finding that bilateral interaction causes convergence of not only overt responses about the same perceptual stimuli but also covert psychophysical functions via perspective-taking fits this view. Also note that, in the laboratory behavioral experiment, participants’ psychophysical functions converged within a pair even in Phase 3, i.e., when performing the task independently. This seems to indicate that a shared stable psychophysical function allows people to agree on a new random target beyond the initially and socially experienced targets. Although previous research already found that shared reality can be generalized to similar but other stimuli, my studies contribute to the theory of shared reality by elaborating the basis of such generalization. These generalized endogenous agreements may foster a shared sense of feeling and thinking the same way. I believe that such a shared generative model is one of the most fundamental characteristics of shared reality.

 People rely on constant interaction with others to develop and sustain shared realities. Extending the applicability of the computational approach to more social ones while focusing on the formation of shared generative models, could be a promising way to better understand emerging problems in our digitally connected but often morally divided world.