Behavioral and neuro-cognitive bases for emergence of norms and socially shared realities via dynamic interaction

Abstract

In the digital era, new socially shared realities and norms emerge rapidly, whether they are beneficial or harmful to our societies. Although these are emerging properties from dynamic interaction, most research has centered on static situations where isolated individuals face extant norms. We investigated how perceptual norms emerge endogenously as shared realities through interaction, using behavioral and fMRI experiments coupled with computational modeling. Social interactions fostered convergence of perceptual responses among people, not only overtly but also at the covert psychophysical level that generates overt responses. Reciprocity played a critical role in increasing the stability (reliability) of the psychophysical function within each individual, modulated by neural activity in the mentalizing network during interaction. These results imply that bilateral influence promotes mutual cognitive anchoring of individual views, producing shared generative models at the collective level that enable endogenous agreement on totally new targets-one of the key functions of social norms.

Fig. 1. Task and results from the behavioral experiment.

a Timelines of the dot-estimation task. In Phases 1 and 3 (solo phases), all participants estimated the number of dots individually. In Phase 2, participants in the individual condition (n = 21) performed the task individually again as a control. Those in the pair condition (n = 42) observed the same dots and estimated the number of dots independently, and then their estimates were shared with each other. b Participants’ estimation weights (w_i) in Phases 1 and 3 in the pair condition. Each point indicates one participant’s estimation weight. c Absolute differences in estimation weights between two participants of the real and shuffled pairs in Phases 1 and 3. The box plots indicate the medians, the first and third quartiles, and the values no further than 1.5 inter-quartile range from the quartiles. *P < 0.05, **P < 0.01. d An illustration of the state-space model that was used to measure the stability of estimation weights within each individual. We assumed that the estimation weight at trial t, w(t), was sampled with observational noise from a latent state, μ(t). The latent state μ(t) was in turn assumed to be sampled with system noise (σ) from the state in the preceding trial, μ(t-1). We used σ as an index of stabilization of estimation weights within each participant. In the bottom row, y indicates the participant’s estimate. e Sigmas of estimation weights in the pair and individual conditions. The bar plots indicate the means across participants. ***P < 0.001.

Fig. 2. Task flow and behavioral results from the functional magnetic resonance imaging experiment.

a The flow of the fMRI experiment. In the pre- and post-interaction phases, participants performed the dot-estimation task individually. In each interaction phase, the participant was paired with the Sherif- or Asch-type partner that had a much stronger underestimation bias initially (w = 0.61) than the average participant of the behavioral experiment. The order of the two partners was counterbalanced across participants. The participant and the partner observed the same dots and estimated the number of dots independently, and then their estimates were shared. b Participants’ estimation weights (w_i) in each phase. Each point indicates one participant’s estimation weight. c Sigmas of estimation weights, indexing stability of the participant’s covert psychophysical function. ***P < 0.001. d Coefficients for similarity (Sim) for the two computer partners in the time-series analysis of participants’ estimates during interaction. **P < 0.01, ***P < 0.001. The bar plots indicate the means across participants.

Fig. 3. Imaging results from the functional magnetic resonance imaging experiment.

a Individual right temporoparietal junction (RTPJ) and dorsomedial prefrontal cortex (DMPFC) regions of interest (ROIs) identified a priori by the functional localizer for cognitive perspective taking based on a theory-of-mind task, conducted separately from the main estimation task. The color bars show the number of overlapped individual ROIs in each voxel. b Parametric modulation for Sim (similarity) with the partner in estimation (Eq. 13) in RTPJ activity during interaction. *P < 0.05. c Parametric modulation for Est (estimation: Eq. 11) in DMPFC activity during interaction, with larger activation for larger estimates. †P < 0.10, *P < 0.05. d Correlations between RTPJ betas (in response to Sim during interaction: Fig. 3b) and the stability (σ) of the participants’ estimation weights in the post-interaction phase. e Correlations between RTPJ–DMPFC functional connectivity during interaction and the stability (σ) of the participants’ estimation weights in the post-interaction phase. The bar plots indicate the means across participants, and the lines in the scatter plots are the linear regression lines.

Fig. 4. Stabilization of estimation weights after interaction in the online behavioral experiment.

Regardless of the estimation biases, participants’ estimation weights became more stable (i.e., smaller σ of estimation weights) after interacting with the Sherif-type partner than with the Asch-type partner. The effect of the Sherif-type partner was more pronounced in the overestimation condition than in the underestimation condition. The bar plots indicate the means across participants.