skip to main content


Title: MGpi: A Computational Model of Multiagent Group Perception and Interaction
Toward enabling next-generation robots capable of socially intelligent interaction with humans, we present a computational model of interactions in a social environment of multiple agents and multiple groups. The Multiagent Group Perception and Interaction (MGpi) network is a deep neural network that predicts the appropriate social action to execute in a group conversation (e.g., speak, listen, respond, leave), taking into account neighbors' observable features (e.g., location of people, gaze orientation, distraction, etc.). A central component of MGpi is the Kinesic-Proxemic-Message (KPM) gate, that performs social signal gating to extract important information from a group conversation. In particular, KPM gate filters incoming social cues from nearby agents by observing their body gestures (kinesics) and spatial behavior (proxemics). The MGpi network and its KPM gate are learned via imitation learning, using demonstrations from our designed social interaction simulator. Further, we demonstrate the efficacy of the KPM gate as a social attention mechanism, achieving state-of-the-art performance on the task of group identification without using explicit group annotations, layout assumptions, or manually chosen parameters.  more » « less
Award ID(s):
1637927
NSF-PAR ID:
10308753
Author(s) / Creator(s):
; ;
Date Published:
Journal Name:
AAMAS '20: Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Systems
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Despite increasing awareness and research about online strategic information operations, there remain gaps in our understanding, including how information operations leverage the wider information ecosystem and take shape on and across multiple social media platforms. In this paper we use mixed methods, including digital trace ethnography, to look beyond a single social media platform to the broader information ecosystem. We aim to understand how multiple social media platforms are used, in parallel and complementary ways, to achieve the strategic goals of online information operations. We focus on a specific case study: the contested online conversation surrounding Syria Civil Defense (the White Helmets), a group of first responders that assists civilians affected by the civil war within the country. Our findings reveal a network of social media platforms from which content is produced, stored, and integrated into the Twitter conversation. We highlight specific activities that sustain the strategic narratives and attempt to influence the media agenda. And we note that underpinning these efforts is the work of resilience-building: the use of alternative (non-mainstream) platforms to counter perceived threats of 'censorship' by large, established social media platforms. We end by discussing the implications on social media platform policy. 
    more » « less
  2. null (Ed.)
    Robots' spatial positioning is a useful communication modality in social interactions. For example, in the context of group conversations, certain types of positioning signal membership to the group interaction. How does robot embodiment influence these perceptions? To investigate this question, we conducted an online study in which participants observed renderings of several robots in a social environment, and judged whether the robots were positioned to take part in a group conversation with other humans in the scene. Our results suggest that robot embodiment can influence perceptions of conversational group membership. An important factor to consider in this regard is whether robot embodiment leads to a discernible orientation for the agent. 
    more » « less
  3. Internships are widely viewed as “door openers” to opportunity, yet students without ample financial, social, and institutional supports are often excluded from these experiences. This exclusion is especially problematic for Latinx students attending Hispanic-Serving Institutions, for whom an internship could be transformative. In this article, we elaborate upon Núñez’s (2014) multilevel model of intersectionality to highlight how agents’ perceptions, social categories, embodied practices, and broader contextual forces interrelate to shape Latinx students’ access to internship opportunities at a university in the border region of Texas. Using inductive thematic, correlational, and social-network analysis techniques to analyze survey (n = 192) and focus group (n = 12) data, we find that gender, academic major, socioeconomic status, and race intersect with organizational (e.g., insufficient information) and contextual (e.g., labor markets) factors to shape students’ access and perceptions. We provide recommendations for disrupting systemic inequalities in internship access and culturally appropriate programming and ideas for future research. 
    more » « less
  4. Abstract

    Both individual and group behavior can influence individual fitness, but multilevel selection is rarely quantified on social behaviors. Social networks provide a unique opportunity to study multilevel selection on social behaviors, as they describe complex social traits and patterns of interaction at both the individual and group levels. In this study, we used contextual analysis to measure the consequences of both individual network position and group network structure on individual fitness in experimental populations of forked fungus beetles (Bolitotherus cornutus) with two different resource distributions. We found that males with high individual connectivity (strength) and centrality (betweenness) had higher mating success. However, group network structure did not influence their mating success. Conversely, we found that individual network position had no effect on female reproductive success but that females in populations with many social interactions experienced lower reproductive success. The strength of individual-level selection in males and group-level selection in females intensified when resources were clumped together, showing that habitat structure influences multilevel selection. Individual and emergent group social behavior both influence variation in components of individual fitness, but impact the male mating success and female reproductive success differently, setting up intersexual conflicts over patterns of social interactions at multiple levels.

     
    more » « less
  5. Customers are rapidly turning to social media for customer support. While brand agents on these platforms are motivated and well-intentioned to help and engage with customers, their efforts are often ignored if their initial response to the customer does not match a specific tone, style, or topic the customer is aiming to receive. The length of a conversation can reflect the effort and quality of the initial response made by a brand toward collaborating and helping consumers, even when the overall sentiment of the conversation might not be very positive. Thus, through this study, we aim to bridge this critical gap in the existing literature by analyzing language’s content and stylistic aspects such as expressed empathy, psycho-linguistic features, dialogue tags, and metrics for quantifying personalization of the utterances that can influence the engagement of an interaction. This paper demonstrates that we can predict engagement using initial customer and brand posts. 
    more » « less