Visual-spatial dynamics drive adaptive social learning in immersive environments

Abstract

Humans are uniquely capable social learners. Our capacity to learn from others across short and long timescales is a driving force behind the success of our species. Yet there are seemingly maladaptive patterns of human social learning, characterized by both overreliance and underreliance on social information. Recent advances in animal research have incorporated rich visual and spatial dynamics to study social learning in ecological contexts, showing how simple mechanisms can give rise to intelligent group dynamics. However, similar techniques have yet to be translated into human research, which additionally requires integrating the sophistication of human individual and social learning mechanisms. Thus, it is still largely unknown how humans dynamically adapt social learning strategies to different environments and how group dynamics emerge under realistic conditions. Here, we use a collective foraging experiment in an immersive Minecraft environment to provide unique insights into how visual-spatial interactions give rise to adaptive, specialized, and selective social learning. Our analyses show how groups adapt to the demands of the environment through specialization rather than homogeneity and through the adaptive deployment of selective imitation rather than indiscriminate copying. We test these mechanisms using computational modeling, providing a deeper understanding of the cognitive mechanisms that dynamically influence social decision-making in ecological contexts. Specialization of learning strategies and selective attention thus offer insights into the adaptive foundations of human social learning.

Collective incentives reduce over-exploitation of social information in unconstrained human groups

“…Our work not only translates and tests predictions from idealized producer-scrounger models in more realistic social scenarios but, thereby, also highlights the fundamental role space and perception play in modulating social decision-making. Moreover, we found that participants calibrated their strategies over time, becoming more (less) likely to use social information in concentrated (distributed) environments, extending previous research highlighting the importance of selective and strategic social learning rather than pure copying or innovation 13,15,20 .…”

“…Although abstracting away from real-world details and controlling the environment participants face can allow researchers to more accurately identify cognitive processes and strategies, ultimately, we aim to understand how people make unconstrained decisions in relevant realworld ecologies. Technological advances now provide us with unprecedented access to the individual-level informational environments and constraints that guide strategic choices in humans and other animals 20,52,53 . Such dynamic data require dynamic statistical inference and Hidden Markov models provide ideal tools to simultaneously extract meaningful patterns from multidimensional time-series data and use internal or external situational factors to predict switches between the identified hidden states.…”

“…Although our model is tailored to the present experimental paradigm (especially with respect to the state-dependent variables), our freely available modeling code and indepth documentation set the scene for future research on the socioecological drivers of social decision-making in human and (nonhuman) animal collectives. In addition to other naturalistic behavioral experiments 20,54,55 , Social Hidden Markov Decision models can, for instance, be applied to human crowd behavior to better understand how situational factors influence movement patterns and potentially cause stampedes 56,57 ; they can be adapted to sports analytics where Hidden Markov models have already been used to identify drive events and defensive assignments in basketball 58,59 or "hot hands" in darts 60 ; they could elucidate leader-follower dynamics in animal societies and help us better understand when and why animals follow the example of others 61,62 ; and they could be applied to GPS data from subsistence foragers 63,64 , cell phone users 65,66 or migratory animals [67][68][69] to infer modes of (collective) search, movement and space use across different spatial and temporal scales.…”

“…To understand the mechanisms governing real-world human collective systems, we need paradigms that allow complex social dynamics to unfold within naturalistic environments. Analyzing behavior in such complex systems requires novel computational models that describe how dynamically interacting individuals make decisions while accounting for their unique (visual) perspectives and spatial constraints 2,19,20 . Such constraints are unavoidable features of the real world and fundamentally shape the costs and benefits of social information use 19,21 ; they are thus prerequisites for testing collective dynamics in more realistic settings and connecting abstract models to reality.…”

“…1a and Supplementary Movie 1). The 3D environment imposes a limited, first-person, field of view as well as realistic spatial constraints creating a natural trade-off between individual exploration of the environment and social information use 20,22 . Participants completed four rounds of the task in a 2 × 2 design (Fig.…”