Social Cognition

Repeated games provide a controlled yet rich framework for studying social interactions. Classic game-theoretic analyses focus on determining equilibrium solutions for completely self-interested, fully informed, and rational agents. Other, arguably more valid approaches, tend to focus on agents with socially-oriented utility functions, or with less-than perfect planning-ahead capabilities. Here, I will present the framework of interacting latent Markov chains in repeated games, leveraging the strengths of both approaches. I will assume that at each decision point, a player's actions are determined by their discrete latent state, which may depend on learned beliefs about the strategies of other players, yet also on the social orientation towards other players (e.g. love, hate, ignorance). Players can transition between such states as a consequence of the actions of all players. I will show the unique advantages of this framework, and contrast it to other common approaches to repeated games, including traditional solutions such as subgame perfect Nash equilibria and Markov perfect equilibria.

This is an in-person presentation on July 22, 2024 (15:20 ~ 15:40 CEST).

How does striving for cognitive consistency affect political polarization? The psychological consistency theories have found large support in empirical research, yet, the temporal, social, and political consequences of individuals striving for cognitive consistency are little understood. In the present work, an agent-based model simulating change in individuals' political beliefs was developed, based on the basic assumption that people strive for cognitive consistency in their political beliefs. While it can be assumed that all people strive for cognitive consistency, research has also shown the understanding of what consistency means for an individual differs. Hence, people differ in their cognitive models. In the present work, a cognitive model is formalized as a correlation matrix that indicates how strongly different beliefs should relate to each other. To derive different cognitive models, correlational class analysis is used on the European Social Survey data. In the agent-based simulation, agents can interact with each other and change their beliefs towards more consistency. Simulation results are analyzed regarding political polarization/consensus, showing that striving for consistency increases polarization, leads to more extreme beliefs, yet is also necessary to maintain diversity in beliefs. Differences in cognitive models amplify these effects. Limitations and different design choices are discussed. Although the macro-level results of the simulation are difficult to validate, it is argued for the value of testing cognitive psychological theory and its consequences in the temporal and social interplay.

This is an in-person presentation on July 22, 2024 (15:40 ~ 16:00 CEST).

Investments in public health or environmental sustainability require funding agencies to choose which proposals have the most valuable outcomes, but little is known about how the subjective value of environmental and health outcomes change in the face of delay and risk. This talk investigates the cognitive mechanisms that influence the subjective value of sea-level rise (environmental), diagnoses of tobacco-use related illness (health), and monetary losses when choosing between outcomes and when pricing outcomes in isolation. Experiment 1 explored how these outcomes are valued under delay and risk independently. Experiment 2 tested the value of these outcomes when they are both delayed and risky. Cognitive models of the choice and pricing tasks were used to compare changes in subjective value between the outcomes and tasks through parameters that represent the underlying cognitive processes involved in determining the value of a delayed and/or risky outcome. Results suggest that pricing favors larger outcomes compared to choice and that this effect is primarily driven by anchoring on the outcome and a higher sensitivity to the outcome during pricing. Monetary and environmental outcomes generally showed more similarities to one another than to health outcomes, with the most serious health outcomes retaining relatively more value in the face of long delays or low probabilities of receipt than monetary or environmental outcomes.

This is an in-person presentation on July 22, 2024 (16:00 ~ 16:20 CEST).

Humans are remarkably adaptive instructors who can adjust advice on complex tasks based on a learner’s current knowledge and goals. However, paradigms in cognitive psychology generally examine pedagogy in constrained and discrete tasks, like categorization or feature learning, with a small set of actions for the instructor to choose from. We examine teaching in continuous domains, where there are theoretically infinite choices, and model how teachers formulate a computationally tractable Bayesian inference about which choice is best. We propose that teachers can reason about learners as agents that update their hypotheses using Gaussian process regression. We tested this in a visual function completion task, in which an agent observes dots placed along a 2D plane and must draw a line that represents the underlying function that produced those dots. We modeled how a teacher, who knows the underlying function, would select points that best help a learner recover the function. Preliminary evidence suggests teachers are sensitive to learners’ priors about continuous functions. For instance, when learners expect a diverse range of function types (linear, quadratic, periodic, etc.) then teachers tend to select examples that help distinguish between those types. Conversely, teachers did not adhere to this pattern if learners saw just one function type. Ongoing analyses will compare human data to a model-based teacher agent that reasons about the learner’s Gaussian process regression to evaluate the utility of showing certain points. Our results provide insight into how teachers formulate pedagogical guidance in computationally tractable ways, even in complex, continuous domains.

This is an in-person presentation on July 22, 2024 (16:20 ~ 16:40 CEST).