ICCM Symposium

The Continued Influence Effect (CIE) refers to the psychological phenomenon where misinformation continues to influence people’s reasoning, even after a correction/retraction has been presented and accepted as true. The CIE has been found to be empirically robust across a variety of manipulation conditions, but there is still no consensus on the theoretical framework underlying this phenomenon. While multiple theoretical frameworks have been proposed, no single explanation has been widely accepted, or can explain the CIE across conditions. In particular, there is a need for an overarching theoretical model that can integrate cognitive, social, and affective factors. This talk will introduce a novel theoretical modeling approach for the CIE that applies Thagard’s Theory of Explanatory Coherence and its computational instantiation, ECHO, to narratives in the CIE. The aim of this discussion is to explain how the coherence of narratives used in CIE paradigms can influence a person’s reliance on misinformation, and how to assess coherence of misinformation and corrections in a quantifiable manner. Moreover, this computational approach also allows for the potential integrations of both cognitive and socio-affective factors. This not only has implications for a model that could both make predictions and explain behavior for the CIE across a variety of conditions, but it also contributes towards the development of an overarching framework for the CIE phenomenon.

A common approach to the issues of information acquisition, attitude formation, and belief maintenance is to view them through a normative prism. Knowledge of ground truth and reliability of trusted sources are assumed, and one is left to try to explain why wrong beliefs are so common. In many ways this is similar to the approaches to cognitive biases that assume rational inference to be easy and straightforward and then make special assumptions about unexplained distortions. Instead, our approach recognizes the difficulty of deriving and maintaining consistent, coherent and compatible belief sets in a world of confusing and conflicting information. The task can then be viewed as a difficult instance of sensemaking that requires determining which sources to trust, what facts is compatible with existing beliefs, and when to revise attitudes considering new information. Given the difficulty and ubiquity of the task, we assume that the core processes directly tap into mechanisms of the cognitive architecture. In particular, there is a mutual dependency between learning trust in sources that reflect the compatibility of their input with currently held views and revising attitudes and beliefs in a way that factors trust in the origin of the information. High-fidelity computational cognitive models can make quantitative predictions of specific effects across a broad range of situations. Currently lacking to enable further progress are datasets that match the degree of precision and fidelity of the models. Leveraging new AI technologies such as large language models might be the key for using abundant naturalistic datasets.

Information and influence operations are not new, but are a growing concern due to advancements in technology enabling the sharing of information with little cost. Experimental research in this area has started to build a foundation to understand how such operations exploit human’s cognitive vulnerabilities. However, there are mixed findings, a lack of consistency, and a stronger focus on artificial tasks that may not generalize to the real world. Furthermore, few computational models exist to predict, explain, and simulate information effects in individuals and models at the group level focus on information as a contagion. In order to make significant progress in understanding information and influence effects, we need a comprehensive and general theory capable of spanning from individuals to social networks. We need strong and consistent experimental methods, wider data collection, computational models, and realistic experimental materials. This symposium aims to bring together the ICCM community to have meaningful discussions about novel ideas for experiment designs and development of computational models capable of simulating information and influence effects within individuals and groups. Talks will include relevant research, novel ideas and frameworks to address research gaps, and discussion about how we can overcome challenges through collaborations.