Large networks of rational agents form persistent echo chambers

Analogies for modeling belief dynamics

Belief dynamics has an important role in shaping our responses to natural and societal phenomena, ranging from climate change and pandemics to immigration and conflicts. Researchers often base their models of belief dynamics on analogies to other systems and processes, such as epidemics or ferromagnetism. Similar to other analogies, analogies for belief dynamics can help scientists notice and study properties of belief systems that they would not have noticed otherwise (conceptual mileage). However, forgetting the origins of an analogy may lead to some less appropriate inferences about belief dynamics (conceptual baggage). Here, we review various analogies for modeling belief dynamics, discuss their mileage and baggage, and offer recommendations for using analogies in model development.

Preferences for advisor agreement and accuracy

Previous research has shown that people are more influenced by advisors who are objectively more accurate, but also by advisors who tend to agree with their own initial opinions. The present experiments extend these ideas to consider people's choices of who they receive advice from-the process of source selection. Across a series of nine experiments, participants were first exposed to advisors who differed in objective accuracy, the likelihood of agreeing with the participants' judgments, or both, and then were given choice over who would advise them across a series of decisions. Participants saw these advisors in the context of perceptual decision and general knowledge tasks, sometimes with feedback provided and sometimes without. We found evidence that people can discern accurate from inaccurate advice even in the absence of feedback, but that without feedback they are biased to select advisors who tend to agree with them. When choosing between advisors who are accurate vs. likely to agree with them, participants overwhelmingly choose accurate advisors when feedback is available, but show wide individual differences in preference when feedback is absent. These findings extend previous studies of advice influence to characterise patterns of advisor choice, with implications for how people select information sources and learn accordingly.

Generalized naming game and Bayesian naming game as dynamical systems

We study the β model (β-NG) and the Bayesian Naming Game (BNG) as dynamical systems. By applying linear stability analysis to the dynamical system associated with the β model, we demonstrate the existence of a nongeneric bifurcation with a bifurcation point β_{c}=1/3. As β passes through β_{c}, the stability of isolated fixed points changes, giving rise to a one-dimensional manifold of fixed points. Notably, this attracting invariant manifold forms an arc of an ellipse. In the context of the BNG, we propose modeling the Bayesian learning probabilities p_{A} and p_{B} as logistic functions. This modeling approach allows us to establish the existence of fixed points without relying on the overly strong assumption that p_{A}=p_{B}=p, where p is a constant.

Political reinforcement learners

Politics can seem home to the most calculating and yet least rational elements of humanity. How might we systematically characterize this spectrum of political cognition? Here, we propose reinforcement learning (RL) as a unified framework to dissect the political mind. RL describes how agents algorithmically navigate complex and uncertain domains like politics. Through this computational lens, we outline three routes to political differences, stemming from variability in agents' conceptions of a problem, the cognitive operations applied to solve the problem, or the backdrop of information available from the environment. A computational vantage on maladies of the political mind offers enhanced precision in assessing their causes, consequences, and cures.

Individuals, Collectives, and Individuals in Collectives: The Ineliminable Role of Dependence

Our beliefs are inextricably shaped through communication with others. Furthermore, even conversation we conduct in pairs may itself be taking place across a wider, connected social network. Our communications, and with that our thoughts, are consequently typically those of individuals in collectives. This has fundamental consequences with respect to how our beliefs are shaped. This article examines the role of dependence on our beliefs and seeks to demonstrate its importance with respect to key phenomena involving collectives that have been taken to indicate irrationality. It is argued that (with the benefit of hindsight) these phenomena no longer seem surprising when one considers the multiple dependencies that govern information acquisition and the evaluation of cognitive agents in their normal (i.e., social) context.

Knowledge through social networks: Accuracy, error, and polarisation

This paper examines the fundamental problem of testimony. Much of what we believe to know we know in good part, or even entirely, through the testimony of others. The problem with testimony is that we often have very little on which to base estimates of the accuracy of our sources. Simulations with otherwise optimal agents examine the impact of this for the accuracy of our beliefs about the world. It is demonstrated both where social networks of information dissemination help and where they hinder. Most importantly, it is shown that both social networks and a common strategy for gauging the accuracy of our sources give rise to polarisation even for entirely accuracy motivated agents. Crucially these two factors interact, amplifying one another's negative consequences, and this side effect of communication in a social network increases with network size. This suggests a new causal mechanism by which social media may have fostered the increase in polarisation currently observed in many parts of the world.

Naïve information aggregation in human social learning

To glean accurate information from social networks, people should distinguish evidence from hearsay. For example, when testimony depends on others' beliefs as much as on first-hand information, there is a danger of evidence becoming inflated or ignored as it passes from person to person. We compare human inferences with an idealized rational account that anticipates and adjusts for these dependencies by evaluating peers' communications with respect to the underlying communication pathways. We report on three multi-player experiments examining the dynamics of both mixed human-artificial and all-human social networks. Our analyses suggest that most human inferences are best described by a naïve learning account that is insensitive to known or inferred dependencies between network peers. Consequently, we find that simulated social learners that assume their peers behave rationally make systematic judgment errors when reasoning on the basis of actual human communications. We suggest human groups learn collectively through naïve signaling and aggregation that is computationally efficient and surprisingly robust. Overall, our results challenge the idea that everyday social inference is well captured by idealized rational accounts and provide insight into the conditions under which collective wisdom can emerge from social interactions.

Incorporating Psychological Science Into Policy Making: The Case of Misinformation

The spread of false and misleading information in online social networks is a global problem in need of urgent solutions. It is also a policy problem because misinformation can harm both the public and democracies. To address the spread of misinformation, policymakers require a successful interface between science and policy, as well as a range of evidence-based solutions that respect fundamental rights while efficiently mitigating the harms of misinformation online. In this article, we discuss how regulatory and nonregulatory instruments can be informed by scientific research and used to reach EU policy objectives. First, we consider what it means to approach misinformation as a policy problem. We then outline four building blocks for cooperation between scientists and policymakers who wish to address the problem of misinformation: understanding the misinformation problem, understanding the psychological drivers and public perceptions of misinformation, finding evidence-based solutions, and co-developing appropriate policy measures. Finally, through the lens of psychological science, we examine policy instruments that have been proposed in the EU, focusing on the strengthened Code of Practice on Disinformation 2022.

Psychological inoculation can reduce susceptibility to misinformation in large rational agent networks

The unchecked spread of misinformation is recognized as an increasing threat to public, scientific and democratic health. Online networks are a contributing cause of this spread, with echo chambers and polarization indicative of the interplay between the search behaviours of users and reinforcement processes within the system they inhabit. Recent empirical work has focused on interventions aimed at inoculating people against misinformation, yielding success on the individual level. However, given the evolving, dynamic information context of online networks, important questions remain regarding how such inoculation interventions interact with network systems. Here we use an agent-based model of a social network populated with belief-updating users. We find that although equally rational agents may be assisted by inoculation interventions to reject misinformation, even among such agents, intervention efficacy is temporally sensitive. We find that as beliefs disseminate, users form self-reinforcing echo chambers, leading to belief consolidation-irrespective of their veracity. Interrupting this process requires 'front-loading' of inoculation interventions by targeting critical thresholds of network users before consolidation occurs. We further demonstrate the value of harnessing tipping point dynamics for herd immunity effects, and note that inoculation processes do not necessarily lead to increased rates of 'false-positive' rejections of truthful communications.

Epistemic Communities under Active Inference

The spread of ideas is a fundamental concern of today's news ecology. Understanding the dynamics of the spread of information and its co-option by interested parties is of critical importance. Research on this topic has shown that individuals tend to cluster in echo-chambers and are driven by confirmation bias. In this paper, we leverage the active inference framework to provide an in silico model of confirmation bias and its effect on echo-chamber formation. We build a model based on active inference, where agents tend to sample information in order to justify their own view of reality, which eventually leads to them to have a high degree of certainty about their own beliefs. We show that, once agents have reached a certain level of certainty about their beliefs, it becomes very difficult to get them to change their views. This system of self-confirming beliefs is upheld and reinforced by the evolving relationship between an agent's beliefs and observations, which over time will continue to provide evidence for their ingrained ideas about the world. The epistemic communities that are consolidated by these shared beliefs, in turn, tend to produce perceptions of reality that reinforce those shared beliefs. We provide an active inference account of this community formation mechanism. We postulate that agents are driven by the epistemic value that they obtain from sampling or observing the behaviours of other agents. Inspired by digital social networks like Twitter, we build a generative model in which agents generate observable social claims or posts (e.g., 'tweets') while reading the socially observable claims of other agents that lend support to one of two mutually exclusive abstract topics. Agents can choose which other agent they pay attention to at each timestep, and crucially who they attend to and what they choose to read influences their beliefs about the world. Agents also assess their local network's perspective, influencing which kinds of posts they expect to see other agents making. The model was built and simulated using the freely available Python package pymdp. The proposed active inference model can reproduce the formation of echo-chambers over social networks, and gives us insight into the cognitive processes that lead to this phenomenon.

How a minority can win: Unrepresentative outcomes in a simple model of voter turnout

The outcome of an election depends not only on which candidate is more popular, but also on how many of their voters actually turn out to vote. Here we consider a simple model in which voters abstain from voting if they think their vote would not matter. Specifically, they do not vote if they feel sure their preferred candidate will win anyway (a condition we call complacency), or if they feel sure their candidate will lose anyway (a condition we call dejectedness). The voters reach these decisions based on a myopic assessment of their local network, which they take as a proxy for the entire electorate: voters know which candidate their neighbors prefer and they assume-perhaps incorrectly-that those neighbors will turn out to vote, so they themselves cast a vote if and only if it would produce a tie or a win for their preferred candidate in their local neighborhood. We explore various network structures and distributions of voter preferences and find that certain structures and parameter regimes favor unrepresentative outcomes where a minority faction wins, especially when the locally preferred candidate is not representative of the electorate as a whole.

Out-group animosity drives engagement on social media

There has been growing concern about the role social media plays in political polarization. We investigated whether out-group animosity was particularly successful at generating engagement on two of the largest social media platforms: Facebook and Twitter. Analyzing posts from news media accounts and US congressional members (n = 2,730,215), we found that posts about the political out-group were shared or retweeted about twice as often as posts about the in-group. Each individual term referring to the political out-group increased the odds of a social media post being shared by 67%. Out-group language consistently emerged as the strongest predictor of shares and retweets: the average effect size of out-group language was about 4.8 times as strong as that of negative affect language and about 6.7 times as strong as that of moral-emotional language-both established predictors of social media engagement. Language about the out-group was a very strong predictor of "angry" reactions (the most popular reactions across all datasets), and language about the in-group was a strong predictor of "love" reactions, reflecting in-group favoritism and out-group derogation. This out-group effect was not moderated by political orientation or social media platform, but stronger effects were found among political leaders than among news media accounts. In sum, out-group language is the strongest predictor of social media engagement across all relevant predictors measured, suggesting that social media may be creating perverse incentives for content expressing out-group animosity.

The evolution of polarization in the legislative branch of government

The polarization of political opinions among members of the US legislative chambers measured by their voting records is greater today than it was 30 years ago. Previous research efforts to find causes of such increase have suggested diverse contributors, like growth of online media, echo chamber effects, media biases or disinformation propagation. Yet, we lack theoretic tools to understand, quantify and predict the emergence of high political polarization among voters and their legislators. Here, we analyse millions of roll-call votes cast in the US Congress over the past six decades. Our analysis reveals the critical change of polarization patterns that started at the end of 1980s. In earlier decades, polarization within each Congress tended to decrease with time. By contrast, in recent decades, the polarization has been likely to grow within each term. To shed light on the reasons for this change, we introduce here a formal model for competitive dynamics to quantify the evolution of polarization patterns in the legislative branch of the US government. Our model represents dynamics of polarization, enabling us to successfully predict the direction of polarization changes in 28 out of 30 US Congresses elected in the past six decades. From the evolution of polarization level as measured by the Rice index, our model extracts a hidden parameter-polarization utility which determines the convergence point of the polarization evolution. The increase in the polarization utility implied by the model strongly correlates with two current trends: growing polarization of voters and increasing influence of election campaign donors. Two largest peaks of the model's polarization utility correlate with significant political or legislative changes happening at the same time.

Analytic Versus Computational Cognitive Models: Agent-Based Modeling as a Tool in Cognitive Sciences

Computational cognitive models typically focus on individual behavior in isolation. Models frequently employ closed-form solutions in which a state of the system can be computed if all parameters and functions are known. However, closed-form models are challenged when used to predict behaviors for dynamic, adaptive, and heterogeneous agents. Such systems are complex and typically cannot be predicted or explained by analytical solutions without application of significant simplifications. In addressing this problem, cognitive and social psychological sciences may profitably use agent-based models, which are widely employed to simulate complex systems. We show that these models can be used to explore how cognitive models scale in social networks to calibrate model parameters, to validate model predictions, and to engender model development. Agent-based models allow for controlled experiments of complex systems and can explore how changes in low-level parameters impact the behavior at a whole-system level. They can test predictions of cognitive models and may function as a bridge between individually and socially oriented models.

Influence and seepage: An evidence-resistant minority can affect public opinion and scientific belief formation

Some well-established scientific findings may be rejected by vocal minorities because the evidence is in conflict with political views or economic interests. For example, the tobacco industry denied the medical consensus on the harms of smoking for decades, and the clear evidence about human-caused climate change is currently being rejected by many politicians and think tanks that oppose regulatory action. We present an agent-based model of the processes by which denial of climate change can occur, how opinions that run counter to the evidence can affect the scientific community, and how denial can alter the public discourse. The model involves an ensemble of Bayesian agents, representing the scientific community, that are presented with the emerging historical evidence of climate change and that also communicate the evidence to each other. Over time, the scientific community comes to agreement that the climate is changing. When a minority of agents is introduced that is resistant to the evidence, but that enter into the scientific discussion, the simulated scientific community still acquires firm knowledge but consensus formation is delayed. When both types of agents are communicating with the general public, the public remains ambivalent about the reality of climate change. The model captures essential aspects of the actual evolution of scientific and public opinion during the last 4 decades.