Vlasceanu, M., Momennejad, I., & Dudik, M. (Submitted). Network structure, gender diversity, and interdisciplinarity predict the centrality of AI organizations. Publisher's VersionAbstract

Artificial intelligence (AI) research plays an increasingly important role in society, impacting key aspects of human life. From face recognition algorithms aiding national security in airports, to software that advises judges in criminal cases, and medical staff in healthcare, AI research is shaping critical facets of our experience in the world. But who are the people and institutional bodies behind this influen- tial research? What are the predictors of influence of AI researchers and research organizations? We study this question using social network analysis, in an exploration of the structural characteristics, i.e., network topology, of research organizations that shape modern AI. In a sample of 158 organizations with 16,385 affiliated authors of published papers in prominent AI conferences (e.g., NeurIPS, FAccT, AIES), we find that both industry and academic research organizations with influential authors are more interdisciplinary, more hierarchical, more gender diverse, and less clustered. Here, authors’ betweenness centrality in co-authorship networks was used as a measure of their influence. We also find that gender minorities (e.g., women) have less influence in the AI community, determined as lower betweenness centrality in co-authorship networks. These results suggest that while diversity adds significant value to AI based organizations, the individuals contributing to the increased diversity are marginalized in the AI field. We discuss these results in the context of current events with important societal implications.

Vlasceanu, M., McMahon, C., Van Bavel, J. J., & Coman, A. (Submitted). Political and non-political belief change elicits behavioral change.Abstract
Beliefs have long been posited to be a predictor of behavior. However, empirical investigations into the relationship between beliefs and behaviors, mostly correlational in nature, have provided conflicting findings. Here, we are interested in the causal impact of beliefs on behaviors. To explore this relationship in an experimental setting, in Study 1, 183 Cloud Research participants rated the accuracy of a set (half correct, half incorrect) of health-related statements (belief pre-test) and chose corresponding campaigns to which they could donate funds (behavior pre-test). They were then provided with relevant evidence in favor of the correct statements and against the incorrect statements. Finally, participants rated the accuracy of the initial set of statements again (belief post-test) and were given a chance to change their donation choices (behavior post-test). We found that evidence changed beliefs and this, in turn, led to behavioral change. In two follow-up studies (N=83, N=393), we replicated these findings with politically charged topics. To increase the ecological validity of our findings, the behavior measured (i.e., monetary donation) came at a personal cost to participants. We found that Democrats changed their behaviors as a function of belief change more for Democratic compared to Republican items. Republicans, however, did not exhibit a difference in their behavior change as a function of belief change between the Democratic and Republican items, pointing to an asymmetric partisan bias in the effect of belief change on behavioral change. These results are of particular relevance for interventions aimed at promoting constructive behaviors such as recycling, donating, or employing preventative health measures. 
Vlasceanu, M., & Coman, A. (Submitted). The Effect of Dyadic Conversations on Coronavirus-Related Belief Change. Publisher's VersionAbstract

In a high-risk environment, such as during an epidemic, people are exposed to a large amount of information, both accurate and inaccurate. Following exposure, they typically discuss the information with each other in conversations. Here, we assessed the effects of such conversations on their beliefs. A sample of 126 M-Turk participants first rated the accuracy of a set of COVID-19 statements (pre-test). They were then paired and asked to discuss either any of these statements (low epistemic condition) or only the statements they thought were accurate (high epistemic condition). Finally, they rated the accuracy of the initial statements again (post-test). We did not find a difference of epistemic condition on belief change. However, we found that individuals were sensitive to their conversational partners, and changed their beliefs according to their partners’ conveyed beliefs. This influence was strongest for initially moderately held beliefs. In exploratory analyses, we found that pre-test COVID-19 knowledge was predicted by trusting Fauci, not trusting Trump, and feeling threatened by COVID-19. Conversely, pre-test COVID- 19 conspiracy endorsement was predicted by trusting Trump, not trusting Fauci, news media consumption, social media usage, and political orientation. In further exploration of the political orientation predictor, we found that Democrats were more knowledgeable than Republicans, and Republicans believed more conspiracies than Democrats.

Vlasceanu, M., & Coman, A. (Forthcoming). The Impact of Information Sources on Covid-19 Knowledge Accumulation and Vaccination Intention. International Journal of Data Science and Analytics. Publisher's VersionAbstract

During a global health crisis people are exposed to vast amounts of information from a variety of sources. Here, we assessed which information source could increase knowledge about COVID-19 (Study 1) and COVID-19 vaccines (Study 2). In Study 1, a US census matched sample of 1060 Cloud Research participants rated the accuracy of a set of statements and then were randomly assigned to one of 10 between-subjects conditions of varying sources providing belief-relevant information: a political leader (Trump/Biden), a health authority (Fauci/CDC), an anecdote (Democrat/Republican), a large group of prior participants (Democrats/Republicans/Generic), or no source (Control). Finally, they rated the accuracy of the initial set of statements again. Study 2 involved a replication with a sample of 1876 Cloud Research participants, and focused on COVID-19 vaccine information and vaccination intention. In both studies, we found that participants acquired most knowledge when the source of information was a generic group of people. Surprisingly, knowledge accumulation from the different information sources did not interact with participants’ political affiliation.

Van Bavel, J. J., ..,, Vlasceanu, M., ..,, & Boggio, P. S. (2021). National identity predicts public health support during a global pandemic. Nature Communications. Publisher's VersionAbstract
The COVID-19 pandemic is a devastating global health crisis. Until vaccines or effective medications are widely administered within nations, the best hope for mitigating virus transmission is by changing collective behavior and supporting non-pharmaceutical interventions. In a large-scale international collaboration (Study 1, N = 49,968 across 67 countries), we investigated why people reported adopting public health behaviors (e.g., spatial distancing and stricter hygiene) and endorsed public policy interventions (e.g., closing bars and restaurants) during the pandemic (April-May, 2020). Respondents who identified more strongly with their nation consistently reported greater engagement in public health behaviors and support for public health policies. Study 2 (N = 42 countries) conceptually replicates the central finding using aggregate indices of national identity (World Values Survey) and a measure of actual behavior change during the pandemic (Google mobility report). Higher levels of national identification were associated with lower mobility (r = -.40). We discuss the implications of links between national identity, leadership, and public health for managing COVID-19 and future pandemics.
Vlasceanu, M. (2021). Cognitive Processes Shaping Individual and Collective Belief Systems. Psychology and Neuroscience . Princeton University. Publisher's VersionAbstract

Misinformation spread is among the top threats facing the world today. In my dissertation I strive to provide a deeper understanding of the socio-cognitive factors that can impact beliefs and of the potential for translating and implementing this understanding into policies aimed at reducing misinformation at a societal level. In the first Part, I introduce a theoretical generative framework for investigating factors influencing beliefs (Chapter 1). I then provide empirical support for this framework (N=7068) at the individual (Part 2, Chapters 2-5) and collective (Part 3, Chapters 6-8) levels of investigation. Finally, in Part 4 I suggest applications and future trajectories (Chapter 9). Starting with the individual level of investigation, in a series of online and lab experimental studies I show how psychological processes such as memory accessibility (Chapter 2), emotional arousal (Chapter 3), predictions errors (Chapter 4), and social norms (Chapter 5) can be leveraged to change people’s beliefs. For instance, in a series of laboratory experiments, I find that strengthening the memory of a statement increases its believability, while weakening its memory in a targeted fashion decreases its believability (Chapter 2). Given the interdependence between memory and emotions, in a series of online experiments, I also show that pairing emotionally arousing images with statements subsequently increases the believability of these statements compared to statements that had been associated with neutral or no images (Chapter 3). In another series of online experiments I explore the effect of prediction errors on belief update. I find that people update their beliefs as a function of the size of the errors they make when evaluating relevant evidence, and that making large errors leads to more belief update than not engaging in prediction, while controlling for the evidence available. Importantly, I find that these effects hold across ideological boundaries (Democrats and Republicans, evaluating Neutral, Democratic, and Republican beliefs; Chapter 4). In the last series of experiments at the individual level, I show that people change their beliefs more in line with evidence portrayed as normative (e.g., shared by many on social media platforms) compared to evidence portrayed as non-normative (Chapter 5.1) and that normativity cues signaled by large groups of people are the most effective at changing beliefs (Chapter 5.2). While the studies outlined in Part 2 focus on investigating phenomena at an individual level, extensive research shows that such cognitive processes are highly sensitive to the social context in which they manifest. Therefore, in Part 3, at the collective level of investigation, I uncover how macro-level societal outcomes can emerge from micro-level psychological processes. I first show how conversational interactions trigger belief change at a dyadic level, as individuals talking to one another change their beliefs to match those of their conversational partners (Chapter 6). Then, I find that collective level outcomes (i.e., collective beliefs) are influenced by both the conversational network structure that characterizes the community (Chapter 7) as well as by individual level mechanisms (Chapter 8). Finally, in Part 4 (Chapter 9) I propose future avenues of investigation, such as evaluating the connections between beliefs and behaviors, focusing on translating these controlled experimental strategies into applied settings. The goal of this translational work is to encourage a more active use of science in everyday life, for example, in developing actionable recommendations for policy makers and communicators to dispel misinformation at a societal level.

Vlasceanu, M., Morais, M. J., & Coman, A. (2021). Network Structure Impacts the Synchronization of Collective Beliefs. Journal of Cognition and Culture. Publisher's VersionAbstract
People’s beliefs are influenced by interactions within their communities. The propagation of this influence through conversational social networks should impact the degree to which community members synchronize their beliefs. To investigate, we recruited a sample of 140 participants and constructed fourteen 10-member communities. Participants first rated the accuracy of a set of statements (pre-test) and were then provided with relevant evidence about them. Then, participants discussed the statements in a series of conversational interactions, following pre-determined network structures (clustered/non-clustered). Finally, they rated the accuracy of the statements again (post- test). The results show that belief synchronization, measuring the increase in belief similarity among individuals within a community from pre-test to post-test, is influenced by the community’s conversational network structure. This synchronization is circumscribed by a degree of separation effect and is equivalent in the clustered and non- clustered networks. We also find that conversational content predicts belief change from pre-test to post-test.
Vlasceanu, M., & Coman, A. (2021). The Impact of Social Norms on Health-Related Belief Update. Applied Psychology: Heath and Well-Being. Publisher's VersionAbstract

People are constantly bombarded with information they could use to adjust their beliefs. Here, we are interested in exploring the impact of social norms on belief update. To investigate, we recruited a sample of 200 Princeton University students, who first rated the accuracy of a set of statements (pre-test). They were then provided with relevant evidence either in favor or against the initial statements, and they were asked to rate how convincing each piece of evidence was. The evidence was randomly assigned to appear as normative or non-normative, and also randomly assigned to appear as anecdotal or scientific. Finally, participants rated the accuracy of the initial set of statements again (post-test). The results show that participants changed their beliefs more in line with the evidence, when the evidence was scientific compared to when it was anecdotal. More importantly to our primary inquiry, the results show that participants changed their beliefs more in line with the evidence when the evidence was portrayed as normative compared to when the evidence was portrayed as non-normative, pointing to the impactful influence social norms have on beliefs. Both effects were mediated by participants’ subjective evaluation of the convincingness of the evidence, indicating the mechanism by which evidence is selectively incorporated into belief systems.

Vlasceanu, M., Morais, M. J., & Coman, A. (2021). The effect of prediction error on belief update across the political spectrum. Psychological Science. Publisher's VersionAbstract

Making predictions is an adaptive feature of the cognitive system, as prediction errors are used to adjust the knowledge they stemmed from. Here, we investigate the effect of prediction errors on belief update in an ideological context. In Study 1, 704 Cloud Research participants first evaluated a set of beliefs, then either made predictions about evidence associated with the beliefs and received feedback or were just presented with the evidence. Finally, they re-evaluated the initial beliefs. Study 2, which involved a US census matched sample of 1073 Cloud Research participants, replicated Study 1. We find that the size of the prediction errors linearly predicts belief update and that making large errors leads to more belief update than not engaging in prediction. Importantly, the effects hold for both Democrats and Republicans across all belief types (Democratic, Republican, Neutral). We discuss these findings in the context of the misinformation epidemic.

Aladesuru*, B. H., Cheng*, D., Harris*, D., Mindel*, A., & Vlasceanu, M. (2020). To Treat or Not to Treat: The Impact of Hairstyle on Implicit and Explicit Perceptions of African American Women’s Competence. Open Journal of Social Sciences. Publisher's VersionAbstract

African American women wearing their natural Afrocentric hair without altering its texture have long been discriminated against in the workplace, at school, in the military, in the justice system, and more. This phenomenon has been found to be mainly driven by the notion that African American women wearing their natural hair are less professional than African American women wearing chemically treated, Eurocentric hair. In prior work, dimensions such as perceived dominance, intelligence, and unpleasantness have been explored as potential mechanisms playing a role in the relationship between African American hair and perceived professionalism. Here, we explore an additional such dimension: perceived competence. In a sample of 186 predominantly Caucasian Cloud Research participants, we found that African American women wearing their natural Afrocentric hair were perceived both implicitly and explicitly as being less competent than African American women wearing Eurocentric hair, and that the implicit and explicit attitudes were not corre- lated. These findings are relevant to understanding barriers that may hinder African American women in their academic and professional careers.

Vlasceanu, M., Tachihara, K., Goldberg, A., & Coman, A. (2020). Lexical associations in a native and non-native language affect retrieval induced forgetting. CogSci. Publisher's VersionAbstract

Recent work suggests that speakers’ lexical networks in their native and secondary languages are organized somewhat differently, with native languages showing greater systematicity. We here test this claim in a new way, by making use of the “Retrieval-induced forgetting” effect (RIF). Specifically, practicing previously encoded information through rehearsal is expected to result in better memory for that information, regardless of which language the information is encoded. The RIF effect involves the suppression of information that is ​associated ​with the practiced information but is itself unpracticed. Since RIF is understood to rely on the association between the practiced and unpracticed memories, we predict it will be weaker when applied in a language with weaker or less systematically organized lexical associations. Results confirm that while the expected practice effect was evident in participants’ native and second languages, the RIF effect was only significant in participants’ native language. We discuss the relevance and implications of this finding for second language speakers.

Vlasceanu, M., Goebel, J., & Coman, A. (2020). The Emotion-Induced Belief Amplification Effect. CogSci. Publisher's VersionAbstract

Exposure to images constitutes a ubiquitous day-to-day experience for most individuals. From mass-media exposure, to engagement with social-networking sites, to educational contexts, we are bombarded with images. Here, we explore the effect that emotional images have on belief endorsement. To investigate this effect, we test whether statements accompanied by emotionally arousing images become more or less believable than the same statements when they are accompanied by neutral images or by no images. We find that emotional images increase statement believability (Experiment 1, replicated in preregistered Experiment 2). We discuss the implications of this finding in the context of interventions aimed at reducing misinformation.

Vlasceanu, M., Morais, M. J., Duker, A., & Coman, A. (2020). The Synchronization of Collective Beliefs: From Dyadic Interactions to Network Convergence. Journal of Experimental Psychology: Applied. Publisher's VersionAbstract

Systems of beliefs organized around religion, politics, and health constitute the building blocks of human communities. One central feature of these collectively held beliefs is their dynamic nature. Here, we study the dynamics of belief endorsement in lab-created 12-member networks using a 2-phase commu- nication model. Individuals first evaluate the believability of a set of beliefs, after which, in Phase 1, some networks listen to a public speaker mentioning a subset of the previously evaluated beliefs while other networks complete a distracter task. In Phase 2, all participants engage in conversations within their network to discuss the initially evaluated beliefs. Believability is then measured both post conversation and after one week. We find that the public speaker impacts the community’s beliefs by altering their mnemonic accessibility. This influence is long-lasting and amplified by subsequent conversations, resulting in community-wide belief synchronization. These findings point to optimal sociocognitive strategies for combating misinformation in social networks.

Vlasceanu, M., & Morais, M. J. (2019). A Possible Neural Mechanism of Intentional Forgetting . Journal of Neuroscience. Publisher's Version PDF
Vlasceanu, M. (2018). Mnemonic Influence:How Memory Impacts Emotion, Reason, and Action. Princeton University Department of Psychology. PDF
Vlasceanu, M., & Coman, A. (2018). Mnemonic accessibility affects statement believability: the effect of listening to others selectively practicing beliefs. Cognition , 180 (November), 238-245. Publisher's VersionAbstract

Belief endorsement is rarely a fully deliberative process. Oftentimes, one’s beliefs are influenced by superficial characteristics of the belief evaluation experience. Here, we show that by manipulating the mnemonic accessibility of particular beliefs we can alter their believability. We use a well-established socio-cognitive paradigm (i.e., the social version of the selective practice paradigm) to increase the mnemonic accessibility of some beliefs and induce forgetting in others. We find that listening to a speaker selectively practicing beliefs results in changes in believability. Beliefs that are mentioned become mnemonically accessible and exhibit an increase in believability, while beliefs that are related to those mentioned experience mnemonic suppression, which results in decreased believability. Importantly, the latter effect occurs regardless of whether the belief is scientifically accurate or inaccurate. Furthermore, beliefs that are endorsed with moderate-strength are particularly susceptible to mnemonically-induced believability changes. These findings, we argue, have the potential to guide interventions aimed at correcting misinformation in vulnerable communities.

Vlasceanu, M., Enz, K., & Coman, A. (2018). Cognition in a social context: A social-interactionist approach to emergent phenomena. Current Directions in Psychological Sciences. Publisher's VersionAbstract


The formation of collective memories, emotions and beliefs is a fundamental characteristic of human communities. These emergent outcomes are thought to occur due to a dynamical system of communicative interactions among individuals. But despite recent psychological research on collective phenomena, no programmatic framework to explore the processes involved in their formation exists. Here, we propose a social-interactionist approach that bridges cognitive and social psychology to illuminate how micro-level cognitive phenomena give rise to large-scale social outcomes. It involves first establishing the boundary conditions of cognitive phenomena, then investigating how cognition is influenced by the social context in which it is manifested, and finally studying how dyadic-level influences propagate in social networks. This approach has the potential to (1) illuminate the large-scale consequences of well- established cognitive phenomena, (2) lead to interdisciplinary dialogues between psychology and the other social sciences and (3) be more relevant for public policy than existing approaches.


Vlasceanu, M., Drach, R., & Coman, A. (2018). Suppressing my memories by listening to yours: The effect of socially triggered context-based prediction error on memory. Psychonomic Bulletin & Review. Publisher's VersionAbstract
The mind is a prediction machine. In most situations, it has expectations as to what might happen. But when predictions are invalidated by experience (i.e., prediction errors), the memories that generate these predictions are suppressed. Here, we explore the effect of prediction error on listeners’ memories following social interaction. We find that listening to a speaker recounting experiences similar to one’s own triggers prediction errors on the part of the listener that lead to the suppression of her memories. This effect, we show, is sensitive to a perspective-taking manipulation, such that individuals who are instructed to take the perspective of the speaker experience memory suppression, whereas individuals who undergo a low-perspective-taking manipulation fail to show a mnemonic suppression effect. We discuss the relevance of these findings for our understanding of the bidirectional influences between cognition and social contexts, as well as for the real-world situations that involve memory-based predictions.