We propose and explore the possibility that language models can be studied as effective proxies for specific human subpopulations in social science research. Practical and research applications of artificial intelligence tools have sometimes been limited by problematic biases (such as racism or sexism), which are often treated as uniform properties of the models. We show that the “algorithmic bias” within one such tool—the GPT-3 language model—is instead both fine-grained and demographically correlated, meaning that proper conditioning will cause it to accurately emulate response distributions from a wide variety of human subgroups. We term this propertyalgorithmic fidelityand explore its extent in GPT-3. We create “silicon samples” by conditioning the model on thousands of sociodemographic backstories from real human participants in multiple large surveys conducted in the United States. We then compare the silicon and human samples to demonstrate that the information contained in GPT-3 goes far beyond surface similarity. It is nuanced, multifaceted, and reflects the complex interplay between ideas, attitudes, and sociocultural context that characterize human attitudes. We suggest that language models with sufficient algorithmic fidelity thus constitute a novel and powerful tool to advance understanding of humans and society across a variety of disciplines.

Exposure to news, opinion, and civic information increasingly occurs through social media. How do these online networks influence exposure to perspectives that cut across ideological lines? Using deidentified data, we examined how 10.1 million U.S. Facebook users interact with socially shared news. We directly measured ideological homophily in friend networks and examined the extent to which heterogeneous friends could potentially expose individuals to cross-cutting content. We then quantified the extent to which individuals encounter comparatively more or less diverse content while interacting via Facebook's algorithmically ranked News Feed and further studied users' choices to click through to ideologically discordant content. Compared with algorithmic ranking, individuals' choices played a stronger role in limiting exposure to cross-cutting content. Copyright © 2015, American Association for the Advancement of Science.

This study suggests one direction of theoretical and methodological coupling of communication research with the digital trace data, utilizing its differences from the traditional social science approach (e.g., sampling vs. population, normal distribution vs. power–law distribution, generalization vs. simulation, deductive vs. inductive, and perceived vs. actual). We propose specific examples of (i) combining communication research with trace data methodologically and theoretically; (ii) collaborating with linguistic psychology complemented with the automated content analysis and natural language processing techniques; and (iii) creating new theoretical inquiries by configuring the granular level of interactivity and underlying dynamics, observing the longitudinal change of interactions, and discovering the neglected presence of outliers and the invisibles. We expect the direction suggested by this study contributes to deepening our understanding of human communication behavior.

Self-reported measures of media exposure are plagued with error and questions about validity. Since they are essential to studying media effects, a substantial literature has explored the shortcomings of these measures, tested proxies, and proposed refinements. But lacking an objective baseline, such investigations can only make relative comparisons. By focusing specifically on recent Internet activity stored by Web browsers, this article's methodology captures individuals' actual consumption of political media. Using experiments embedded within an online survey, I test three different measures of media exposure and compare them to the actual exposure. I find that open-ended survey prompts reduce overreporting and generate an accurate picture of the overall audience for online news. I also show that they predict news recall at least as well as general knowledge. Together, these results demonstrate that some ways of asking questions about media use are better than others. I conclude with a discussion of survey-based exposure measures for online political information and the applicability of this article's direct method of exposure measurement for future studies.

\n Does content curation by Facebook introduce ideological bias?\n \n [Also see Report by\n \n Bakshy\n et al.\n \n ]\n \n

Does the consumption of ideologically congruent news on social media exacerbate polarization? I estimate the effects of social media news exposure by conducting a large field experiment randomly offering participants subscriptions to conservative or liberal news outlets on Facebook. I collect data on the causal chain of media effects: subscriptions to outlets, exposure to news on Facebook, visits to online news sites, and sharing of posts, as well as changes in political opinions and attitudes. Four main findings emerge. First, random variation in exposure to news on social media substantially affects the slant of news sites that individuals visit. Second, exposure to counter-attitudinal news decreases negative attitudes toward the opposing political party. Third, in contrast to the effect on attitudes, I find no evidence that the political leanings of news outlets affect political opinions. Fourth, Facebook’s algorithm is less likely to supply individuals with posts from counter-attitudinal outlets, conditional on individuals subscribing to them. Together, the results suggest that social media algorithms may limit exposure to counter-attitudinal news and thus increase polarization. (JEL C93, D72, L82)

This article presents a review of communication research on user-generated content with a special focus on studies which include a content analysis. The trends of research on this comparatively new and rapidly developing subject are systematically discussed and desiderata are identified. The evaluation is based on a content analysis of pertinent approaches in nine relevant international peer-reviewed journals published from 2004 to 2012. From the results, the article concludes that user-generated content is approached by scholars from a variety of perspectives and offers scope for interdisciplinary cooperation but also notes that several of the challenges posed by the continuously changing nature of the content are not fully met.

There is widespread public and academic interest in understanding the uses and effects of digital media. Scholars primarily use self-report measures of the quantity or duration of media use as proxies for more objective measures, but the validity of these self-reports remains unclear. Advancements in data collection techniques have produced a collection of studies indexing both self-reported and log-based measures. To assess the alignment between these measures, we conducted a pre-registered meta-analysis of this research. Based on 106 effect sizes, we found that self-reported media use correlates only moderately with logged measurements, that self-reports were rarely an accurate reflection of logged media use and that measures of problematic media use show an even weaker association with usage logs. These findings raise concerns about the validity of findings relying solely on self-reported measures of media use.

\r\nThere have been growing concerns regarding the potential impact of social media on democracy and public debate. While some theorists have claimed that ICTs and social media would bring about a new independent public sphere and increase exposure to political divergence, others have warned that they would lead to polarization through the formation of echo chambers. The issue of social media echo chambers is both crucial and widely debated. This article attempts to provide a comprehensive account of the scientific literature on this issue, shedding light on the different approaches, their similarities, differences, benefits, and drawbacks, and offering a consolidated and critical perspective that can hopefully support future research in this area. Concretely, it presents the results of a systematic review of 55 studies investigating the existence of echo chambers on social media, providing a first classification of the literature and identifying patterns across the studies’ foci, methods and findings. We found that conceptual and methodological choices influence the results of research on this issue. Most importantly, articles that found clear evidence of echo chambers on social media were all based on digital trace data. In contrast, those that found no evidence were all based on self-reported data. Future studies should take into account the possible biases of the different approaches and the significant potential of combining self-reported data with digital trace data.\r\n

This article addresses questions of ideological polarization and the filter bubble in social media. It develops a theoretical analysis of ideological polarization on social media by considering a range of relevant factors. Over recent years, fake news and the effect of the social media filter bubble have become of increasing importance both in academic and general discourse. The article reviews the assumption that algorithmic curation and personalization systems place users in a filter bubble of content that decreases their likelihood of encountering ideologically cross-cutting news content. At the intersection of new media, politics and behavioural science, the article establishes a theoretical framework for further research and future actions by society, policymakers and industries.