Daniel Karell

Generative AI is an advancement of machine learning technologies, offering sociologists new methods for understanding social life while intervening in a society already shaped by algorithmic systems. Yet one feature sets it apart from earlier computational tools and also makes it distinctly sociological: its interactivity. These systems are trained on social exchanges, refined by human feedback, and used through dialogue and delegation, so that interaction runs throughout their development and use. We conceptualize this interactivity and introduce a framework for understanding how generative AI intervenes in social processes across individual, organizational, and institutional levels. The framework describes how generative AI expands users, functions as a proxy for society, and expresses power, in sociological practice and in society. We conclude with three dynamics that will challenge sociology in the near future: the transformation of scientific practice, the reorganization of social interaction, and the acceleration of algorithmic governance.

Although data centers have rapidly emerged as a political flashpoint, there is little systematic evidence to date on whether their arrival has electoral consequences. We link the precise locations of data centers in Virginia -- which has the United States' largest concentration of such facilities -- to voters' geocoded residential addresses and validated turnout histories, allowing us to compare individual-level changes in turnout between the state's 2021 and 2025 elections according to voters' proximity to operating data centers. Across two independently assembled measures of data-center openings, voters living near an active facility became more likely to vote in 2025. The estimated increase is approximately 0.4 to 0.6 percentage points among voters living within 2.5 to 5 kilometers of a facility. Estimates remain positive at farther distances, although they become less precise. We also find suggestive evidence that the increase in turnout is most concentrated among non-affiliated registered voters and in some regions with many facilities. These results offer insights for states with high numbers of data centers, like Ohio and Texas, and indicate that data-center development was already influencing political participation before the 2026 midterm elections.

How can prediction markets affect politics? While traditional markets have long influenced politics, we argue that some of prediction markets' unique features allow them to affect political speech through distinct forms of manipulation, fulfillment, and legtimization. We then test for the possibility that one or more of these mechanisms are operating by analyzing the relationship between "mention markets", which allow traders to wager on specific words spoken by political elites, and public comments by Donald Trump and the Chair of the Federal Reserve. We find that words introduced into mention markets become significantly more likely to appear in Trump's remarks during the days following market creation, with larger effects for markets when there is higher trading volume. By contrast, we detect no comparable effects in Federal Reserve press conferences. These results suggest that prediction markets are playing a role in shaping public political speech when financial stakes are high and speech is less constrained by accountability. The findings also encourage further research directly examining the mechanisms linking prediction markets to political speech, as well as other political outcomes.

Large language models (LLMs) can be used to persuade people on a range of issues, particularly through user-driven strategies such as personalizing messages and dialogues intended to change minds. However, their capacity to influence opinions through subtle, latent ideological framing remains relatively understudied. We investigate whether AI-generated historical summaries affect social and political opinions through a preregistered experiment (N = 1,912). Participants read Wikipedia or GPT-4o summaries of two historical events, with AI summaries maintaining factual accuracy while exhibiting different types of framing biases. Default AI summaries led to more liberal opinions compared with Wikipedia, demonstrating the persuasive capability of LLM's latent biases. Summaries purposefully induced with a liberal framing also led to more liberal opinions, regardless of readers’ ideologies. Summaries constructed with a conservative framing produced conservative shifts primarily among conservative readers. These findings demonstrate that the use of AI for learning history can influence opinions through both intrinsic and intentional framing mechanisms, even when the content remains factually accurate. As AI becomes integral to information acquisition, recognizing pathways of influence based not only on user-manipulated content but also on models’ latent biases is essential for understanding AI's broader societal impacts.

Many people now use AI chatbots to obtain summaries of complex topics, yet we know little about how this affects knowledge acquisition, including how the effects might vary across different groups of people. We conducted two experiments comparing how well people recalled factual information after reading AI-generated or human-written historical summaries. Participants who read AI-generated summaries scored significantly higher on knowledge tests than those who read expert-written blog posts (Study 1) or Wikipedia articles (Study 2). These improvements were present regardless of whether readers knew the content was AI-generated or if the AI summaries were politically biased. Moreover, AI summaries improved recall across various demographic groups, including gender, race, income, education, and digital literacy levels. This suggets that using AI tools for everyday factual queries does not create new knowledge inequalities but could still amplify existing ones through differential access. Our findings indicate that the increasingly routine use of AI for information-seeking could enhance factual learning, with implications for education policy and addressing inequality.

Generative artificial intelligence (AI) offers new capabilities for analyzing data, creating synthetic media, and simulating realistic social interactions. This essay introduces a special issue that examines how these and other affordances of generative AI can advance social science research. We discuss three core themes that appear across the contributed articles: rigorous measurement and validation of AI-generated outputs, optimizing model performance and reproducibility via prompting, and novel uses of AI for the simulation of attitudes and behaviors. We highlight how generative AI enable new methodological innovations that complement and augment existing approaches. This essay and the special issue’s ten articles collectively provide a detailed roadmap for integrating generative AI into social science research in theoretically informed and methodologically rigorous ways. We conclude by reflecting on the implications of the ongoing advances in AI.

How does online speech affect offline attacks? While a growing literature has examined this link in right-wing violence in the West, much less is known about its importance in the religiously divided societies of the Global South. Furthermore, existing research has overwhelmingly focused on negative externalities of social media, while paying comparatively little attention to their conciliatory effects. We advance the scholarship in both of these areas by analyzing 22.4 million posts from Koo, an Indian social media network popular among India’s Hindu nationalists. We combine these data with information on attacks on religious minorities in India from 2020 through 2022. We find that the frequency of hashtags with a Hindu-chauvinist connotation are associated with increased attacks on Muslims and Christians. We also find that the frequency of hashtags alluding to the overcoming of religious divisions is associated with fewer attacks. These results survive a battery of robustness checks and supplemental tests. Additionally, the observed relationships disappear during exogenous Internet outages, consistent with the effect being driven by online speech. Importantly, since the content we study is not overtly aggressive and conveys values rather than factual claims, it does not classify as hate speech, misinformation, or disinformation. This suggests that the scholarly debate on what kinds of online speech influence offline harm has to be broadened and that censorship and fact-checking can fall short of addressing online speech’s negative consequences in religiously divided societies.

How do artificial intelligence (AI) tools affect people's knowledge? Existing scholarship supports contrasting expectations. Some research shows that people often find AI-generated ("synthetic") texts clearer, more accessible, and more enjoyable to read, which can foster learning from the texts. However, other research shows that synthetic texts are derivative and homogeneous, making them less interesting and thus potentially inimical to learning. In two experimental studies, we find that people who read AI-generated summaries of historical events more accurately recall factual information about the events compared to people who read human-written summaries of the same event. In the first study (N=193), participants who read synthetic text describing the Seattle General Strike of 1919 correctly answered more fact-based questions than participants who read a text written by experts. The preregistered second study, with a sample nationally representative of the United States (N=1907) and additionally testing a second historical event, replicates the results of the first study. Study 2 also indicates that synthetic historical summaries have a positive effect regardless of whether subjects know they were generated by AI, as well as whether or not the summaries exhibit political biases. Furthermore, while synthetic summaries improve knowledge for participants across all levels of educational attainment, the largest improvements occur at the lowest levels. Our findings show how AI-generated texts can enhance learning and reduce, rather than amplify, educational inequalities.

A large literature on international labor migration explores how to improve low-skilled migrants’ experience of pursuing and obtaining overseas employment. Much of this scholarship focuses on describing and mitigating difficult, and sometimes exploitative, conditions in the host country. Scholars have paid less attention to factors in home countries that may affect aspiring migrants’ experience of looking for overseas work. We help address this gap by conducting a field experiment in the high out-migration country of Pakistan to examine whether receiving new information about employment brokers and overseas opportunities improves aspiring migrants’ subjective and objective experience of the job-seeking process. After analyzing the effects of information on those who lack alternative sources of information, we report mixed findings. These highlight the need to think carefully about ways to improve aspiring migrants’ job-seeking experiences, especially with regard to the common assumption that if low-skilled migrants simply “knew more” — that is, had more relevant and accurate information — they would not allow themselves to have negative migration experiences; our findings suggest this is not necessarily always the case.