Repository for the Spring 2026 Computational Social Science Workshop
Time: 11:00 AM to 12:20 PM, Thursdays Location: Room 142, 1155 E. 60th St
Amit Goldenberg is an assistant professor in the Negotiation Organization & Markets unit, an affiliate with Harvard’s Department of Psychology, and a member of the Digital Data and Design Institute (D^3). Professor Goldenberg's research focuses on what makes people emotional in social and group contexts, and how such emotions can be changed when they are unhelpful or undesired. He is particularly interested in how technology is used for both emotion detection and regulation.
In his work Professor Goldenberg tries to integrate between experimental psychology that examines behavior at the individual and collectives levels. To integrate these domains, he employs a multi-method approach that combines behavioral experiments, analysis of data from digital media, computational modeling and AI.
Exploring the Impact and Limits of AI Emotional support
Large language models demonstrate impressive social-emotional abilities. Yet, as we deploy these capabilities and brace for their profound impact on our lives, it is crucial to further understand the nature and limitations of AI's emotional support. In my talk, I aim to explore how LLMs such as GPT-4 compare to humans in providing emotional support and to examine the main reasons behind these perceived differences. I will also discuss how the quality of emotional support is influenced by the awareness that it comes from AI. Finally, I will consider why people may choose to share emotional experiences with humans or with AI, what drives these preferences, and how they impact satisfaction and subsequent interactions. These findings begin to establish a broader understanding of human-AI emotional support.
Gary King is the Albert J. Weatherhead III University Professor at Harvard University – one of 25 with Harvard’s most distinguished faculty title – and Director of the Institute for Quantitative Social Science. King develops and applies empirical methods in many areas of social science, focusing on innovations that span the range from statistical theory to practical application.
King is an elected Fellow in 8 honorary societies (National Academy of Sciences, American Statistical Association, American Association for the Advancement of Science, American Academy of Arts and Sciences, Society for Political Methodology, National Academy of Social Insurance, American Academy of Political and Social Science, and the Guggenheim Foundation) and has won more than 55 prizes and awards for his work. King was elected President of the Society for Political Methodology and Vice President of the American Political Science Association. He has been a member of the Senior Editorial Board at Science, Visiting Fellow at Oxford, and Senior Science Adviser to the World Health Organization. He has written more than 190 journal articles, 30 open source software packages, and 8 books.
King proposed the now widely accepted standard for fairness in legislative redistricting known as “partisan symmetry,” and the methods used by courts and parties to detect when partisan gerrymandering violates it. His “ecological inference” methods for inferring individual behavior from aggregate data are used in most jurisdictions applying the Voting Rights Act to detect racial gerrymandering. His book with Keohane and Verba, Designing Social Inquiry, helped launch the modern subfield of qualitative methods in political science; his book Unifying Political Methodology had a similar role for quantitative political methodology. His “Replication, Replication” article helped initiate the data sharing movement in political science, and his ongoing international “Dataverse” project supports the movement across fields. His “anchoring vignettes” approach to cross-cultural survey comparability has been used in more than 100 countries by researchers, governments, and others. King has pioneered “politically robust” research designs that make possible unusually large randomized experiments in politically difficult circumstances – including the largest ever randomized health policy experiment, to evaluate the Mexican universal health insurance program, and the only large scale randomized news media experiment in the US. He has reverse engineered Chinese censorship and fabrication of social media posts, improved Social Security Trust Fund forecasts, and developed empirical methods and software widely used in academia, government, and private industry for automated text analysis, rare events, missing data, measurement error, causal inference, interpreting statistical results, and for forecasting elections, mortality rates, and international conflict.
King’s work is widely read across scholarly fields and beyond academia. He was listed as the most cited political scientist of his cohort; among the group of “political scientists who have made the most important theoretical contributions” to the discipline “from its beginnings in the late-19th century to the present”; and on lists of the most highly cited researchers across the social sciences. King’s many former students and postdocs now hold positions at leading universities and companies around the world. He has collaborated with hundreds of scholars, including many of his students, on research for publication. He has served on more than 30 editorial, nonprofit, and corporate boards; as founding editor of The Political Methodologist, and on the governing councils of the American Political Science Association, Inter-university Consortium for Political and Social Research, Society for Political Methodology, Midwest Political Science Association, Center for the Advanced Study in the Behavioral Sciences, and Institute for Data, Science, and Society.
King is also co-founder and an inventor of the original technology for Crimson Hexagon (merged with Brandwatch, acquired by Cision), Learning Catalytics (acquired by Pearson), Thresher (acquired by Two Six Technologies, a Carlyle Company), OpenScholar (acquired by Monomyth Group), Perusall, and QuickCode. He has received 17 patents for these technologies.
King is a proud graduate of SUNY New Paltz (B.A., 1980) and the University of Wisconsin-Madison (M.A., Ph.D., 1984). He taught at NYU for three years before coming to Harvard in 1987.
Who's to Blame for Survey Instability: Respondents with Nonexistent Preferences or Researchers with Flawed Measures?
Neither. For at least 75 years, survey researchers have found that about 25% of respondents give different answers when asked the same question twice (even if no material changes occur and respondents do not remember being asked the first time). This “survey instability” problem casts doubt on a vast research enterprise spanning large areas of academia and industry, is core to many ongoing substantive debates, and requires a resolution for proper survey design and analysis methods. We collect a wide variety of observational and experimental evidence, including 59 unique surveys. We first show that instability barely drops after accounting for both existing explanations, i.e., when respondents have fixed knowledge of their preferences and researchers use high quality, unbiased survey instruments. We trace a large component of survey instability to a different source recognized only in fields with non-survey measurement instruments — intrinsic human stochasticity. We then trace the decision making, cognitive, psychological, and individual characteristic precursors of this stochasticity and reveal their wide ranging implications for understanding respondents, avoiding inattention, designing surveys, and building statistical analysis methods.
Reading List
Diyi Yang is an assistant professor in the Computer Science Department at Stanford University, also affiliated with the Stanford NLP Group, Stanford HCI Group and Stanford Human Centered AI Institute. Her research focuses on human-centered natural language processing and human-AI interaction. She is a recipient of IEEE “AI 10 to Watch” (2020), Microsoft Research Faculty Fellowship (2021), NSF CAREER Award (2022), an ONR Young Investigator Award (2023), and a Sloan Research Fellowship (2024). Her work has received multiple paper awards or nominations at top NLP and HCI conferences. She received her Ph.D. from Carnegie Mellon University, as well as a B.S. from Shanghai Jiao Tong University.
The Human Half of Human-AI Interaction
Recent advances in large language models (LLMs) have transformed human-AI interaction; however, much research has focused on the AI half, scaling models and improving benchmarks, while the human half, how people use, are shaped by, and can be modeled with these systems, remains an open challenge. In this talk, we first examine the rise of AI companions, studying how users form relationships with LLMs and how this shapes their well-being. We then present CARE, an LLM-based training system that upskills novice counselors through simulated patient practice and structured feedback. We further introduce scaling laws for social simulation, characterizing when LLMs can faithfully simulate human behaviors and where they fall short. Together, these projects highlight how studying, designing, and simulating human-AI interaction can guide the development of AI systems that meaningfully support people.
Reading List
- Can LLM-Simulated Practice and Feedback Upskill Human Counselors? A Randomized Study with 90+ Novice Counselors
- The Rise of AI Companions: How Human-Chatbot Relationships Influence Well-Being
Nikhil Garg is an assistant professor of operations research and information engineering at Cornell Tech, the Jacobs Technion-Cornell Institute, and Cornell Engineering. He uses algorithms, data science, and mechanism design approaches to study democracy, markets, and societal systems at large.
Garg has received the NSF CAREER Award, the INFORMS George Dantzig Dissertation Award, an honorable mention for the ACM SIGecom Dissertation Award, and a Forbes 30 under 30 for Science. He has also received several other best paper awards.
He received his M.S. and Ph.D. from Stanford University, as well as a B.S. and B.A. from the University of Texas at Austin. He has spent considerable time collaborating with government agencies and non-profits, including the New York City Department of Parks and Recreation, New York City Public Schools, New York Public Library, and Uber.
In addition, Garg’s work has been featured in publications such as the New York Times, the Washington Post, Science Magazine, and Smithsonian Magazine.
Recommendations in the NYC High School Match (and other high-stakes settings)
Algorithmic systems are now used in high-stakes settings, including to help find jobs, schools, and therapists. Building public interest recommender systems in such settings bring both individual-level (enabling exploration, diversity, data quality) and societal (fairness, capacity constraints, monoculture) challenges. I will talk about an ongoing collaboration with the NYC Public Schools, in which we designed and deployed an informational intervention to help students from underserved middle schools discover high-performing, nearby high schools where they have a strong individual admissions likelihood. However, recommending specific programs brings a methodological challenge, congestion: if many applicants are recommended the same program, affecting admissions likelihoods, then the recommendations may be self-defeating. Time permitting, I'll also overview other directions in tackling such challenges, including (a) algorithmic monoculture and LLM homogeneity, (b) a platform to help discharge patients to long-term care facilities, and (c) feed ranking algorithms on Bluesky for research paper recommendations and an analysis on how scientists communicate on social media.
Reading List
- Correlated Errors in Large Language Models
- Paper Skygest: Personalized Academic Recommendations on Bluesky
Shiping Tang is one of Asia’s most influential and innovative social scientists. He was elected as one of the three vice-presidents (2025-26) of the International Studies Association (ISA). He is the first Chinese scholar to be elected to this position. He has published five single-authored volumes so far. In 2024, he was honored as one of the three Distinguished Scholars at the Global IR Section (GIRS) at the ISA Convention in San Francisco, along with Barry Buzan and Cristina Rojas. He has published five single-authored volumes so far. His more recent two books include: The Institutional Foundation of Economic Development, (Princeton University Press, 2022) and On Social Evolution: Phenomenon and Paradigm (Routledge, 2020).
Prof. Tang has a very broad research interest and has published widely, covering international relations, comparative politics, institutional economics, methodology, philosophy of the social sciences, political theory and sociology. He has also developed powerful platforms for complex decision making based on Computational Social Sciences (CSS).
Polarization versus the Median Voter Theorem: An Agent-based Modeling Simulation
Many Western democracies have experienced a rising tide of political polarization. While existing studies have singled out some key drivers, they have yet to pose a more fundamental question: does the “median voter theorem (MVT)”, a central pillar of classical theories of democracy, always hold? This article argues that the MVT critically rests upon three unrealistic assumptions: “rational” voters, a two-party system, and a single dividing issue among voters and parties. Because political systems in the real world violate the three assumptions and political competition as a dynamic process almost inevitably changes and transforms a political system, MVT premised on the three assumptions may not hold in the real world. We then use agent-based modeling (ABM) simulation to simulate possible outcomes of various political systems. Our ABM exercises show that MVT only holds when there is only a single dividing issue (e.g., income). With more than one dividing issues, political competition almost inevitably drives a two-party system into polarization. In contrast, a three-party system is less prone to political polarization, even with more than one dividing issues, ceteris paribus. Our discussion holds critical implications for understanding and tackling political polarization in both mature and new democracies.
Yian Yin is an Assistant Professor of Information Science at Cornell University. His research interests lie at the intersection of network science and computational social science, with a particular focus on the science of science. He applies and develops novel computational tools to understand how individual, social, and environmental processes independently and jointly promote (or inhibit) scientific progress and innovation achievements.
Scientific Production in the Era of Large Language Models: Early Evidence from Large-scale Preprint Data
The rapid adoption of AI across disciplines is reshaping the landscape of scientific production. While both enthusiasm and concern about generative AI in research are rising, systematic empirical evidence on the impact of large language models (LLMs) remains limited. In this talk, I draw on several large-scale analyses to examine how LLM use affects the productivity of individual scientists, reshapes attention to prior work, introduces hallucinated content into the scientific record, and creates new challenges for peer review. Taken together, these findings provide macro-level evidence on the impact of generative AI on science, highlighting the need for institutions, journals, funding agencies, and the broader public to rethink how scientific work should be evaluated in this new era.
Reading List
Chenhao Tan is an Associate Professor of Computer Science and Data Science at the University of Chicago, and directs the Chicago Human+AI Lab. He earned his PhD in Computer Science from Cornell University and dual bachelor's degrees in computer science and economics from Tsinghua University. His research focuses on human-centered AI, communication & intelligence, AI & Scientific Discovery, and AI alignment. His work has been covered by major news media outlets, including the New York Times and the Washington Post. He also won a Sloan research fellowship, an NSF CAREER award, an NSF CRII award, a Google research scholar award, research awards from Amazon, IBM, JP Morgan, and Salesforce, a Facebook fellowship, and a Yahoo! Key Scientific Challenges award.
Science in the Age of AI
As AI becomes increasingly capable of following instructions and conducting analyses, I believe that scientists will increasingly play the role of selector and evaluator. In this talk, I will introduce our recent work in building an ecosystem for the future of AI & Scientific Discovery. I will share our work in AI-enabled research evaluation and hypothesis generation. First, I will present ongoing work that formalizes the evaluation of research outcomes beyond the paper itself and use AI to conduct robust evaluation of research evaluation, with a case study on mechanistic interpretability. Second, rather than treating AI hallucinations as obstacles to eliminate, we leverage data and literature to steer AI creativity toward generating effective hypotheses. I will also introduce HypoBench, a dedicated benchmark for evaluating hypothesis generation, which reveals significant room for potential improvement of current AI models.
Reading List
- The Story is Not the Science: Execution-Grounded Evaluation of Mechanistic Interpretability Research
- HypoBench: Towards Systematic and Principled Benchmarking for Hypothesis Generation
- Literature Meets Data: A Synergistic Approach to Hypothesis Generation
Carolyn Rosé is the Kavčić-Moura Professor of Language Technologies and Human-Computer Interaction at Carnegie Mellon University. Her research advances the concept of Sociotechnical Artificial Intelligence from a highly multidisciplinary perspective, exploring human-AI complementarity and multi-agent human-AI teaming. Dr. Rosé's work bridges computational linguistics, sociolinguistics, and learning sciences to develop innovative AI systems that monitor, analyze, and support communication processes. Her contributions include advances in automated discourse analysis, conversational agent technologies, and multimodal modeling, with applications in education, health, and workplace collaboration.
His research centers on the diversity and evolution of languages. He wants to understand where the ~7,500 languages extant today come from (with a special emphasis in the last 12,000 years, the Holocene), how they will change in the advent of the human-machine era, and what is that languages have done to our species, our cognitions, behaviors, and cultures. He fully embraces a transdisciplinary and question-guided approach, drawing from data science, human biology, cognitive sciences, comparative linguistics, evolutionary anthropology, computational social sciences, natural language processing, and cultural evolution. A substantial proportion of his work involves inferences with small, sparse, incomplete, imbalanced, noisy and non-independent observational data.
LLM Agents as Facilitators of Effective Group Collaboration Processes
Supporting collaborative interaction is an ideal context for exploring the capabilities and limitations of LLM-based conversational agents. The ability to extract information in context and produce a coherent sounding text can be used to generate reflection triggers. In two recent studies, we have employed LLM-based conversational agents with the goal of triggering human reflection and learning during collaborative software design. As humans engage in collaborative design, they employ their own abilities to reason abstractly, to decompose problems, and apply principles productively. Reflection is a valuable activity for promoting human learning in these settings. However, what humans are able to do in terms of abstraction and reasoning as part of their creative problem solving is precisely what is most difficult for LLM agents to do. In contrast to claims of "super-human performance" in the media, in this talk we will explore the complementarity of human intelligence and Artificial Intelligence. We will begin with results of a classroom study where LLM-based conversational agent support for collaborative software development was successful in increasing student learning. From there we will move on to argue in favor of a research agenda for exploiting the complementarity both in terms of applying AI capabilities to the betterment of human learning as well as inspiring further extension of technical capabilities from insights derived from observation of human reflection and learning in collaborative design.
Reading List