Keynotes

Lisa Argyle [↑]

Arti-“fickle” Intelligence: Using LLMs as a Tool for Inference in the Political and Social Sciences

Chair: TBA

July 22, 4:30pm–6:00pm

Abstract

Generative large language models (LLMs) are incredibly useful, versatile, and promising tools. However, they will be of most use to political and social science researchers when they are used in a way that advances understanding about real human behaviors and concerns. To promote the scientific use of LLMs, we suggest that researchers in the political and social sciences need to remain focused on the scientific goal of inference. To this end, we discuss the challenges and opportunities related to scientific inference with LLMs, using validation of model output as an illustrative case for discussion. We propose a set of guidelines related to establishing the failure and success of LLMs when completing particular tasks, and discuss how we can make inferences from these observations. We conclude with a discussion of how this refocus will improve the accumulation of shared scientific knowledge about these tools and their uses in the social sciences.

Bio

Dr. Lisa P. Argyle is currently an Assistant Professor of Political Science at Brigham Young University. Starting in Fall 2025, she will begin at Purdue University as an Associate Professor of Political Science. She earned a Ph.D. from the University of California, Santa Barbara, after which she was a postdoc at Princeton University. Dr. Argyle blends political psychology with computational social science to study political attitudes and participation in the United States. Since the release of GPT-3 in 2020, a primary focus of her research has been generative AI, exploring how we can use Large Language Models as a tool to improve social science research and democratic societies. Her goal is to use surveys, experiments, and artificial intelligence tools to better understand how people talk about politics in their everyday lives, and how to improve those conversations.

Kathleen M. Carley [↑]

Computational analysis of social and organizational systems

Chair: TBA

July 22, 9:00am–10:30am

Abstract

The US National Academy in its decadal survey of the Social Sciences argued that today we are in an era of Networks+. In other words, rarely do people use social network or network science techniques in isolation; rather, these are augmented with other computationally based techniques such as computational linguistics and machine learning. In the area of social-cybersecurity, a field concerned with identifying, characterizing and mitigating online harms such as undue influence, disinformation, hate-speech and extremism, network science is frequently combined with various AI techniques to build more effective tools to support interventions, to improve imperial analysis, and to simulate behavior. A particular area of concern is influence, that is "who is influencing whom on line, how and to what effect?". The BEND framework has been proposed as a way of systematically characterising such information operations, and it has been operationalized using network science + AI techniques in a way that enables the analyst in detecting, characterizing, mitigating, and assessing the impact of influence activities, those being influenced, and the influencer. In this presentation the BEND framework and its operationalization using network science+AI will be described. Then the insights that can be drawn using this approach will be illustrated using data from various events around the world. The presentation with a call for future research and an itemization of limitations and gaps where new techniques and additional research are needed.

Bio

Dr. Kathleen M. Carley is a Professor of Computer Science in the Software and Societal Systems Department in the School of Computer Science at Carnegie Mellon University, IEEE Fellow, and Director of the Center for Computational Analysis of Social and Organizational Systems (CASOS) and Director of the center for Informed DEmocracy And Social‐cybersecurity (IDeaS) both at Carnegie Mellon University. She joined Carnegie Mellon in 1984 as Assistant Professor Sociology and Information Systems. In 1990 she became Associate Professor of Sociology and Organizations, in 1998 Professor of Sociology, Organizations, and Information Technology, and in 2002, attained her current role as Professor of Computation, Organization, and Society. She is also the CEO of Carley Technologies Inc. aka Netanomics.

Dean Eckles [↑]

Effect sizes and decisions

Chair: TBA

July 22, 9:00am–10:30am

Abstract

Social scientists often aim to not only determine whether some effect exists, but to quantify it. This often requires much larger or otherwise much more informative studies. It also requires choices of how to quantify effects. What relationship do various such quantifications have to scientific knowledge and, of particular relevance in applied work, to decisions? Using some examples from our recent experiments and other prominent, recent studies, I highlight how some effect size measures are typically decision-irrelevant and do not facilitate gaining generalizable knowledge. I also examine how we can use relevance to decisions to help select quantities of interest in the context of interventions in networks.

Bio

Dean Eckles is a social scientist and statistician. At the Massachusetts Institute of Technology, Dean is the William F. Pounds Professor in the MIT Sloan School of Management and associate director of the Institute for Data, Systems & Society in the Schwarzman College of Computing. Much of his research examines how interactive technologies affect human behavior, especially by mediating social influence. Dean also works on methods for inferring cause–effect relationships and on applied statistics more generally. He is a co-organizer of the Conference on Digital Experimentation (CODE@MIT). He was previously a scientist at Facebook and Nokia. Dean completed five degrees, including his PhD, at Stanford University.

Amir Goldberg [↑]

The Sociology of Interpretation: A Computational Approach

Chair: TBA

July 23, 4:30pm–6:00pm

Abstract

Culture shapes how we make sense of the world. People occupying different cultural positions often interpret the same realities in different ways. Recent advances in computational linguistics now make it possible to systematically measure these divergent interpretations at scale. In this talk, I introduce the Categorization-Association model of interpretation—a framework for capturing shared interpretation as well as processes of interpretive coordination and divergence. Drawing on computational analyses across multiple cultural domains—from American politics to popular understandings of leisure, work, and artificial intelligence—I show how culture reveals a fractured or "broken" geometry of meaning. This work illustrates how computational methods can help deepen our sociological understanding of culture, while underscoring the importance of grounding computational social science in robust theory.

Bio

Amir Goldberg is a Professor of Organizational Behavior and (by courtesy) Sociology at the Stanford Graduate School of Business, where he is the founding co-director of the Computational Culture Lab. His work draws on computational methods to measure and model culture, and its evolution, in organizations, markets, and beyond. He is currently serving as the Organizations Department Editor at Management Science.

Laura Nelson [↑]

Why Qualitative Research Needs Computational Social Science

Chair: TBA

July 23, 9:00am–10:30am

Abstract

Computational methods are playing a larger role in qualitative social science, but often with the goal of scaling or automating qualitative inquiry. Scaling and automating approaches are rooted in quantitative traditions, which are focused on measurement, generalizability, and evaluation against fixed benchmarks. But qualitative research is grounded in a different logic. It aims to understand meaning, context, and how people experience the world differently depending on their social position. This talk argues that if we want to use computational methods for qualitative research, we need to develop them differently than the current approach. For example, while machine learning is often framed as a tool for classification, its strength in identifying emergent patterns in large amounts of qualitative data aligns with inductive and abductive reasoning long used in qualitative research. Evaluating classification algorithms against a pre-determined "ground truth" stunts their development for inductive/abductive exploration and qualitative logics. In this talk I will share examples of how computational tools can support qualitative goals—not by making qualitative work more quantitative, but by shifting how we build, evaluate, and use these tools. This includes recognizing multiple perspectives in the data, treating researchers as active interpreters, and developing criteria that reflect qualitative rather than quantitative values.

Bio

Laura K. Nelson is an associate professor of sociology at the University of British Columbia, where she also directs the Centre for Computational Social Science. She uses computational methods to study social movements, gender, culture, and institutions, and to advance qualitative computational text analysis methods. She has published in outlets such as American Journal of Sociology, American Sociological Review, and Sociological Methods & Research, among others.

Brandon Stewart [↑]

Design-Based Supervised Learning: A General Framework for Using LLM Annotations and Other Predicted Variables in Downstream Analyses

Chair: TBA

July 22, 4:30pm–6:00pm

Abstract

TBA

Bio

TBA

Arnout van de Rijt [↑]

Luck and success in millions of life courses

Chair: TBA

July 23, 4:30pm–6:00pm

Abstract

This talk probes the role luck and success in the life course. We ask whether when people get lucky the trajectory of their life success increasingly diverges from that of their unlucky counterpart. We study this question theoretically using basic models of positive feedback. Empirically we look at the lives of thousands of Americans tracked in panel survey data and millions of Swedes captured in register data. We focus on income as measure of success and study a host of different ways by which people might be lucky or unlucky at different stages of the life course. We use causal identification strategies to isolate pairs of egos and alteregos who led similarly lives before the event and then compare their life courses afterwards.

Bio

Arnout van de Rijt is Professor of Sociology at the European University Institute (EUI). Van de Rijt holds a PhD in Sociology from Cornell University (2007). Van de Rijt is Editor-in-Chief of Sociological Science, an open-access journal for social scientists committed to advancing a general understanding of social processes. He is President of the International Network of Analytical Sociologists (INAS), which bring together scholars with a common interest in social mechanisms and micro-macro dynamics. Van de Rijt is elected member and council member of the European Academy of Sociology (EAS), a fellowship with the aim to advance excellence in sociological scholarship. His research spans the broad areas of social network analysis, computational social science, collective action and social stratification. Van de Rijt received the 2010 Lynton Freeman award from the International Network for Social Network Analysis (INSNA) for his contributions to social network analysis. He received the 2017 Raymond Boudon award for early career achievement from the EAS for his work on cumulative advantage.

Duncan Watts [↑]

Integrating explanation and prediction in computational social science

Chair: TBA

July 24, 9:00am–10:30am

Abstract

TBA

Bio

TBA

Sarah Williams [↑]

Urban policy and big data for public good

Chair: TBA

July 24, 4:30pm–6:00pm

Abstract

Williams will explain how we can use data as a tool for empowerment rather than oppression, something Williams calls "Data Action," which is also the title of her recent book. Data Action seeks to provide guidance for using data toward the benefit of society, learning from the ways we have used data unethically in the past and illustrating ways we can use it more ethically and creatively in the future. Williams will illustrate the seven Data Action principles through her diverse research projects spanning topics of Central American migration, popular transit in Africa, ghost cities in China, and translating New York City’s zoning text. Williams will also show her most recent work on People Powered AI: How to use Gen AI for Civic Engagement.

Bio

Sarah Williams is an Associate Professor of Technology and Urban Planning at the Massachusetts Institute of Technology (MIT) where she is also Director of the Civic Data Design Lab and the Leventhal Center for Advanced Urbanism. Williams’ combines her training in computation and design to create communication strategies that expose urban policy issues to broad audiences and create civic change. She calls the process Data Action, which is also the name of her recent book published by MIT Press. Williams is co-founder and developer of Envelope.city, a web-based software product that visualizes and allows users to modify zoning in New York City. Before coming to MIT, Williams was Co-Director of the Spatial Information Design Lab at Columbia University’s Graduate School of Architecture Planning and Preservation (GSAPP). Her design work has been widely exhibited including work in the Guggenheim, the Museum of Modern Art (MoMA), Venice Biennale, and the Cooper Hewitt Museum. Williams has won numerous awards including being named one of the top 25 technology planners and Game Changer by Metropolis Magazine. Check out her latest exhibition, Visualizing NYC 2021, at the Center for Architecture in New York City.