Quantitative Social Science in the Age of AI
Preface
I propose to consider the question, “Can machines think?” This should begin with definitions of the meaning of the terms “machine” and “think”.
Turing (1950, 433)
Artificial intelligence (AI) is an overloaded term. To some “[i]t is the science and engineering of making intelligent machines, especially intelligent computer programs.” (McCarthy 2007). To others it is “the study of agents that receive percepts from the environment and perform actions” (Russell and Norvig 2021, vii). Still others, provocatively, say that AI “is neither artificial nor intelligent” (Crawford 2021, 8).
The lack of agreement on what AI means is reflected in a lack of agreement on how it should be used. Some are forced to use it, while others are adopting it enthusiastically. Even if there were agreement today, the rapid changes we have seen over the past few years mean this would soon be outdated.
Regardless of what it is called, or how it ought to be used, AI is re-shaping universities and society. The quantitative social sciences are no exception. All quantitative social scientists now have to contend with how it is changing our work.
Quantitative social science is a nebulous term that often encompasses the more applied researchers from economics, political science, communication, and sociology, to name a few. Imai (2017, 1) describes it as an interdisciplinary field built on the data and computational revolutions where “scholars analyze data to understand and solve problems about society and human behavior”.
Despite our different disciplinary training, many of us in the quantitative social sciences often have more in common with each other than with those in our own disciplines. We are united by a shared focus on writing code to analyze data, typically applying statistical methods, to answer questions. We are also often the ones teaching quantitative methods courses in our home departments and advising students on how to approach technical problems.
Some in the quantitative social sciences are brand new to AI—their first experience was winter 2026 when Claude Code became especially popular. Some quantitative social scientists got excited when ChatGPT was released in November 2022, or when GPT-3 was released in the middle of 2020. Still others were relatively early to what is now AI in earlier guises such as natural language processing or text-as-data. But very few of us have a deep history and comfort with AI, compared with, say, statistics. Having missed the earlier cycles of “AI winter” and “AI spring” (Mitchell 2021) many computational social scientists are now taken with recent AI progress. We should all remember the warning of Cantwell Smith (2019, 2) that “…respected scientists were serious [about the magnitude of the impending upheaval] 50 years ago, too”, but there are certainly a large number of changes that must be considered seriously.
Review, albeit from machines not peers, can now be obtained in seconds; entire research papers can be generated in a matter of minutes; and data cleaning that used to take weeks can be coded in hours. What to make of our hard-won coding and analytical skills? What is the purpose of a research paper? And while AI might be able to generate a paper in minutes, it might now take hours for a human to identify the error at the heart of it, because heuristics such as sloppy writing or code no longer apply. How do we move forward in such a world?
Science is not just what is in textbooks, it is also the shared agreement of the community of scientists. When on the research frontier that shared agreement may be about what questions are worth exploring and how to do so, and eventually, over the course of time, that becomes a shared understanding of what is correct and how to determine correctness. This edited volume is an attempt to explore the messy frontier of what AI means for quantitative social scientists and to share that exploration with others.
This collection emerged from a conference at the University of Toronto on this topic on 1 May 2026. Draft papers were presented and discussed. Over the next few months additional contributions were received, refinements were made, and the conversations sparked by this conference and collection continued.
The contributions in this edited volume are a marker of how quantitative social scientists from a diverse range of disciplines are thinking about AI in the middle of 2026. One is struck not only by the contributions themselves, but also by the depth of scholarship that they build on, and it is exciting to learn of so much new work. I look forward to seeing the community develop shared agreement over the coming years.
The edited collection outline is as follows:
Chapter 1 The Machines Talk Back: Modeling Turn-Taking Text as a Stochastic Process by AJ Alvero and Anna Seo Gyeong Choi focus on generative AI and the chat interface that has become so popular. Their stochastic process model of sequential text-based interactions will be important as this form of interaction with generative AI continues in importance.
Chapter 2 AI All The Way Down: Infinite Regress and the Future of Quantitative Social Science by Lisa P. Argyle, Ethan C. Busby, and Joshua R. Gubler introduces infinite regress, a concern that despite increased AI-enhanced output in society there will be decreased understanding and actual scientific and social benefit. They make the point that while this is not necessarily immediately disqualifying, it does need to be reckoned with more openly.
Chapter 3 New, but Normal? Evaluating AI’s Fitness for Purpose in Survey Research by Soubhik Barari, Trent D. Buskirk, and Joshua Y. Lerner re-imagine AI as normal methodology in survey research and subject to familiar standards of methodological justification. Such a re-imagining allows us to appreciate the strengths and appropriate use-cases of AI, while acknowledging the need for context-specific standards.
Chapter 4 As AI Lowers the Cost of Research, Adjudication and Attention Will Become the Bottlenecks by Ryan Briggs argues that adjudication, such as identifying whether a claim is coherent and properly tested will be initially made more difficult by AI. If, as he expects, AI progress increases the ease of doing research then the focus would shift toward adjudication and filtering.
Chapter 5 The Shifting Production Function: AI, Reproducibility, and the Future of Quantitative Social Science by Abel Brodeur and Bruno Barbarioli look at how AI may allow automated reproduction at scale. They highlight some of the concerns about changing norms and standards, and then propose a research agenda for the next five years, and a disclosure checklist that should be filled out when quantitative research is conducted with AI-assistance.
Chapter 6 All Models Are Wrong, But Some Are Trending by Mike Burnham identifies ways in which AI will create a fight for attention caused by changing filters. He identifies some particular opportunities including new forms of publication, and potential changes to tenure and hiring standards.
Chapter 7 Research Transparency and Collaboration in the Era of Generative AI Requires Open, Clean Code by Chris Cochrane and Michael Cowan focus on coding and the excitement of what is now possible to compute, balanced with concern about increased dependencies. Based on hard-won experience, they advocate for the adoption of not just open code, but open code that is also clean.
Chapter 8 Failing Faster, Learning Better: Agentic AI and Empirical Social Science Research by Charles Crabtree, Valentina González-Rostani, and Jae Yeon Kim examine how agentic AI can be used to produce better work. They consider innovation as a search problem, and then consider two failures: when promising ideas are screened out, and when weaker ideas are pursued. While they highlight the potential for agentic AI to help reduce these failures, they conclude with the importance of human judgment.
Chapter 9 A relational approach to agentic AI in social science research: Projects, roles, and tasks by Thomas Davidson provides a framework for how AI is used in the research process. He identifies different approaches and then discusses the implications of each, to compare and contrast.
Chapter 10 Fault Lines: Inequality and the Future of Quantitative Social Science in the Age of Artificial Intelligence by Alicia Eads and Fedor Dokshin introduce inequality into the conversation. They highlight different access, use, and returns, to AI, for different groups of researchers, and they emphasize that without deliberate choices, existing inequalities will persist and perhaps get worse.
Chapter 11 AI for Me, Not (Yet) for Thee? Desirable Difficulties and Deliberate Friction with LLMs by Andrew Heiss describes how AI creates a tension between being both a skeptic of AI in the learning process, while at the same time being a careful user in his own work. He highlights the importance of creating deliberate friction to ensure that experienced researchers maintain their expertise and that learners are able to develop their own skills.
Chapter 12 AI and the Measurement Imperative by Rehan Mirza and Sandra González-Bailón focuses on measurement as central to the social science mission to understand what is before defining what should be. They argue that the latter is a fundamentally human question and should not be outsourced to automated workflows. They also illustrate how social science can benefit from AI technologies while improving current benchmarks to better assess AI capabilities and the broader social impacts.
Chapter 13 AI Allows More Diversity in the Forms of Social Science by Kevin Munger is informed not just by his experience as a researcher, but also as an editor. He identifies the PDF as a bottleneck to research in social science, and then discusses how AI-enabled approaches could improve on it including to enable synthesis at scale, the potential for updating, and high-throughput exploration.
Chapter 14 Epistemic Standards and the Next Generation of Scholars by Andreea Musulan and Jean-François Godbout identify AI as decoupling research production from the development of a researcher’s capacity. They argue for the need for new methodological standards and training practices.
Chapter 15 Quantifying Culture from Embeddings to Generation by Laura K. Nelson and Tom Einhorn focus on the measurement of culture and begin with embeddings. They then trace this to transformer-based language models and highlight the paradox that language models allow for the analysis of cultural associations across a wider context than before, while at the same time being less tied to the situated perspectives through which culture is experienced and learned by humans.
Chapter 16 With Great Powers: A Practical Guide to Agentic AI for Social Science Research by Simone Paci discusses how the AI revolution has changed research practice, resulting in new inequality. He then builds a practical workflow based around researcher control and radical transparency.
Chapter 17 The Human Element in Social Science Research by Alexis Palmer looks at how to evaluate researchers and who selects into doing that research. She highlights that the current focus on producing papers needs to change.
Chapter 18 Cultivating Social Science Data in the Age of Agentic AI by Pratik Sachdeva and Dennis M. Feehan write about their excitement for the potential to more quickly and easily extract structured insights from large, unstructured data sources. They see this as raising the value of social science theory because these new data will engender new theory and highlight the continued importance of measurement.
Chapter 19 Optimal Editorial Screening Under Cheap Testing by Gabriel Sekeres takes seriously the challenge of a journal editor deciding which submissions to screen and publish. The key parameter examined is how their decision should change as the cost of conducting a test has decreased.
Chapter 20 Intelligent Machines Are Social Actors by Milena Tsvetkova thinks of AI as social actors, and then examines how quantitative social science should change in such a hybrid society. She concludes that an overhaul of quantitative social science training is needed.
Chapter 21 Curation and Reproducibility in an Artificial Intelligence World: Challenges and Solutions for Scientific Research by Lars Vilhuber considers the role of AI in legitimate scientific work, grappling with how to curate AI-supported research tools, input data and outputs. He focuses on the perspective of someone having to verify reproducibility and support the curation of research compendia. He argues that such challenges are not new, but that AI magnifies old challenges.
Chapter 22 On Benchmarks by Eva Vivalt focuses on benchmarks noting the increased demand for benchmarks in quantitative social science. She then notes that once a benchmark exists, it changes the incentives for researchers, and so not only do we need to be worried about building good benchmarks, but also how researchers will respond to the subsequent incentives.
Chapter 23 Computational Social Scientists Need to Start Caring About the Competitiveness of the AI Market by Patrick Y. Wu highlights the importance of seriously considering how many suppliers there are in the AI market, and that this is a primary methodological concern. For instance a consolidated market could result in correlated errors and a lack of independence between models.
Finally, I would like to thank David Grubbs and Soumya Jain at Chapman & Hall/CRC for publishing this collection.
Rohan Alexander
Toronto, Canada
Summer 2026
rohan.alexander@utoronto.ca