←
Thomas Davidson793 posts
Thomas Davidson avatar

Thomas Davidson

@thomasrdavidson

Assistant Professor of Sociology | Computational social science, politics, and AI

◷ Joined October 2016▧ 141 Photos and videos
Thomas Davidson avatar
Thomas Davidson@thomasrdavidson·15 Dec 2025

New paper in Nature Human Behaviour. I use a conjoint experiment to evaluate the capabilities of the latest models for context-sensitive moderation and compare the results with those from human subjects, demonstrating how social science techniques can enhance AI auditing. 💻🤖💬

Post attachment
Thomas Davidson avatar
Thomas Davidson@thomasrdavidson·Oct 24

Over the past 2 weeks @claudeai found bugs affecting core functionality in two different R packages just through routine use for research and teaching. The first PR was just merged. Less satisfying than finding these myself, but what amazing time for software development!

Emma Zang 臧熙璐 avatar
⟳ Thomas Davidson retweeted
Emma Zang 臧熙璐@DrEmmaZang·Oct 19

📢 Call for Papers Now Open! AI for Social Science Research Methods 2027 at Yale University! Excited to announce that we are now accepting paper submissions for the 2027 AI for Social Science Research Methods Conference, taking place April 28–29, 2027, at Yale University! We have an incredible lineup of confirmed speakers, including @kinggary, @ben_golub, Sanjog Misra, and Felix Elwert, among many others! As AI transforms how we conduct social science research, we're bringing together leading scholars from sociology, economics, political science, statistics, computer science, and beyond to explore new methodological frontiers, tackle emerging challenges, and shape the future of AI-driven social science. We welcome innovative methodological and empirical work on AI applications in social science research, including new tools, methods, and approaches to evaluation. 📅 Submission deadline: November 30, 2026 📍 Yale University | April 28–29, 2027 🔗 Conference details and submissions: web.cvent.com/event/3942a257… Please spread the word to colleagues and students who might be interested. We hope to see many of you at Yale next spring!

Workshop on Online Abuse and Harms avatar
⟳ Thomas Davidson retweeted
Workshop on Online Abuse and Harms@WOAHWorkshop·Jun 2

📢 Call for Reviewers: WOAH @ #EMNLP2026 WOAH is looking for reviewers to help evaluate submissions for our 10th edition at EMNLP 2026. Interested in serving as a reviewer? 🔗 docs.google.com/forms/d/e/1F… Thank you for helping us make #WOAH2026 a success! #NLP #AI #OnlineSafety

William J. Brady avatar
⟳ Thomas Davidson retweeted
William J. Brady@william__brady·May 27

✨New paper out @SpringerNature✨ For 8 weeks around the 2024 US election, we randomly assigned 2,000 people to use social media algos we custom-built. Do engagement-based algorithms amplify intergroup, moral & emotional content + does that distort how we see political norms? 🧵

Post attachment
PNAS Nexus avatar
⟳ Thomas Davidson retweeted
PNAS Nexus@PNASNexus·Mar 5

AI-generated summaries of history led to more liberal opinions compared to Wikipedia, while summaries by chatbots prompted to use a conservative framing produced more conservative opinions—but primarily among conservative readers. In PNAS Nexus: ow.ly/oYQH50YpOHO

Post attachment
Emma Zang 臧熙璐 avatar
⟳ Thomas Davidson retweeted
Emma Zang 臧熙璐@DrEmmaZang·Mar 6

Call for Submissions: AI for Social Science Methodology @yaledatascience • Keynote: @NAChristakis • Panel with editors of leading journals on publishing AI research • Mentoring roundtables for early-career scholars • Generous travel support Discussion-driven, high-quality research. 📩 Submit your work: yalefds.swoogo.com/socialsci… Please share with colleagues working at the AI × social science frontier! @Yale @PopAssocAmerica

Chris Bail avatar
⟳ Thomas Davidson retweeted
Chris Bail@chris_bail·Mar 3

Want to learn about computational social science *for free* and identify new research partners across academic fields? Apply to one of the 2026 Summer Institutes in Computational Social Science (described in yellow in the attached map) here: sicss.io/locations

Post attachment
Thomas Davidson avatar
Thomas Davidson@thomasrdavidson·15 Dec 2025

Overall, these results show that MLLMs can make more context-sensitive moderation decisions than text-based classifiers. These systems still make mistakes, and context can cut both ways, eliminating some biases while enabling others. Human oversight thus remains essential.

Thomas Davidson avatar
Thomas Davidson@thomasrdavidson·15 Dec 2025

Additionally, some models are particularly sensitive to visual identity cues (AI-generated profile pictures). This demonstrates how different data modalities result in different levels of algorithmic bias.

Post attachment
Thomas Davidson avatar
Thomas Davidson@thomasrdavidson·15 Dec 2025

When considering the identity of the author, some MLLMs make judgments similar to human subjects. e.g., less likely to flag Black users for using reclaimed slurs, a common false positive. But the results also reveal less normative decisions regarding so-called "reverse racism".

Post attachment
Thomas Davidson avatar
Thomas Davidson@thomasrdavidson·15 Dec 2025

I find that multimodal large language models (MLLMs) follow a consistent hierarchy of offensive language to humans, but there is also extensive heterogeneity across models.

Post attachment