Call for Papers: Inaugural Large Language Models and the Social Sciences Conference (LLMSS 2026)
Date: 14 – 16 October
Venue: City University of Hong Kong (CityU)
The Large Language Models and the Social Sciences 2026 (LLMS 2026) conference is an interdisciplinary forum for scholars and practitioners to explore the intersection of advanced large language models (LLMs) and social science research. It will bring together AI researchers (computer scientists, engineers, machine learning experts, applied mathematicians), industry practitioners (e.g., health sciences, legal and fintech) government representatives and social scientists from fields like political science, sociology, economics, and communication. Participants will share cutting-edge developments in LLM technology, novel applications to social data, and insights gained from cross-disciplinary collaboration. The conference is organized by Peking University’s Analytics Lab for Global Risk Politics, CityU’s Computational Social Sciences Lab, and University of Oxford’s Nuffield College Talking to Machines Initiative, with support from the European Political Science Association (EPSA).
LLMS 2026 comes at an early stage of integrating LLM-based methods into social science workflows. This conference provides a timely international forum to showcase innovative research, exchange ideas, and receive constructive feedback. The workshop-style format – including research presentations with designated discussants and Q&A – is designed to foster open discussion, collaboration, and methodological advancement among participants from diverse fields. We especially encourage interdisciplinary work and aim to cultivate a feedback-oriented environment that sparks new partnerships between the technical AI community and social scientists.
Organized by Raymond Duch (University of Oxford), this IMEBESS/EPSA Prague 2026 pre-conference workshop brings together researchers exploring applications of Large Language Models (LLMs) in the social sciences.
Workshop Themes
The Oxford Talking to Machines (T2M) project has conducted a series of workshops over the past two-years exploring methods for incorporating large language models (LLMs) into how we conduct social science research. T2M is currently running a large-scale replication project, Synthetic Replication Game that explore the potential of LLMs to augment human samples in social science experiments. At the recent IMEBESS (Valencia) and EPSA (Madrid) meetings we organized pre-conference workshops bringing together scholars who shared the T2M interests in synthetic replications and incorporating LLMs into social science research.
Building on the success of these previous workshops, we will again organize a pre-conference workshop for the 2026 Prague meetings of the IMEBESS and the EPSA.
The LLM Workshop focus will be on the application of LLMs to research being conducted in the social sciences. The workshop will give particular attention to the evolving tools that we are using to leverage AI in social science research; to showcasing the innovative research designs that incorporate AI and the insights they have generated; to addressing the challenges of conducting cross-national research with AI tools. The Synthetic Replication Game (SRG) team of over 150 participants have made impressive advances in synthetically replicating human subjects in experiments – these will be highlighted at the workshop.
Previous workshops have covered many broad themes related to LLMs and social science and we expect to continue this tradition – themes such as: the role of AI agents as subjects in experiments; fine-tuning and RAG strategies; the role of LLMs in the analysis of conversations and open-ended survey responses; LLM guidance in the design of experimental protocols and treatment arms; technical advances including the incorporation of non-text data such as images, video and audio, transformer-based models, and DNNs; and the ethical challenges of incorporating LLMs into social science research; efforts to address safety and LLM model development and deployment. The workshop format offers presenters a unique opportunity to showcase their work and get constructive and thoughtful feedback.
The pre-conference workshop takes place Wednesday July 8th, a day before IMEBESS and EPSA Prague 2026. The workshop will be held in the same venue as the main meetings – the Prague Convention Center. We cannot provide travel or accommodation support for participants. For those who are accepted to participate, we will provide a workshop dinner on July 8th and lunch during the workshop.
Core Research Themes
We invite submissions spanning a broad range of topics at the nexus of LLM development and the social sciences. Specific areas of interest include, but are not limited to, the following themes. (Submissions may be theoretical or applied, and we also welcome papers that introduce new tools, datasets, libraries, or platforms to support LLM-driven social science research).
Data Generation & Collection
- LLMs in Experimentation: Using AI agents as survey respondents, interviewers, experimental subjects or confederates in social science studies.
- Adaptive Experiment Design: LLM-guided design of surveys and experiments (e.g. dynamically generating or refining treatment vignettes and protocols).
- Synthetic Data & Augmentation: Generating synthetic text or multimedia data with LLMs to augment training datasets or simulate social scenarios.
- Dataset Curation with LLMs: Large-scale compilation of text, image, video, and audio corpora for social research, and LLM-assisted data labeling and annotation.
LLM Applications in Social Science Research
- Text Analysis & Classification: LLM-based methods for content analysis, topic classification, sentiment analysis, and embedding-based representations of social science text data.
- Causal Inference from Text: Using LLMs to design or interpret experiments (e.g. estimating treatment effects from text and image vignettes created for studies).
- Open-Ended Responses & Conversations: Analyzing interviews, focus group transcripts, social media conversations, and open-ended survey responses with the help of LLMs.
- Domain-Specific Applications: Innovative uses of LLMs in domains such as political communication, public policy, economic modeling, law, and cultural analysis.
LLM Development & Adaptation
- Interpretability and Explanation: Techniques for interpreting LLM decisions and outputs (model explainability, transparency, probing of internal representations).
- Fine-Tuning & Domain Adaptation: Methods for fine-tuning or adapting LLMs to domain-specific tasks and datasets in the social sciences (including low-resource or specialty corpora).
- Prompt Engineering: Strategies for prompt design and conditioning that improve LLM performance, reliability, and factuality in social science use-cases.
- Evaluation & Benchmarking: Developing evaluation protocols, benchmark tasks, and metrics for LLMs (including performance benchmarking on social science tasks, bias and fairness audits, and behavioral analysis of models).
- Model Behavior & Emergent Capabilities: Characterizing LLM behavior, biases, and emergent capabilities (e.g. consistency, truthfulness, adaptability), and identifying failure modes or hallucinations.
- Multimodal & New Architectures: Advances in LLM architectures, including multimodal models that integrate text with images, audio, or video, and other emerging model innovations relevant to social science data.
Tools, Platforms & Infrastructure
- Research Tools and Libraries: Development of software tools, libraries, or frameworks (preferably open-source) to facilitate LLM-based analysis and workflows for social scientists.
- Platforms and Pipelines: LLM-driven research platforms, toolchains, or pipelines that integrate LLMs into data collection, analysis, or visualization processes (for example, interactive analysis notebooks, API integrations, or collaborative environments for human-AI research).
- Scalability and Deployment: Practical challenges and solutions for deploying LLMs in real-world settings – including scalable model serving, computing infrastructure for large models, cost-efficient inference, and integrating LLMs into organizational or public-facing applications.
- Reproducibility & Best Practices: Infrastructure and methodological best practices to ensure reproducible LLM experiments (versioning of models/prompts, evaluation standardization, and result validation).
Ethics, Policy & Societal Impact
- Ethical & Responsible AI: Ethical challenges in using LLMs for research and in deployment (e.g. issues of privacy, consent, misinformation, and the responsible design of human-AI interactions).
- Fairness and Bias: Identifying and mitigating biases in LLM training data or outputs; ensuring fairness and equity when applying LLMs across different populations or languages.
- Policy and Governance: Implications of widespread LLM adoption for institutions and policy – including regulation of AI, legal considerations, and governance frameworks for LLM use in society.
- Societal & Economic Impact: The broader impact of LLMs on society and the economy (such as effects on labor markets, education, media, and public discourse) and how social scientists can study these changes.
- Robustness & Reproducibility: Ensuring reproducibility and robustness in LLM-based research findings (validation of results, model stability, and transparent reporting of methods), as well as discussions of safety and alignment for LLMs used in sensitive social contexts.
Conference Format
- Invited Keynote Talks: Talks by leading experts from academia and industry at the forefront of LLM research and applications.
- Research Paper and Poster Sessions: Presentations of accepted papers and poster displays for work-in-progress, with ample time for questions and discussion.
- Interactive Workshop Presentations: Seminar-style sessions where researchers can demo tools, share data, or conduct mini-tutorials on specialized methods.
- Methodological Demonstrations: Live demonstrations of new software, libraries, or experimental techniques relevant to LLMs and social science.
- Structured Feedback & Discussion: Dedicated discussant feedback for each presented paper and open-floor discussions to provide in-depth, constructive critique and foster collaboration.
Keynote Speakers
Issa Dahabreh
Associate Professor of Epidemiology and Biostatistics
Harvard T.H. Chan School of Public Health
Issa Dahabreh, MD ScD, is Associate Professor of Epidemiology and Biostatistics at the Harvard T.H. Chan School of Public Health, and Section Head for Epidemiology and Data Science at the Richard A. and Susan F. Smith Center for Outcomes Research. His research develops methods for causal inference, evidence synthesis, and the use of trial and external data to improve the design and analysis of studies informing clinical and public health decisions.
Scott A. Hale
Associate Professor & Director
Oxford Internet Institute, University of Oxford
Dr Scott A. Hale is an Associate Professor and Director of the Oxford Internet Institute, a multidisciplinary department at the University of Oxford focused on understanding the effects of new technologies on society. He is also Director of Research at Meedan, a technology nonprofit building public interest AI tools. At Oxford, he leads the Equitable Access to Quality Information Lab (eaqilab), which focuses on developing and applying new machine learning approaches to understand and improve how people discover, evaluate, and make use of online information in their daily lives. He is particularly interested in LLM alignment and multilingual Natural Language Processing applications as well as putting research into practice with community and media organizations.
Pierre Landry
School Director & Professor
Political Science
Professor Landry’s undergraduate training was in economics and law at Sciences-Po in Paris. He received his Ph.D. in Political Science at the University of Michigan and is an alumnus of the University of Virginia (MA in Foreign Affairs) and the Johns Hopkins–Nanjing University program at the Center for Chinese and American Studies in Nanjing. His research interests focus on Asian and Chinese politics, comparative local government, quantitative comparative analysis and survey research, and he has written on governance and the political management of officials in China. Besides articles and book chapters in comparative politics and political methodology, he is the author of “Decentralized Authoritarianism in China” with Cambridge University Press (2008). He is also the co-investigator of the Barometer on China’s Development (BOCD) at the Universities Service Centre for China Studies (Chinese University of Hong Kong) and serves on the international advisory committee of the Centre.
King-wa Fu
Professor
Journalism and Media Studies Centre (JMSC), The University of Hong Kong
King-wa Fu is a Professor at the Journalism and Media Studies Centre (JMSC), The University of Hong Kong. His research interests include China’s information governance, media and political participation, computational social sciences, health and the media, and the younger generation’s media use. He was a visiting Associate Professor at the MIT Media Lab and a Fulbright-RGC Hong Kong Senior Research Scholar in 2016–2017, a China-US Scholar in 2021–2022 at Boston University, and a Richard von Weizsäcker Fellow at the Robert Bosch Stiftung since 2023.
Jian Hua Jonathan Zhu
Chair Professor of Computational Social Science
City University of Hong Kong
Prof. Jian Hua Jonathan Zhu is Chair Professor of Computational Social Science at City University of Hong Kong and Director of the Centre for Communication Research (CCR). He is a Chair Professor in the Department of Media and Communication and in the Department of Data Science.
Paper Submission
Submissions should be made through the conference submission system (COMS):
We welcome full research papers as well as extended abstracts describing ongoing work, preliminary results, or promising projects that would benefit from feedback.
Key Dates
- Abstract submission deadline: 7 August 2026
- Notification of acceptance: 15 August 2026
Further deadlines (e.g. camera-ready paper submission) will be communicated to accepted authors.
Registration
- Early registration (until 31 August 2026): €550
- Regular registration (1 September 2026 onward): €650
- Registration deadline: 15 September 2026
Cancellation Policy
If you must cancel your conference registration, please notify us as soon as possible at melanie.sawers@nuffield.ox.ac.uk. Refunds will be processed under the following terms:
- Cancellations before 15 September 2026: Full refund minus a €50 processing fee.
- Cancellations after 15 September 2026, or failure to attend: No refund available.
Call for Pre-Conference Workshops
LLMSS 2026 is assembling a hands-on, code-along workshop programme on applied AI/LLM methods for the social sciences, held Tuesday 13 October — the day before the main conference. The confirmed line-up already spans AI public opinion polling, digital twins, the “Silicon Jury,” and prediction-powered inference (PPI++). We are inviting proposals for additional half-day workshops.
We’re looking for: hands-on, code-along sessions (not lecture-only) on applied methods at the intersection of AI/LLMs and social science research — text-as-data, causal inference and RCTs, fine-tuning and domain adaptation, generative agents and multi-agent simulation, or other methodological frontiers relevant to the conference themes.
To propose a workshop, send the following to Melanie Sawers (melanie.sawers@nuffield.ox.ac.uk) by 31 August 2026:
- Workshop title and a short description
- Target audience and any prerequisites
- Format and a rough session outline (e.g. framing → worked example → hands-on exercise → discussion)
- Hands-on component: tools, packages, or datasets involved
- Organiser name(s) and short bio(s)
Workshops are capped at 30–50 participants to keep the hands-on experience genuine. Accepted workshop organisers receive a complimentary conference registration; the conference provides the room and A/V, and organisers are responsible for their own materials (we recommend cloud-based notebooks — Google Colab or Hugging Face Spaces — to avoid setup issues on the day).
Places in the pre-conference workshops are otherwise allocated through conference registration.
Pre-Conference Workshops
Public Opinion · Digital Twins · The Silicon Jury · Valid Inference
Tuesday 13 October 2026 · City University of Hong Kong
A full-day, hands-on workshop programme on applied AI/LLM methods for the social sciences. The core track introduces the Talking to Machines digital twin approach — from measuring public opinion, to generalising belief measurement, to designing strategy with synthetic panels — alongside additional sessions on validly incorporating LLM predictions into research.
- Morning – 09:00–12:30 – AIPOP
- Afternoon I – 13:30–15:30 – Talking to Digital Twins
- Afternoon II – 15:45–17:45 – The Silicon Jury
- Additional – Time TBA – Predicting Human Behavior with LLMs
Workshop Programme
AIPOP — Artificially Intelligent Public Opinion Polling
Leads: Raymond Duch, Laurenz Günther & Matias Fuentes Becerra
As traditional surveys grow more expensive and response rates collapse, large language models offer a provocative alternative: inferring public opinion from the digital traces people already leave behind. This hands-on workshop walks participants through PoSSUM — our Protocol for Surveying Social Media Users with Multimodal LLMs — the AI polling method whose state-by-state forecasts of the 2024 US presidential election tracked, and at times outperformed, the leading poll aggregators. We move through the full pipeline: designing the digital interview, building a “silicon” sample of social-media users, and producing bias-corrected estimates with multilevel regression and post-stratification (MrP) in R, validated against ground-truth election results. A featured sub-theme is our Swiss “Silicon Politicians” study, which predicts how individual politicians and citizens vote in referendums from their social-media traces alone — and showcases the app we built around it. Throughout, we keep the harder question in view, drawing on the recent APSA report Public Opinion in the Age of AI: when does simulating respondents enrich measurement, and when does it risk manufacturing the opinion it claims to observe?
Aimed at pollsters, political scientists, and survey methodologists comfortable with R; no machine-learning background required.
| Time | Session |
|---|---|
| 09:00–09:30 | Why AI polling? |
| 09:30–10:10 | PoSSUM and the 2024 US election case |
| 10:10–10:25 | Coffee break |
| 10:25–11:30 | Hands-on: Build a silicon sample and run MrP |
| 11:30–12:05 | Swiss ‘Silicon Politicians’ demo |
| 12:05–12:30 | Discussion & Q&A |
| 12:30–13:30 | Lunch break |
Talking to Digital Twins — The Wisdom of Digital Twins
Leads: Raymond Duch & Matias Fuentes Becerra
The afternoon extends the morning’s method beyond elections. The same architecture — unobtrusively harvesting someone’s public posts, repeatedly interviewing an LLM “digital twin” built from them, and applying panel econometrics to the result — turns scattered, self-selected commentary into a balanced, forward-looking panel of on-demand forecasters. We work through three applications. In financial markets, we build digital twins of “finfluencers” and interview them daily, recovering their stock-level beliefs even on days they post nothing, and show that these signals predict the cross-section of S&P 500 returns without look-ahead bias (drawing on our Talking to Digital Twins study, Bowles et al., 2026). We then turn to forecasting and prediction-market applications, and to a measurement problem the method is unusually suited to: eliciting views on sensitive topics people are reluctant to volunteer — treating silence itself as a belief state. Participants build a twin, run a repeated-interview protocol, assemble the panel, and evaluate a simple forecast.
Aimed at financial economists, quantitative and survey researchers, and data scientists comfortable with Python or R; the morning session is helpful but not required.
| Time | Session |
|---|---|
| 13:30–13:55 | From polls to markets |
| 13:55–14:30 | Finfluencer twins and S&P 500 signals |
| 14:30–15:15 | Hands-on: Build a digital twin |
| 15:15–15:30 | Backtesting and prediction markets |
| 15:30–15:45 | Short break |
The Silicon Jury — Optimising Legal Strategy with Digital Twins
Leads: Laurenz Günther with Raymond Duch
This capstone session turns the digital twin from a measurement instrument into a design tool. The same architecture that reads opinion can be run as a closed loop: propose a strategy, have a panel of LLM twins evaluate it, and search for the version that resonates best with the target audience. We introduce the engine on familiar ground — generating and optimising political-party platforms against a twin panel, using the same party-policy optimisation harness demoed in the morning session — then re-point that identical harness at the courtroom. Participants build twins of jurors and judges, stage a synthetic mock trial, and vary the argument — opening framing, evidence order, narrative emphasis — reading the twins’ verdict probabilities as the case is re-argued. Throughout we keep validity and ethics in view: how to validate synthetic jurors against the experimental mock-trial literature, how the contamination of well-known cases threatens prediction, and why these are tools for stress-testing arguments before they are made, never for replacing adjudication.
Pitched to engage the legal community — law schools, litigation and legal-technology firms, and judiciary-adjacent researchers — alongside computational social scientists. Aimed at intermediate participants comfortable running Python notebooks; no legal or machine-learning background required.
| Time | Session |
|---|---|
| 15:45–16:10 | Closed-loop optimisation |
| 16:10–16:50 | Political platform optimisation |
| 16:50–17:30 | Juror and judge twins |
| 17:30–17:45 | Ethics, validation & Q&A |
Predicting Human Behavior with LLMs — A Practical Toolkit for Valid Inference
Lead: David Broska
Researchers in academia and industry increasingly propose using LLM predictions of human behavior to pilot, augment, or even replace human data collection. When do these predictions support valid inferences, and when do they mislead? This workshop introduces a practical toolkit for answering that question and for using them in scientifically defensible ways. Drawing on recent frameworks for using and validating LLM predictions as behavioral evidence (Broska et al. 2025; Hullman et al. 2026), participants will learn to assess when predicted responses can be trusted, recognise common failure modes, and combine human and synthetic samples for greater precision without introducing bias. Used carefully, predictions do not replace human samples but help researchers design more informative studies.
Attendees will leave with the concepts and hands-on experience needed to decide whether and how to incorporate LLM predictions into their own research.
Practical Details
- Code-along workshops with notebooks and datasets provided.
- Laptop and web browser required.
- Suitable for PhD students and researchers across the social and data sciences.
- Places allocated through conference registration.
Accommodation
Participants are responsible for arranging their own accommodation. Please refer to the conference website (see Programme section) for suggested hotels and practical information for visitors. Early booking is recommended as October is a busy period in Hong Kong.
Contact
For any queries or difficulties with the submission system, please contact Melanie Sawers (conference coordinator) at melanie.sawers@nuffield.ox.ac.uk. We are happy to assist with questions about the conference scope, submissions, or logistics.
Venue & Accessibility
- Location: City University of Hong Kong – a modern, fully accessible campus located at Tat Chee Avenue, Kowloon Tong, Hong Kong.
- Public Transit: The campus is directly accessible via the Kowloon Tong MTR station (connected to several major lines).
- Amenities: Immediately adjacent to Festival Walk shopping mall (with numerous restaurants, cafés, and shops).
- Airport Access: Approximately 30–40 minutes from Hong Kong International Airport, either by taxi or by taking the Airport Express train to Kowloon Station and a short taxi ride to campus.
Explore Hong Kong
Take the opportunity to explore Hong Kong’s vibrant culture and sights during your visit:
- Victoria Peak: Enjoy panoramic views of the famous skyline and Victoria Harbour.
- Kowloon Walled City Park: A historic site and tranquil garden (located near CityU).
- Mong Kok Markets: Experience the bustling street markets, local shops, and diverse dining options in Mong Kok.
- Tsim Sha Tsui Promenade: Stroll along the waterfront for stunning harbour views, especially beautiful at night.
- October Weather: Hong Kong’s autumn weather is typically warm and comfortable (≈26–29°C / 79–84°F), ideal for sightseeing.