Abstract

– Theo Araujo

Lecture:

Trust calibration in generative AI: Investigating daily human-AI interactions

As conversational generative AI agents become part of everyday life, a pressing question emerges: how do individuals calibrate their trust in systems that adapt to their behavior and play an increasingly central role in their online interactions? This lecture outlines an integrated framework for studying trust calibration in generative AI, covering what trust calibration means in practice, the challenges that emerge from daily human-AI interaction, and how computational social science methods can help us study this process.

Workshop:

Digital data donation: Studying social phenomena in partnership with citizens 

Access to digital trace data is both a central need and a growing challenge for computational social scientists — platform APIs are increasingly restricted, and alternatives are scarce. Data donation, a citizen science approach where individuals share their own digital data with researchers, offers a promising path forward. This workshop walks through the full data donation workflow using the open-source PORT tool, and includes a hands-on brainstorming session to help participants apply it to their own research.

– Claudia Wagner

Foundations of Computational Social Science: From Origins to Infrastructure

How did computational social science emerge, and what defines the field today? This lecture takes a journey from the early days of digital data to the current landscape shaped by large-scale data collection, AI methods, and interdisciplinary collaboration. We introduce the core pillars of CSS—data, computational methods, and theory—and discuss how they jointly enable new ways of understanding social phenomena. Finally, we consider how research infrastructure—such as data collections, tools, and collaborative environments—supports CSS research, and what challenges and opportunities lie ahead.

– Luca Pappalardo

Human-AI feedback loops and their impact on society

Recommender systems and AI assistants have become ubiquitous, fundamentally reshaping human decision-making across digital platforms. This creates a continuous, self-reinforcing feedback loop: user choices generate the training data for AI models, which then refine their outputs to further influence future user preferences. This dynamic differs significantly from traditional human-machine interaction, often yielding complex and unintended systemic consequences. We will discuss the concept of human-AI co-evolution as a framework for understanding these interactions. We will evaluate the strengths and limitations of current methodologies, highlighting critical shortcomings in capturing feedback mechanisms. By examining diverse real-world ecosystems (social media, urban, online retail, and generative AI platforms) we will identify the multidisciplinary challenges inherent in establishing this field of study. Finally, we conceptualize these hurdles across three increasing levels of abstraction (scientific and legal) to provide a roadmap for future research into the long-term impacts of AI-driven influence.

– Debora Nozza

A Roadmap for the Everyday Use of LLMs: Emerging Risks and Research Directions

Large Language Models (LLMs) are increasingly part of everyday life, shaping how people seek information, advice, and guidance. This rapid shift raises new challenges that extend beyond traditional NLP benchmarks, as models can influence decisions, beliefs, and perceptions in subtle but powerful ways. In this talk, I will reflect on recent research and ongoing work aimed at identifying these challenges and exploring how we can design LLMs that foster safer, more trustworthy, and more pluralistic interactions.