Invited Talks

Representational Defaults and the Challenge of Pluralism for Generative AI

June 01, 2026

Talk, Tech for Impact Seminar at IBM, Virtual

In the talk titled, I will explore a central tension in decisions in generative AI representations: current approaches to “fixing” representational harms often push toward convergence — consensus on what is appropriate, standardized evaluations, universal guidelines. But lived experiences of representation demand the opposite: multiplicity, context-dependence, the ability to be legible in some contexts while pushing boundaries in others. In this talk I provide high-level guidance that required a focus not only on the distributional effects of disparities in representations, but also the process of creating with generative AI, shaping what possibilities are thinkable, imaginable, and relevant. I enumerate design opportunities: namely 1) context-conditional defaults and friction, for a system that aims to broaden our horizons (or set of possibilities), and 2) artifacts for representational provenance, that aim to carry some of the context alongside AI-generated content, creating opportunities for critical and measured engagement. Overall, the proposed affordances asked us to think about what a representation is intended to depict and stand for, who it’s for, and how it that context is carried (or removed) as it is ported into different contexts.

Representations - How We Shape Them and How They Shape Us

December 01, 2025

Talk, University of San Diego, San Diego, CA, USA

This talk traces the evolving meaning of representation: from the early days of symbolic AI, when researchers sought to model human cognition through explicit world-models, to today’s generative systems that learn statistical representations from datasets. In the shift from modeling to generation, representation has transformed the question of how machines know the world to how machines participate in constructing it. I first situate representation historically within AI’s technical lineage, tracing the move from symbolic reasoning to data-driven learning. I contrast these definitions with social dimensions of representation: what it means to be represented in media and cultural narratives, and how visibility shapes identity and belonging. Finally, I examine how generative AI systems contribute to both technical and social senses of representation, shaping what is visible, imaginable, and possible. As generative systems begin to shape what can be seen and conceived, this new representational order demands renewed ethical attention. It calls for new methods of evaluation, new frameworks for accountability, and new forms of deliberation about which representational baselines ought to guide collective notions of representation.

User Agency Across the ML Pipeline

December 01, 2024

Talk, California State University Fullerton, Virtual

User agency (i.e., the ability to affect change) is vital for preserving freedom, free will, and independence. As machine learning models are deployed in high-stakes domains like finance, hiring, and health, protecting agency becomes imperative. This talk examines two facets of user agency: overturning undesirable predictions over time and impacts on cognitive reasoning. Grounded in principles of user control and transparency, this work ensures machine learning systems empower users rather than constrain them.

Beyond Behaviorist Representational Harms: A Plan for Measurement and Mitigation

August 01, 2024

Talk, Microsoft's Aether Fairness & Inclusiveness Community Meeting, Virtual

Algorithmic harms are commonly categorized as either allocative or representational. This study specifically addresses the latter, focusing on an examination of current definitions of representational harms to discern what is included and what is not. This analysis motivates our expansion beyond behavioral definitions to encompass harms to cognitive and affective states. The paper outlines high-level requirements for measurement: identifying the necessary expertise to implement this approach and illustrating it through a case study. Our work highlights the unique vulnerabilities of large language models to perpetrating representational harms, particularly when these harms go unmeasured and unmitigated. The work concludes by presenting proposed mitigations and delineating when to employ them. The overarching aim of this research is to establish a framework for broadening the definition of representational harms and to translate insights from fairness research into practical measurement and mitigation praxis.

Interdisciplinary Paths in Computer Science

May 01, 2023

Talk, GradWIC x Kearny High School UCSD Campus Visit Keynote Speaker, San Diego, CA, USA

Discussing interdisciplinary paths to computer science with local high schoolers. Purely technical paths aren’t the only one!

Tell How You Want It: Recourse and Explainability for Generative Language Models

August 01, 2022

Talk, DeepMind Sociotechnical AI Effort Meeting, London, UK

Researchers and developers increasingly rely on toxicity scoring to moderate generative language model outputs, in settings such as customer service, information retrieval, and content generation. However, toxicity scoring may render pertinent information inaccessible, rigidify or “value-lock” cultural norms, and prevent language reclamation processes, particularly for marginalized people. In this work, we extend the concept of algorithmic recourse to generative language models: we provide users a novel mechanism to achieve their desired prediction by dynamically setting thresholds for toxicity filtering. Users thereby exercise increased agency relative to interactions with the baseline system. A pilot study (n=30) supports the potential of our proposed recourse mechanism, indicating improvements in usability compared to fixed-threshold toxicity-filtering of model outputs. Future work should explore the intersection of toxicity scoring, model controllability, user agency, and language reclamation processes – particularly with regard to the bias that many communities encounter when interacting with generative language models.

Jennifer Chien