My research group:

Overarching Research Themes

*Extracted by GPT-4, there may be inconsistencies.* #### *Ethics in AI Interactions* My research group explores the ethical implications of artificial intelligence in human interactions. We investigate biases present in large language models, as highlighted in the important paper [Rejected Dialects: Biases Against African American Language in Reward Models](http://maartensap.com/publications.html#mire2025rejectedDialects), which sheds light on inherent biases against specific dialects. Additionally, we emphasize the importance of user interaction through the lens of [Rel-A.I.: An Interaction-Centered Approach To Measuring Human-LM Reliance](https://arxiv.org/abs/2407.07950), focusing on how users’ reliance on AI can shape outcomes. Our work aims to enhance accountability and value alignment in AI systems, as illustrated in the paper [HAICOSYSTEM: An Ecosystem for Sandboxing Safety Risks in Human-AI Interactions](http://arxiv.org/abs/2409.16427). #### *The Dynamics of Narratives* My research group explores the complexities of narratives, particularly in how they shape and reflect social dynamics. A significant contribution to this area is the paper [Modeling Empathic Similarity in Personal Narratives](https://arxiv.org/abs/2305.14246), which examines how empathy can be quantified within storytelling. We also delve into the impact of narratives on perceptions in the digital age, as discussed in the paper [HEART-felt Narratives: Tracing Empathy and Narrative Style in Personal Stories with LLMs](https://arxiv.org/abs/2405.17633). This theme underscores the importance of understanding narrative flow, supported by findings from [Quantifying the narrative flow of imagined versus autobiographical stories](https://www.pnas.org/doi/10.1073/pnas.2211715119). #### *AI Agents and Social Intelligence* My research group explores the intricate relationships between AI agents and social intelligence. Our work focuses on evaluating the social reasoning capabilities of language models, as highlighted by the paper [SOTOPIA: Interactive Evaluation for Social Intelligence in Language Agents](https://arxiv.org/abs/2310.11667), which provides a framework for assessing AI agents' social dynamics. We also address the ethical trade-offs faced by these agents, elaborated in [AI-LieDar: Examine the Trade-off Between Utility and Truthfulness in LLM Agents](https://arxiv.org/abs/2409.09013). Additionally, our investigations into the limitations of social intelligence in AI are informed by the findings in [Neural Theory-of-Mind? On the Limits of Social Intelligence in Large LMs](https://aclanthology.org/2022.emnlp-main.248/). #### *Understanding Toxic Language* My research group explores the complexities and challenges of addressing toxic language in AI systems. Our recent study [PolygloToxicityPrompts: Multilingual Evaluation of Neural Toxic Degeneration in Large Language Models](https://arxiv.org/abs/2405.09373) surveys biases and deterioration of language models across different languages. Furthermore, we examine the perspectives of those combating online hate in our paper [Counterspeakers’ Perspectives: Unveiling Barriers and AI Needs in the Fight against Online Hate](https://arxiv.org/abs/2403.00179), which seeks to identify gaps in current approaches. Our work also addresses broader contextual factors that influence interpretations of hate speech, as detailed in [COBRA Frames: Contextual Reasoning about Effects and Harms of Offensive Statements](http://arxiv.org/abs/2306.01985).