Ashley
Lewis
PhD Candidate at The Ohio State University
"Hallucination in AI Dialogues: Detection and Mitigation"
"Hallucination in AI Dialogues: Detection and Mitigation"
Don’t miss out and join visionaries, innovators, and thought leaders from all over the world at the Women in Tech Global Conference.
Vote by Sharing
Unite 100 000 Women in Tech to Drive Change with Purpose and Impact.
Do you want to see this session? Help increase the sharing count and the session visibility. Sessions with +10 votes will be available to career ticket holders.
Please note that it might take some time until your share & vote is reflected.
Session: Hallucination in AI Dialogues: Detection and Mitigation
Large language models often produce text that sounds confident and authoritative — but not necessarily true. In this talk, I unpack why these “hallucinations” (or more accurately, confabulations) occur, and what we can do to detect and reduce them. Drawing on my research in dialogue systems and recent experiments with retrieval-augmented generation, I introduce the VISTA Score, a new method for evaluating factual accuracy in conversational AI. I also discuss practical strategies for mitigating confabulation through retrieval design, prompt engineering, and data cleaning — and why our word choices (“hallucination” vs. “confabulation”) shape how non-experts think about AI systems. The goal is to bridge technical insight and public understanding: building systems — and habits — that recognize the difference between fluency and truth.Key Takeaways
- Understand why large language models “hallucinate” — and why confidence doesn’t always mean correctness.
- Learn practical strategies for detecting and reducing hallucination in conversational AI systems.
- Explore new approaches like VISTA Score, a dialogue-specific framework for evaluating factual accuracy.
- Recognize how language choices (“hallucination” vs. “confabulation”) shape public perceptions of AI.
- Gain tools for promoting AI literacy — helping teams and users engage critically and confidently with AI-generated content.
Bio
Ash Lewis is a PhD candidate in Computational Linguistics at The Ohio State University, where she studies how to make conversational AI systems more factual, reliable, and human-aware. Her research focuses on mitigating hallucinations in dialogue through efficient, data-driven methods such as knowledge distillation, self-training, and synthetic data generation. She is currently developing a virtual tour guide for the COSI museum in Columbus, Ohio, an AI system that engages visitors in conversations about language and science, and was part of a team that placed third internationally in Amazon’s AlexaPrize Taskbot Challenge. Her work bridges computational linguistics and AI ethics, emphasizing practical strategies for building systems that know the difference between fluency and truth.