Conversational AI: Practical Challenges in Talking to People

This is a hybrid in-person and online event. Pre-registration is required for either.
Conversational AI systems today speak with remarkable confidence, often giving the impression of understanding and reasoning. However, teams deploying these systems often quickly encounter familiar problems: drift, hallucinations, contradictory answers, and conversations that quietly lose their original purpose. Why do systems that seem so capable end up behaving so unpredictably?
In this talk, Elena Gostrer will examine these behaviors from a practical, product-engineering perspective. Rather than exploring model internals, this talk will focus on what actually happens when humans interact with probabilistic language models – and why traditional software assumptions fail in conversational settings.
Elena will also discuss what architectural patterns teams are adopting to keep in control, and she’ll highlight why combining generative AI with explicit structure, state, and constraints is becoming essential. Attendees will leave with a clearer understanding of why conversational AI breaks, what helps it behave more reliably, and how to think differently about designing human-AI interactions.
Speaker(s): Elena Gostrer,
925 Thompson Place, Sunnyvale, California, United States, 94085, Virtual: https://events.vtools.ieee.org/m/541387