Building Trust in Conversational AI
summary
Study interviews 150 users across retrieval, generative and hybrid chatbots and reports specific gains: 30–35% trust increase, 48% better understanding of reasoning, 27% more willingness for complex tasks and ~40% higher error tolerance. Explanations use lime shap counterfactuals: lime aids single responses, shap gives global feature importances but is computationally heavy, counterfactuals are intuitive but struggle with real time latency. Evaluation uses faithfulness stability and comprehensibility and notes privacy and bias risks that remain unresolved.
AI-written summary · about
chapters (10)
- 0:00Introduction to the Talk
- 0:39Understanding Explainable AI
- 1:30Evolution of Chatbots
- 2:13Key Explainable AI Techniques
- 3:26Research Methodology
- 4:32Evaluation Metrics and Results
- 5:46User Impact and Insights
- 6:36Challenges in Implementing Explainable AI
- 7:30Future Research Directions
- 8:11Conclusion
related talks
16mTrust by Design: Building Reliable AI for Sensitive Environments
12mSafety Perceptions of Generative AI Conversational Agents: Uncovering Perceptual...
29mBuilding Trust in AI Agents
01h 04mBuilding Expertise Into Generated Conversations
34mScaling Trust, Not Automation - Rethinking Quality for the AI Era
39mHow Chatty Leads Builds AI Agents Businesses Actually Want
24mHow to Build Trustworthy AI
43m3 Steps to Create Trustworthy Generative AI Products