Vector Institute Launches Zero-Hallucination Chatbot for Canadian Air Passenger Rights
Researchers at the Vector Institute for Artificial Intelligence have developed a new chatbot designed to help Canadian air travelers understand and exercise their legal rights. Addressing the confusion caused by complex regulations and frequent flight disruptions, the tool breaks down official documents into easily digestible information. The system employs a novel zero-hallucination approach, utilizing GPT-4 for query decomposition and dense retrieval to extract accurate passages from a curated knowledge base. Instead of generating synthesized answers, it presents relevant source excerpts and links, ensuring reliability. A comparative user study involving fifteen participants demonstrated that the chatbot offers higher usefulness and satisfaction scores than manual Google searches, with strong quantitative retrieval performance metrics. This innovation not only empowers passengers but also supports volunteers at the non-profit Air Passenger Rights by reducing their workload. The project highlights the potential of retrieval-based AI methods in high-stakes legal contexts, setting a new standard for accuracy in AI-driven customer service and information dissemination within the aviation sector.
Wire timeline
Vector Institute Launches Zero-Hallucination Chatbot for Canadian Air Passenger Rights
Researchers at the Vector Institute for Artificial Intelligence have developed a new chatbot designed to help Canadian air travelers understand and exercise their legal rights. Addressing the confusion caused by complex regulations and frequent flight disruptions, the tool breaks down official documents into easily digestible information. The system employs a novel zero-hallucination approach, utilizing GPT-4 for query decomposition and dense retrieval to extract accurate passages from a curated knowledge base. Instead of generating synthesized answers, it presents relevant source excerpts and links, ensuring reliability. A comparative user study involving fifteen participants demonstrated that the chatbot offers higher usefulness and satisfaction scores than manual Google searches, with strong quantitative retrieval performance metrics. This innovation not only empowers passengers but also supports volunteers at the non-profit Air Passenger Rights by reducing their workload. The project highlights the potential of retrieval-based AI methods in high-stakes legal contexts, setting a new standard for accuracy in AI-driven customer service and information dissemination within the aviation sector.
Vector Institute for Artificial Intelligence