Trusted AI

Trustworthy Group Recommendation Recommender Systems Trusted AI

Description: We study the problem of trustworthy group recommendation, where the focus goes beyond individual user preferences and dives into collective tastes and interests. Our objective is to build novel methods and useful tools for group recommendation with fairness, and drive different use cases. We extend this problem to the team formation domain, where we present an AI-based system to aid in collaborative teaming for researchers responding to funding agency proposals. Our approach leverages NLP techniques to extract and normalize technical skills from various data sources, facilitating matching and teaming based on constraints. We have gathered initial feedback from university researchers to deploy the prototype system and published a dataset for broader use.

LLM and Chatbot Testing Trusted AI

Description: Increasingly powerful Large Language Model (LLM) based chatbots, like ChatGPT and Bard, are becoming available to users that have the potential to revolutionize the quality of decision-making achieved by the public. We find that although the outputs of the chatbots are fluent and plausible, there are still critical gaps in providing accurate and reliable information using LLM-based chatbots.

Rating of AI Systems Neuro-Symbolic AI Trusted AI

Description: AI systems, such as facial recognition and sentiment analyzers, often show model uncertainty, leading to algorithmic bias, particularly concerning protected attributes like gender and race. This research seeks to understand and reduce bias by establishing causal relationships. Assigning ratings to AI systems based on this helps users make informed choices.