Wednesdays from 4:30 pm – 5:30 pm
The aim of this series is to hear about the innovative and exciting research that is taking place within our college, thus both faculty and students are encouraged to give a colloquium presentation. Talks should be either 40 - 50 minutes in length for a single presenter colloquium or 20 - 25 minutes in length for a joint colloquium event. If you would like to sign-up to give a CNHS Colloquium Talk, email Efren Ruiz.
Date: 25 March 2026
Where: STB 118
Speaker: Dr. Travis Mandel, UHH Computer Science and Data Science Department
Title: Towards Fast and Effective Personalization of AI Assistance
Abstract: AI assistance is becoming a critical part of workflows in domains like writing and computer programming. However, there remains a divide - some expert users find AI tools indispensable, while others have chosen to minimize AI assistance due to concerns about quality and reliability. While most work focuses on adapting the content of AI responses to meet a user’s preferences, in this paper we study the problem of efficiently adjusting the quantity of AI assistance to minimize the time it takes to perform a task. We formulate this as a filtering problem, where AI responses need to be filtered based on both AI confidence and learned properties of a user’s personality. We identify several important subproblems in this space, and address them with theoretically-motivated approaches. We evaluate our approach in simulation on multiple domains, including image analysis tasks using data from the Big Island, as well as assisted coding tasks to demonstrate the generality of our approach.