I am a PhD student in Social & Engineering Systems and Statistics at MIT working in the Clinical Machine Learning group led by David Sontag. My research focuses on building methods to improve decision-making abilities of Human-AI teams through appropriate reliance and deferral. Applications of my research include healthcare domains such as radiology and pregnancy complications as well as in information retrieval and code generation.
A focus of my research has been on developing new methods that allow AI classifiers to know when to defer to humans and when to make decisions on their own (learning to defer). When humans have to make the final decision, I have created onboarding methods to allow humans to know when to effectively use AI classifiers. More recently, I’ve started working on understanding how programmers interact with coding assistants such as GitHub Copilot and how to improve that experience.
Past experiences: I received my undergraduate degree in computer engineering from the American University of Beirut in 2019. In the summer of 2022 I was a research intern in the HAX team at Microsoft Research working on Copilot, in the summer of 2021 I interned at ASAPP where I worked on how customer service agents interact with AI suggestions.
You can reach me at mozannar@mit.edu
We provide a method to decide when to display suggestions in AI-assisted programming systems based on the programmer’s feedback. We show that we can avoid displaying a significant fraction of suggestions that would have been rejected.
We build new machine learning algorithms that can predict whether a patient is pregnant and whether they will have a high-risk pregnancy. We then integrate these algorithms into a user interface and evaluate it with nurses.
We provide algorithms that can provably minimize the learning to defer objective and provide an experimental benchmark to study human-deferral algorithms.
We study how programmers interact with the AI code-recommendation system Copilot and develop a taxonomy of programmer activities. We also predict programmer acceptance of AI code-recommendations.