Explainable AI from Cognitive Science perspectives
Project Description
In this project, students will have hands-on experiences in conducting research on explainable AI. It will involve evaluation and human-centric benchmarking of large language models such as GPT or Llama, or vision language models such as CLIP, from the perspectives of human cognition and human-AI interaction. More specifically, we will compare human cognition and AI models in both their behavior/performance and underlying mechanisms through both computer science and cognitive science methods.
Supervisor
HSIAO, Janet Hui-wen
Quota
3
Course type
UROP1000
UROP1100
UROP2100
Applicant's Roles
The applicant will either help collect or analyse human data to be compared with AI models, or help experimenting on AI models for evaluation or benchmarking purposes. Basic knowledge about cognitive science/experimental psychology methods, quantitative data analysis, or machine learning/programming skills are required.
Applicant's Learning Objectives
- To learn about the literature on explainable AI from human cognition perspectives.
- To leann about methods for alignment, evaluation, or humen-centric benchmarking of AI models and have hands-on experience in comparison studies between humans and AI models.
Complexity of the project
Challenging