Generative AI and Sustainability Governance
Project Description
This project investigates how corporate executive and board members' sustainability credentials can be systematically surfaced from public disclosure documents. Participants to the project will explores how large language models can be applied to parse, interpret, and classify executives’ and board members’ biographies.
Supervisor
HUANG, Hao
Quota
5
Course type
UROP1000
UROP1100
UROP2100
UROP3100
UROP3200
UROP4100
Applicant's Roles
Students will work with a corpus of extracted biographical narratives for individual executive and board members. Using generative AI-assisted text mining techniques, participants will design prompts and workflows that help an AI model flag language associated with sustainability-related experience such as prior roles in environmental policy, clean energy ventures, conservation organizations, or ESG oversight committees. The resulting AI-generated classifications will then be validated and refined against human judgment to build a more reliable detection pipeline.
Applicant's Learning Objectives
Working with unstructured text: Gain hands-on practice reading and structuring qualitative disclosure data pulled from corporate filings.
Sustainable governance literacy: Build a working understanding of ESG principles and why environmentally qualified executives and directors matter to corporations.
LLM-driven classification design: Learn to construct and iterate on generative AI prompts and pipelines that automatically detect and categorize domain-specific expertise within text, while critically evaluating where AI outputs need human correction.
Sustainable governance literacy: Build a working understanding of ESG principles and why environmentally qualified executives and directors matter to corporations.
LLM-driven classification design: Learn to construct and iterate on generative AI prompts and pipelines that automatically detect and categorize domain-specific expertise within text, while critically evaluating where AI outputs need human correction.
Complexity of the project
Challenging