Bias Analysis of AI Models for Undergraduate Student Admissions
Project Overview
The document explores the application of generative AI in higher education, particularly emphasizing its role in the admissions process and the critical need for bias detection and mitigation in predictive models. Utilizing a comprehensive six-year dataset from a large urban university, the research examines the effects of test-optional admissions policies on demographic representation and the potential biases these AI-driven models may perpetuate in admissions decisions. The findings reveal that while AI can significantly enhance the comprehension of admissions criteria and improve decision-making processes, it also poses risks of bias against sensitive populations, highlighting the importance of implementing robust evaluation and mitigation strategies to ensure equitable outcomes. This underscores the dual nature of generative AI in education, where its benefits must be balanced with the responsibility to address ethical concerns related to fairness and inclusivity.
Key Applications
Machine Learning-based AI models for predicting admissions
Context: Higher education admissions at a large urban research university, targeting prospective students applying to the School of Science.
Implementation: Developed AI models using admissions data before and after the implementation of test-optional policies, analyzing their biases with respect to sensitive demographic variables.
Outcomes: Identified significant bias in admissions predictions based on gender, race, and first-generation status; demonstrated how admissions policies affect demographics.
Challenges: Bias in predictive models can lead to unfair admissions outcomes; the need for nuanced understanding of how to evaluate fairness.
Implementation Barriers
Technical
AI models can introduce new biases and fairness issues based on the datasets used for training. Conduct thorough evaluations of AI models for errors and biases.
Proposed Solutions: Develop separate metrics for bias and fairness.
Policy
Changing admissions policies from test-required to test-optional can lead to significant shifts in admitted demographics, requiring further research. Ongoing evaluation of admissions policies and their impacts on demographic representation.
Project Team
Kelly Van Busum
Researcher
Shiaofen Fang
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Kelly Van Busum, Shiaofen Fang
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai