Machine Learning Bias

Meyer, Sophie, School of Engineering and Applied Science, University of Virginia
Wayland, Kent, Engineering and Society, University of Virginia

Machine learning has become an increasingly popular technology, allowing computers to learn and make decisions based on a given data set. It can be applied to various areas, including image classification, chatbots, GPS routing, and loan approvals. In loan approvals, machine learning is used to determine who should receive a housing loan, with the computer being trained with data from past loan applicants to facilitate this decision-making. However, AI systems are not inherently neutral. They are developed by humans who may inadvertently or deliberately incorporate their biases into the system. These biases can be amplified by machine learning models and lead to discriminatory outcomes. For example, the use of machine learning models to determine loan eligibility can perpetuate housing loan discrimination, as described in my STS research paper. Therefore, it is crucial that the developers of AI systems take into account the potential for bias and work towards developing systems that are unbiased, ethical, and socially responsible. To do this, existing methods of reducing bias in these systems can be examined. My technical report functions as a guide for meta-analysis of techniques for de-biasing machine learning models.

BS (Bachelor of Science)
Issued Date: