Balancing Algorithms: Fairness in High-Stakes Decisions with Machine Learning

In today’s rapidly evolving tech landscape, machine learning (ML) models are pivotal in making decisions that carry significant consequences — from shaping career paths to deciding financial approvals. These algorithms have become gatekeepers in numerous facets of life. Yet, a pressing question remains: Are these machine-driven decisions fair?

Researchers at the University of California San Diego and the University of Wisconsin—Madison have delved into this issue, assessing the fairness of ML models in high-stakes decision-making. Their work, shared at the 2025 Conference on Human Factors in Computing Systems (CHI), reveals a compelling investigation into public views on fairness within AI-driven processes.

Led by Associate Professor Loris D’Antoni, the research highlights a critical flaw in relying on a single ML model. Even models deemed “equally proficient” can yield differing results, thereby challenging the perception of objectivity and fairness. The study participants voiced their reservations about dependence on one model’s outcome, noting that while randomization of model results seems an inadequate solution to contradictions, a singular reliance isn’t the preferred solution either.

This study underscores a significant gap: the expectations of fairness from the public versus the current norms in ML operations. Notably, first author Anna Meyer pointed out the crowd’s preference for fairness, and how it clashed with traditional ML and philosophical methodologies.

As a response, the researchers recommend a hybrid approach to decision-making, particularly in high-stakes scenarios. Incorporating a mix of various models coupled with human insight could rectify fairness disparities. By harnessing human judgment alongside machine predictions, this strategy aims to resolve inconsistencies, offering a more equitable framework for AI applications.

Given ML’s expanding influence over vital societal decisions, these findings could propel transformative changes in algorithm creation and deployment. By advocating for comprehensive evaluations and restoring a role for human discernment, researchers hope to align AI progress with ethical standards and societal aspirations for fairness. Embracing this balanced approach could ensure that AI remains an ally, respecting human dignity and values as it integrates further into the societal fabric.