A paradox looms in our increasingly AI-dependent world: while artificial intelligence holds the promise of improving lives, its algorithms lack imagination and consciousness, rendering them unable to transcend the status quo reflected in their training data. This becomes particularly concerning given that our current societal structures are far from perfectly meritocratic or fair.
Jingyuan Yang, an assistant professor of information systems and operations management at the Costello College of Business at George Mason University, highlights the complexities surrounding this paradox. “The standard view is that fairness is a tax on efficiency. The way conventional systems are structured, fairness checks are added almost as an afterthought that is assumed to negatively impact system performance,” she explains. This raises an important question: is the optimized world promised by AI destined to replicate or even worsen existing inequalities?
Yang’s ongoing research, conducted in collaboration with Pengzhan Guo from Duke Kunshan University and Keli Xiao of Stony Brook University, proposes a compelling alternative. Their work leverages AI to explore a concept they term “fairness-performance complementarity,” suggesting that, under certain conditions, fairness and performance can actually reinforce one another.
“Our ‘fairness-by-design’ framework utilizes reinforcement learning, a type of machine learning (ML). Unlike most algorithms, ours involves multiple agents competing for finite resources within a dynamic environment, not a static one,” Yang asserts. This approach aims to mirror real-world scenarios where diverse individuals vie for limited resources over time.
The researchers integrated fairness into their framework in two distinct stages. Initially, the design encourages high-performing agents to explore choices that may maximize their rewards. “In this framework, high-performing agents are held in an exploratory mode for longer, while lower-performing agents settle into stable paths sooner,” Yang elaborates. Subsequently, options abandoned due to reward-seeking behaviors are redistributed, prioritizing lower-performing agents for access to the best opportunities. This dual-stage process theoretically heightens fairness while maintaining individual choice and performance levels.
Yang summarizes the essence of their research: “The exploratory activity of the high performers releases opportunities that the system channels down toward the weaker performers. Theoretically, this increases fairness while retaining individual choice and without constraining performance.”
To validate their framework, the researchers employed a dataset encompassing detailed job histories of 6.5 million professionals spanning a 20-year period. Yang notes, “In the real-world data, we see a high degree of disparity, without very much redistribution of elite opportunities from relatively advantaged to disadvantaged employees.” The algorithm converted this data into opportunities for hypothetical agents, analyzing career paths through the lens of both performance and fairness. Performance was gauged by the aggregate rewards earned by all agents, while fairness assessed the extent to which initial performance disparities were addressed over subsequent decisions.
The results from the “fairness-by-design” framework surpassed those of eight alternative ML methods, demonstrating improvements in both fairness and performance. The researchers also adapted the system to reflect changes in individual preferences throughout a career. Early-career professionals often prioritize employer reputation and advancement, while later-career individuals focus more on job security and stability. Even with these adjustments, the framework succeeded in enhancing overall career path quality and promoting upward mobility.
In a follow-up study utilizing the New York Yellow Taxi Trip record database, the framework was applied to generate route recommendations for hypothetical cab drivers with varying performance records. The scope was smaller, with 263 locations and a two-hour timeframe, compared to the earlier study’s 4,282 companies and 20 years. Nevertheless, the taxi study similarly revealed that a more equitable distribution of high-quality routes resulted in higher average income per minute for the entire system.
Yang believes the adaptability of their framework positions it as a potential governance mechanism in various AI contexts, including healthcare scheduling, course registration in higher education, and the provision of digital services. While her research is ongoing, she argues that it challenges conventional perspectives on AI. “Our formal proof establishes the conditions under which fairness and performance reinforce each other, and our experiments show those conditions are achievable in realistic settings. That gives our work both theoretical and experimental grounding,” Yang concludes.
See also
IBM Launches Chicago Quantum Hub, Creating 750 AI Jobs and Expanding MIT Research Lab
OpenAI’s AI Model Achieves 81.6% Diagnostic Accuracy, Surpassing Human Doctors in ER Tests
AI Study Reveals Generated Faces Indistinguishable from Real Photos, Erodes Trust in Visual Media
Gen AI Revolutionizes Market Research, Transforming $140B Industry Dynamics
Researchers Unlock Light-Based AI Operations for Significant Energy Efficiency Gains
