Why fairness in temporal resource allocation?
Published:
Artificial Intelligence has become an inextricable part of the human experience. With ``smart’’ tools embedded in almost every application, AI is going to make decisions for us whether we like it or not. This realization was the primary source of motivation for my research. If AI is going to be omnipresent, it is imperative to develop techniques to allow its seamless integration into society while maintaining accountability. When algorithms are used to make decisions that can affect real people[1], the analysis of their disparate impact becomes necessary for ensuring fairness. Following this thought, I started off my Ph.D. with a broad vision of improving AI transparency and fairness. In one of my initial projects, I got to examine a state-of-the-art ridesharing matching model called NeurADP[5]. NeurADP used approximate dynamic programming to efficiently match passengers to taxis with high capacity, allowing them to share rides with others. It was highly competitive, and able to service a much higher fraction of the people wanting rides. While analyzing the dynamics of this system, we discovered a large disparity in how well different regions of a city were served, with suburbs being disproportionately under-served. This, of course, raised the question: What was an acceptable trade-off for efficiency?