Pushing Algorithmic Fairness with Models and Experiments

Performance refers to the objective achievements of an individual. In contrast, success is a collective measure, indicating a community’s reaction to an individual’s performance. In most areas of human activity, we assume that performance uniquely determines success and we often rely on success measures to gauge performance – especially in science, where citations, impact factors, or h-indices are constantly used to gauge quality, to assign recognition, and to allocate resources. Yet, success is strongly susceptible to effects that have nothing to do with performance, like social influence, gen- der, or reputation. Despite this fundamental difference between performance and success, we have a poor quantitative and mechanistic understanding of how success emerges from performance in science. Here, we propose to uncover the relation between performance and success in science, addressing the central question: Given equal performance, how do social effects lead to unequal success? We do so by distinctly pushing the state of the art: We will 1) set up a platform to conduct never attempted large-scale randomized controlled experiments with thousands of participants to identify and measure the social mechanisms leading to unequal success, 2) create realistic dynamical models of success incorporating these mechanisms, 3) untangle social factors from performance in current success measures, to ultimately propose fairer proxies of performance and correct algorithmic bias. The aim of this proposal is not defining new bibliometric indicators, but pursuing fundamental research with a universal, long-lasting scientific and societal impact: We aim to uncover universal social mechanisms, which are also at play in domains other than science, and we will study the effect of these mechanisms on metrics of success with important implications for early recognition of talent and resource allocation.