How Is Socially Responsible Academic Performance Prediction Possible?: Insights From a Concept of Perceived AI Fairness

Introduction

Students' academic learning and work are increasingly taking place online or on digital learning platforms. With the digitization of examinations and administrative processes, higher education institutions generate large amounts of data. The so-called big data enables artificial intelligence (AI) technologies, such as machine learning (ML), to predict academic performance in higher education (Alyahyan & Düştegör, 2020; Daniel, 2015). Some of the attested potentials are already being realized via automated admission systems, such as Parcoursup in France (Frouillou et al., 2020), automated grading (Kotsiantis, 2012), support for administrative or research tasks, and learning analytics (Daniel, 2015; Ekowo & Palmer, 2016). The latter includes a wide variety of applications that enable, for example, real-time performance feedback and advice or performance prediction of exam and study performance. Higher study success is expected to materialize through dropout and performance prediction systems, ultimately preventing student dropouts (Arnold & Pistilli, 2012; Attaran et al., 2018).

However, the use of big data and AI systems in higher education always carries several risks and may lead to potential damages of material (i.e., misallocation of resources) and social nature. For instance, one primary concern is that the data will inherently contain biases, that the algorithms themselves will thus perpetuate or even produce stereotypes, and therefore have discriminatory effects on particular students (Attaran et al., 2018; Ekowo & Palmer, 2017; Fazelpour & Danks, 2021). To take an example from higher education, prospective students could conceivably be disadvantaged in an automated admission process because they belong to a population that is statistically less likely to graduate (Muñoz et al., 2016). In this sense, the fairness aspects of algorithmic decision-making (ADM) are increasingly receiving attention in interdisciplinary research activities (e.g., Lee, 2018; Shin & Park, 2019; Starke et al., 2021). Many scholars have focused on the distribution of goods and their translation into different mathematical fairness notions (e.g., Verma & Rubin, 2018). Recently, however, the focus has shifted to individual perceptions of fairness, which are becoming more important in examining the public understanding and acceptance of AI systems and the legitimacy of AI-driven decision-making (Simmons, 2018; Wong, 2020). Therefore, we bring to the fore students' fairness perceptions of algorithmic decisions in higher education. Consequently, using an academic performance prediction (APP) system as an example, we analyze in this chapter the fairness challenges involved in the use of such systems in higher education that a) need to be considered in implementing APP within academic institutions and b) need to be investigated by conducting empirical research before and during the said implementation process. To this end, we refer to the four-dimensional concept of organizational justice (Greenberg, 1993), which is concerned with designing intra-organizational decision-making processes to achieve or maintain the highest possible satisfaction and commitment of organizational members.