Platform as Pedagogy: Sociopolitical Formation of/in Artificial Intelligence Educational Technologies

• In Event: Platform Pedagogies: Examining the Public Good in Learning, Schooling, and Educating Across Emerging Platforms

Abstract

Objectives and Theoretical Framework
This paper examines the pedagogical relations inscribed in and emerging from AI educational technologies through a case study of Essay Helper (pseudonym), a learning analytics platform that uses supervised machine learning to provide automated, formative feedback on student essay-writing. Now integrated in over 10,000 K-16 schools, Essay Helper reflects the growing enchantment of educational institutions with leveraging AI to optimize instruction and learning (Long & Siemens, 2011; Pea & Jacks, 2014). Scholars have studied the efficacy of such technologies for assessing student work (Warshauer & Grimes, 2008) and augmenting instructional decision-making (Knight, Shum, & Littleton, 2013); however, we know less about the contingent processes by which such platforms are shaped – or those that unfold from their uses. Merging perspectives from platform studies (van Dijck, 2013), computational cultural studies (Author, 2016a), and new materialist and Black feminist thought (Wynter, 2007), we attend to such dimensions by analyzing public and internal documentation from Essay Helper to map the entanglements of algorithmic reasoning, education policy, corporate profit motives, and embodied literacy practices that animate the iterative development of the platform and the transactions of teaching and learning it yields. In particular, we consider how the pedagogy of the platform both inherits and reconstitutes racial and linguistic relations of difference.

Methods and Data Sources
To examine Essay Helper as a case, we conducted a close reading of fourteen texts provided by the company. These included: three grant proposals; three conference proceeding articles, a press release; three internally-circulated technical reports; and four whitepapers. The texts spanned four-years of the platform’s development – from its origins to its acquisition and scale-up. As such, they not only provide insight into the present composition of the platform, but also the formative shifts in its algorithm and interface over time. This perspective allows for analysis of the platform not as a static artifact, but a sociohistorical one (Author, 2017), whose architecture shapes and is shaped by the material-discursive relations of its context.

Results and Significance
Findings center on two themes. First, variations in the platform’s purpose and function in response to the accrued influence of funders, policies, and technical constraints. For instance, where early prototypes considered the role of intra-class peer-review on student-writing, these features were abandoned as funders encouraged tighter alignment to Common Core standards – thereby embedding racialized and linguistic norms of state assessments into the tacit pedagogy of the platform. Second, how this process redefined “revision” in the logic of the platform. For instance, “revision” became less an open-ended iteration conditioned by the emergent thinking of the writer or the needs of an imagined audience, and more a circumscribed iteration that molded writing to fit the contours of an aggregate dataset. Such insights extend the literature on the interplay of technical systems in education (Author, 2016b; Williamson, 2017) to include analysis of how platforms inherit and produce structural relations of difference. We conclude by discussing the implications for educational equity, and offering suggestions for how learning analytics platforms might be configured and used otherwise.

Authors

• T. Philip Nichols, Baylor University

• Ezekiel J. Dixon-Roman, The University of Pennsylvania

• Ama Nyame-Mensah, Analytics Made Accessible LLC