Transgressing local, national, global spheres: the blackboxed dynamics of platformization and infrastructuralization of primary education

ABSTRACT

This article analyzes how platformization and infrastructuralization are currently reshaping the educational sector by engaging in ‘sphere transgressions’, resulting in the merging of a local and national public sector into a transnational and global digital market. It elaborates on the adaptive learning application Bingel as a case-in-point to exemplify how sphere transgressions are conducive to data accumulation across national markets and sectors into transnational and global data infrastructures. Zooming in on these processes as ‘sphere transgressions’ we ask: how are local student data becoming prime assets in the global flow of digital resources? How does this benefit the financial basis of tech firms rather than serving the need for openness and transparency of educational institutions? The conclusion expands on the implications of these sphere transgressions for the future of national education as a public good.

KEYWORDS:

• Sphere transgression
• cloud classroom
• platformization
• infrastructuralization
• personalization
• edtech

1. Introduction

In 2019, Malmberg, market leader of legacy educational publishers in Dutch primary education, announced its learning platform Bingel as a tool that combines the ‘highest pedagogical quality’ with the ‘world’s most advanced personalized digital technology’ (Malmberg, 2019). Bingel’s personalization technology was provided by Knewton—an American company with 40 million users worldwide with whom Malmberg had started to cooperate in 2014. For infrastructural services Knewton relies on Amazon Web Sevices (AWS), the biggest cloud service provider in the world. Two years later, in 2021, Malmberg suddenly changed course: it announced to phase out Knewton as adaptive technology operating in the back-end of Bingel. Instead, Malmberg started using adaptive technology which had been developed in-house, in a European context, by parent-company Sanoma Learning.

At first sight, the story of Bingel’s integration with and parting from Knewton seems to be a simple story about the first wobbly steps in digitization of a Dutch company with a long legacy of developing educational learning materials for (local) schools, students and teachers. Malmberg has been an educational publisher in The Netherlands since 1886 and is known for its close and trusted cooperation with local schools. But behind the chronicle of a local product being (re)developed by a national educational publisher, hides a grander narrative of an evolving blackboxed ecosystem of platforms interweaving the locality of public schools and classrooms, across national and European boundaries, to become part of a corporate global digital infrastructure. In schools’ struggle to incorporate educational technology (edtech), the Bingel case addresses the entangled roles of national and transnational edtech intermediaries and global platform companies, epitomizing fundamental issues of data-sovereignty and infrastructural dependency: educational processes and products are shaped into platformized assets, national companies become dependent on international dataprocessing services, and local public schools hook into corporate global cloud infrastructures.

This article analyzes how (ed)tech companies are currently reshaping the educational sector by engaging in ‘sphere transgressions’ (Sharon, 2021). Building on the work of Walzer (1983), Sharon maintains that tech companies use the advantage they have built up in the digital world to access the sphere of health care, resulting in an ‘uncomfortable merging of what are often held to be by nature or by normative design separate spheres of human life—the market and health’ (Sharon, 2021, p. 316). Our research analyzes this process of sphere transgression for the sector of education, which has resulted in the merging of a local and national public sector into a transnational and global digital market. The sphere transgressions investigated in this article are conducive to data accumulation across national markets and sectors into transnational and global data infrastructures. This benefits publishing houses which in the past decade reinvented themselves as edtech companies, bringing models of value-extraction that are key to the platform economy (e.g., data assetization) to the public education sector (Komljenovic, 2021). Ultimately, these processes empower the Big Tech companies that have built the infrastructures on which (public) sectors are increasingly dependent.

Research in the emerging field of platform studies in education has started to investigate how global Big Tech platforms (e.g., Google Classroom) and (trans)national edtech applications (e.g., ClassDojo) reshape classroom pedagogies in line with platform values and logics (e.g., Apps et al., 2023; Perrotta & Selwyn, 2020; Manolev et al., 2019). Other work has studied specific enactments of sphere transgressions in education, either in general terms of neoliberalisation—(ed)tech companies advancing neo-liberal and market values in school classrooms (Ball & Grimaldi, 2022)—or company specific advances into education as described in terms of ‘Googlization’ and ‘Amazonification’ (Kerssens & Van Dijck, 2022; Williamson et al., 2022). These studies are very relevant in addressing how digital (education) platforms and expanding platform ecosystems are rapidly reshaping public education. Yet these works pay little attention to the complex web of organizational and technological interdependencies and intermediaries underpinning the actual platformization of primary school classrooms. Unlike previous research, we want to focus not so much on the pedagogical power of a particular digital education platform, or the ecosystem power of one specific tech company (e.g., Google, Amazon), but on the ‘glocal’ dynamics of platformization and infrastructuralization driving these sphere transgressions, resulting in the integration of public education institutions within transnational and global ecosystems of digital services.

The structure of this article is as follows. In the next section, we explain how the ecosystem of platforms in primary school classrooms manifests as a hierarchical stack of services. In the following section, the terms platformization and infrastructuralization are deployed in relation to educational sector specific processes of personalization and datafication. The third and fourth section will elaborate on the adaptive learning application Bingel as a case-in-point. We will illustrate how behind Bingel’s seemingly familiar facade of learning material operates a complex socio-technical architecture that is wielded by an intricate web of corporate owners whose interests and business models constitute the economic drivers behind platformization and infrastructuralization. In section 5, we reflect on these transgressions, drawing attention to tech companies’ global infrastructural powers. The conclusion expands on remedying these sphere transgressions at local, (trans)national and global levels.

2. Platformization and infrastructuralization of education

In the 1990s and early 2000s, educational publishing companies in The Netherlands invested substantial amounts of money and expertize in their digital remake: to switch from books and pens to digital devices and software. Only as recent as fifteen years ago, educational technology was still commonly referred to as a collection of digital learning materials—standalone software applications, to be run on local servers and hardware (laptops, desktops). In this pre-cloud era, legacy publishers continued substantial investments in digitizing their product line and in-house technical expertize, while remaining closely connected to, and invested in, a national sector of educational professionals; they offered digitized learning materials in the national language that were tailored toward the national curriculum. After all, local schools, teachers and students comprised their consumermarket while also serving as a source of input for the development of new materials. Due to their collaborations, legacy publishers had intimate knowledge of what happened in local classrooms and in educational policies at the national level. Starting around 2010, legacy publishers faced increasing competition from digitally-native edtech startups who quickly grew a market for networked learning in the cloud. These startups started to weave educational materials into an integrated system of software, hardware, and infrastructure driven by a businessmodel based on data-driven services, rather than standalone products.

Since 2010, we have seen this rapidly expanding ‘ecosystem’ being developed by a handful of American Big Tech companies comprising ‘heterogenous assemblages of technical devices, platforms, users, developers, payment systems, etc. as well as legal contracts, rights, claims, standards, etc.’ (Birch & Cochrane, 2022, p. 2). Platformization has been defined as ‘the interpenetration of the digital infrastructures, economic processes, and governmental frameworks of platforms in different economic sectors and spheres of life’ (Poell et al., 2019, p. 6). Almost all sectors of society became integrated in this global ecosystem, resulting in what Van Dijck et al. (2018) have called a ‘platform society’. In the following years, platformization became tightly interwoven with infrastructuralization: large tech companies providing material and social digital infrastructures such as cloud services, machine learning tools, analytics, and other services that shape user interaction. Platforms became increasingly understood and examined as extensive digital data infrastructures that host a variety of services across sectors (Van Dijck et al., 2018; Nieborg & Helmond, 2019; Poell et al., 2019). Moreover, societal actors as well as internet companies began to ‘rely on the properties of platforms to replace or mesh with existing infrastructures to gain economic advantages’, raising questions about the privatization of public services or utilities (Plantin & Punathambekar, 2019, p. 164).

The result of platformization and infrastructuralization is a ‘stacked’ modular ecosystem of digital networked services (Van Dijck, 2021) which intricate intertwining with the educational sector needs further dissection. To unpack schools’ integration with interlinked platform and infrastructural services, we distinguish between the application layer and the infrastructural layer. From the perspective of a school, the platform stack starts with the application layer—the landscape of educational webapplications that are integrated into everyday classroom practices. Provided as on-demand services through the Internet, educational webapplications are available to teachers and students anytime, anywhere, through a webbrowser installed on any hardware device (Google Chromebook, Microsoft Surface Tablet, Apple iPad) connected to the Internet, even if the majority use of cloud services are connected to teaching and learning practices in the physical classroom during schooltime. This upper-level application capability is known in the field of cloud computing as Software as a Service (SaaS) (Mell & Grance, 2011). The SaaS application layer typically consists of two main types of systems: digital learning platforms (DLP) and learning management and support systems (LMS) (Kerssens & Van Dijck, 2021).¹ Digital learning platforms are applications ‘that in form and content are aimed at instructing or testing knowledge, skills and developing attitudes in schools’ (ibid., p. 253), often in core subject areas (e.g., math, language) that are part of the curricula of the national education system. Learning management and support systems (LMS) includes a range of software services ‘involved in the organization and management of digital learning’ (ibid.), amongst which software dedicated to functions such as single-sign-on access, class management and monitoring, lesson development, instruction and presentation, productivity, tracking and analyses, and (parental) communication.

This infrastructural layer is built from two software sublayers commonly distinguished in cloud computing—Platform as a Service (PaaS) and Infrastructure as a Service (IaaS)—and a bottom physical layer of hardware resources (e.g., server, storage and network components) providing necessary material support to the two software sublayers of the infrastructural part of the stack (Mell & Grance, 2011). Cloud-based educational applications are built on PaaS services—such as databases, learning analytics, AI and machine learning—offered to edtech companies as ready-to-use building blocks for software development. Smaller platform companies provide (sector-)specific Paas services—such as Knewton providing its learning analytics services to edtech application developers. But the majority of PaaS services is offered by the three biggest cloud providers Google Cloud Platform (GCP), Amazon Web Services (AWS) and Microsoft Azure (MA), which also dominate the IaaS service layer (e.g., computing power and storage) and the physical layer (ACM, 2022) on which PaaS services rely. The service model of global cloud vendors extends across all layers of the cloud stack. They flexibly rent out server and storage space on public data servers, whilst providing hundreds of PaaS services (mostly to software developers), and some SaaS services to professional users (e.g., Google Workspace for Education, Microsoft Office, Amazon Lexa).

In this stacked ecosystem, educational SaaS applications such as DLPs also extend across PaaS and IaaS layers. A subset of DLPs known as adaptive learning platforms (ALPs), such as Bingel, are illustrative in terms of a layered organization—one that operates in function of twinned processes of personalization and datafication. ALPs enable the personalization of learning by using algorithmic analytics that can dynamically tailor didactic content to fit a child’s individual progress or need for more practice in a specific skill. Personalization functions as a key driver of a shift towards routine collection and analysis of enormous quantities of children’s personal data (Kucirkova, 2022)—not only demographic data but also behavioral, cognitive, developmental data. Personalization works better with large amounts of data culled from many learners. Optimizing this process requires a reiterative shift between data pulled from local classrooms to transnational data collection and processing: computational analytics are not based on the data of one class, but on hundreds of schools, or even all national data. These masses of aggregated and often anonymized data fed into the system serve to optimize the adaptive learning service—a process that requires enormous computational processing power in real time. Such optimization of proprietary platform services—including the training of their machine learning models—and the development of new data-driven products (e.g., learning analytics dashboards) through aggregation and processing of non-identifiable educational data describes a form of assetization (Komljenovic, 2020) by which edtech companies transform educational data resources ‘into capitalized property’ (Birch et al., 2021, p. 2).

The twinned processes of personalization and datafication are intricately intertwined with platformization and infrastructuralization. When talking about the platformization of education we refer to the transformation of educational content, activities and processes to become part of a (corporate) platform ecosystem, including its economies, (data) infrastructures and technical architectures (Kerssens & Van Dijck, 2021). Platformization drives infrastructuralization, turning global platforms into digital infrastructures consisting of a wide range of interconnected services on which users become increasingly dependent (Plantin et al., 2018). In turn, platformization and infrastructuralization promote the invisible transgression of spheres: from local into global and from public into market sphere—promoting the commodification of boundary assets (Sharon, 2021). As Birch and Bronson (2022) explain, the notion of boundary assets is important to understand Big Tech’s power to operate cross-sectoral: they promote seamless modular integration and scalability across sectors so they can benefit from the integration of user data and they can ‘set the terms of engagement through contractual arrangements (e.g., terms and conditions), representing privately-made rules and standards controlled by Big Tech’ (p. 10).

The journey of one ALP (Bingel) to nestle itself in the ecosystem of platforms and infrastructure, offers a unique prism onto the ‘glocal’ constellation of data flows. In the global digital economy, classroom learning analytics and local data flows cannot be seen as separate from ‘an increasingly commercial data ecology’driven by ‘a variety of stakeholders and data interests’ (Prinsloo et al., 2023). Transgressions of spheres underlying these processes, however, are invisible to schools, teachers and students. In classrooms, DLPs appear to their users as standalone, bounded and easy-to-use learning materials offered by local and trusted publishers (e.g., Malmberg). Software’s appearance in classrooms, however, no longer matches the nested architecture of SaaS applications. In the world of cloud, webapplications form collections of modular online services loosely coupled through application programming interfaces (APIs) (Narayan, 2022). API-based integrations cement webapplications like Bingel into seamless and often invisible stacks of layered software services—for hosting, storage, databases, analytics, machine learning and artificial intelligence—running on different servers in varying locations and developed, managed and updated by varying (for-profit) tech organizations. When teachers and students use Bingel, they need to know nothing about how the adaptive learning service in class is embedded into lower levels of the platform stack. While this simplifies the use of the underlying cloud resource, serving teachers’ and students’ relationship to trusted front-ends, it also facilitates a blackboxing of ‘lower-level’ services which now act ‘as an integral, albeit largely invisible, presence in the[ir] back-end’ (Williamson et al., 2022, p. 245).

This is why it is so instructive to look beneath the surface of a single application, into the deeper layers of the complex layered structure which is barely discernable to its users. And this is just the socio-technical part of the stack; in the next section, we turn to the political-economic aspects of these dynamics, again using Bingel and its owner/developer Malmberg as a prism: How do ownership structures and business models drive the dynamics of a glocal digitized classroom? And how do these boundary crossings between the local/national and the transnational/global level relate to the sphere transgressions between the public sector of education and the market?

3. Platformization: turning national education into a transnational market

The story of Bingel and its owner/developer Malmberg exemplifies the blackboxed dynamics underpinning the platformization and infrastructuralization of the Dutch classroom not just as a socio-technical process, as explained in the previous section, but also as an economic process. In this section, we scrutinize the intricate intertwining of both levels showing how a local content publisher becomes part of a digital landscape where multinational companies reign. In the past decade, ownership and business models of legacy educational publishers have been reconfigured around cloud infrastructures, pushed by the need to keep up with a changing edtech market (Accenture, 2022). We will use Bingel to explain how a national market of educational publishing, facing fierce competition not only from Dutch startups but also from the Big Tech conglomerates, transformed into a transnational market.

Malmberg, as mentioned, has operated as an educational publisher in The Netherlands since 1886; in 2004, it was incorporated by Sanoma Publishers, an international publishing house based in Finland. As a Sanoma subsidiary, Malmberg was enabled to strengthen its leadership in the national market of Dutch primary education, which it shares with three main competitors ThiemeMeulenhoff, Zwijsen and Noordhoff. Starting around 2012, these legacy edu-publishers met fierce competition from edtech startups (Snappet, Gynzy) for their share of the DLP-market. In 2012, Snappet was the first company to launch an adaptive learning technology at scale in the Netherlands, allowing them a considerable advantage in the digital classrooms’ transition to the cloud.² In 2015, Snappet got competion from edtech startup Gynzy which expanded its highly used productline of smartboard software with a personalized learning platform. In 2019, legacy publisher Malmberg entered the personalized learning market with Bingel. Today, all three platforms are used on a day-to-day basis in math and language education in primary schools.

In contrast to edtech startups and independent ‘cloud native’ companies Snappet and Gynzy, legacy publishers could not innovate as quickly. Cloud natives could invest in digital innovation without cannibalizing on their traditional book-oriented business model, which was still profitable for traditional publishers. In addition, publishers had insufficient in-house technological expertize, which, in combination with a rapidly shrinking market share, formed a significant incentive for purchasing adaptive learning technology as a finished product instead of building a system of their own. For these reasons, Sanoma Learning, pushed by Malmberg, started looking for a partner to provide the necessary adaptive learning capacities and expertize. In 2014, Sanoma Learning contracted with New York-based edtech company and global leader in adaptive learning technology Knewton (Sanoma, 2014).

Knewton had been founded in 2008 in the USA, offering a platform technology for personalized learning specifically enabling third parties to build adaptive learning applications. With the big hype surrounding personalized learning, Knewton quickly raised hundreds of millions of venture capital, including early investments by large American educational publishing houses, such as Pearson in 2011 and Houghton-Mifflin Harcourt in 2013, who all started building their own personalized learning applications on top of Knewton’s adaptivity platform. Thereafter, Knewton partnered with leading publishers and tech companies across the world—including Sanoma Learning in Europe. Sanoma’s initial goal was to use Knewton to build ‘a new generation of personalized learning solutions’ to be used at scale in all European Sanoma markets (Sanoma, 2014). So after the deal was closed, in 2014, Sanoma ordered Dutch publisher Malmberg to develop the adaptive backend architecture of Bingel to integrate with Knewton’s data-driven intermediary platform. To understand what happened next, it is relevant to first look into the technical integration and how this is related to the underlying business model of platformed services.

Data intermediaries like Knewton provide a platform service; through API-based integration, they power Bingel with adaptive recommendations and analytics for constructing personalized learning paths for students and helping teachers to identify struggling learners in need of intervention (Knewton, n.d.). As a first step in this process of technical integration, Bingel developers adapted Knewton to the content ecosystem of publisher Malmberg—that is, they organized Malmberg content into Knewton’s preformatted data model known as the ‘knowledge graph’. This knowledge graph maps publisher content into a form that is processable by Knewton engines by specifying relationships between instructional items and the concepts that those items assess or instruct (Wilson & Nichols, 2015). Only by specification of the knowledge graph, the three ‘inferential engines’ of the Knewton platform—the psychometrics engine, the recommendation engine, and the analytics engine—are able to automatically provide learning applications embedded into Knewton with real-time and accurate content recommendations and student data analytics. Through its psychometric engine, Knewton builds up an understanding of students’ mastery of the different concepts as specified in the knowledge graph by making psychometric inferences on student knowledge and abilities. Basically, this engine estimates a student’s knowledge level and abilities relative to particular instructional items or questions. This engine is in turn deployed to power Knewton’s other two inferential engines—the recommendation engine generates ‘personalized content recommendations’ for students, while Knewton’s analytics engine ‘provides real-time inferences for predictive analytics and student reports’ (Wilson & Nichols, 2015, p. 14).

Importantly, for publishing companies such as Malmberg, Knewton’s technologies provided a technological base and enabling resource for building personalized learning applications adapted to the content-specific contexts of national markets. At the same time, technical integration of educational webapplications with lower-level platform services such as Knewton stimulated a more encompassing process of ‘platforming third-party education providers on the cloud’ (Williamson et al., 2022, p. 231). As it turned out, this type of outsourcing to American platforms lowered the treshhold to enter global markets.

4. Infrastructuralization: turning a transnational market into a global asset

Platformization alone, indeed, is not enough to describe the technical infrastructure upon which Bingel and Knewton came to rely; infrastructuralization explains how a transnational edtech market turned into a global market of data assets. For a seamless operation of personalized recommendations, the three inferential engines that function as part of the ‘application layer’ of Knewton’s service provided to third-party clients, rely on a ‘core layer’ responsible for the real-time processing of massive volumes of learning data flowing through Knewton (Wilson & Nichols, 2015). Knewton’s recommendation and analytics engines are built to benefit from the scale of data collected from its personalized learning applications built on top of the platform. To enable processing and analyzing large volumes of personalized learning data at scale and in real time, Knewton started operating its data and analytics infrastructure on Amazon Web Services (AWS, n.d.). Unknown to schools, and probably unknown to some of Knewton’s partnering publishers, its core layer integrates with several PaaS services of AWS, such as Amazon Relational Database Service and Amazon Redshift. Moreover, Knewton uses Amazon Elastic Map Reduce (Amazon EMR)—‘the industry-leading cloud big data solution for petabyte-scale data processing, interactive analytics, and machine learning’ (AWS, n.d.)—to analyze large data sets across the Knewton platform. Moreover, at the IaaS level of the infrastructural layer, Amazon Simple Storage Service (Amazon S3) provides storage capability and through AWS CloudFormation at least 1000 Amazon Elastic Compute Cloud (Amazon EC2) instances are provisioned to Knewton.

In other words, Bingel ended up as a cloud-based application, seamlessly composed as a stack of software services—some of which are delivered and developed by Malmberg and Sanoma (learning materials, user interface), others by Knewton (adaptive learning engine) and yet others by Amazon (cloud storage, computing capacity, machine learning). Dutch local schools and teachers that used Bingel knew next to nothing about Knewton’s data architecture or business model, which was mainly built on data extraction. Knewton’s ambitions to build an educational technology market around a profitable data extraction model is exemplified by Knewton’s original mission statement as described by its former chief executive Jose Ferreira: ‘The larger goal for Knewton […] is to create the world’s most valuable repository on how people learn’ (Farr, 2014). Knewton’s analytics engines were used at scale in digital education platforms developed by national publishers in a large number of countries across the world. But while its users—students and also teachers—continuously fueled its engines with massive amounts of data about learning, they had little to no insights in its processing complexities. For Knewton, on the other hand, these data represented tremendous asset value, not only in the sense of an unparalled repository of knowledge on learning (which can serve as key input for the development of new platform services), but also as resources enabling Knewton to continuously refine its personalization algorithms.

Meanwhile, as analytics and machine learning services offered by Big Tech cloud providers became affordable and simple to use, an increasing number of (inter)national edtech companies started to explore and unlock the benefits of these technologies, creating a web of interdependencies. In the Netherlands, all educational webapplications across the DLP and LMS division, get their computing capacity, storage and server space supplied by one of the three big American cloud providers. Yet the most far-reaching integrations—spanning deep into IaaS and PaaS—are constructed by adaptive learning platforms. Competitor Snappet’s adaptivity engine is complexly integrated with and dependent on a stack of AWS services for computing, data processing, analytics and machine learning. Gynzy, in turn, is built on the Google Cloud Platform (GCP) using BigQuery for large scale data processing and analytics across it educational application. While in contrast to Malmberg, both Gynzy and Snappet operated independently from ‘middle-ware’ such as Knewton, these edtech platform companies were just as firmly linked to and incorporated into US-cloud and computing services.

As gradually became clear, only Big Tech providers are able to provide the scale of computing services necessary to run these adaptive learning technologies on-demand and at affordable rates (rents) because they can spread their services over ever-increasing numbers of projects (Rieder, 2022). Knewton became fully dependent on cloud giant Amazon for powering its engine in real-time. In terms of the scale and scope, AWS turned Knewtons transnational assets (educational data) into global services. Besides serving Knewton, AWS also provides a platform infrastructure for computing power, data storage, analytics, and machine learning to many companies across the globe and across sectors (Williamson et al., 2022). This development, again, changed the strategy of legacy publishers vis-à-vis digital native edtech companies. Why would (European) legacy publishers outsource the most precious part of their digital education platforms—the data-fueled algorithmic personalization engines—to an American edtech multinational, thereby loosing control over their most valuable assets: learning data generated by students and teachers?

The pressure to open up the complex mechanisms behind data flows and data extraction started in 2016, and came both from market actors as well as from government and civic society actors. To start with the market actors: in 2017, Knewton changed its business strategy, and started selling a personalized learning system named Alta to schools and colleges directly, hence bypassing legacy publishers (McKenzie, 2017). While scaling up Alta, from now on, would be Knewton’s main focus, it pledged to ‘maintain its existing commercial partnerships where they continue to make sense for both parties’ (Ibid.). That same year, leading publisher investor Pearson decided to phase out its partnership with Knewton while starting to develop its own in-house adaptive learning technology in close collaboration with multinational corporation IBM, providing Pearson a possibility of integrating with the AI technology Watson (Wan, 2017). Multiple contractors followed Pearson’s move, including Sanoma/Malmberg in 2021, which started to develop its own personalized learning engine across its franchises in Europe, even though Malmberg had just started to roll out its Knewton-powered Bingel tool two years before.

One of the reasons why Sanoma—and other publishing companies such as Pearson—terminated the partnership is allegedly Knewton’s inability to live up to its promises in its early venture capital-seeking days. Its platform service was not an off-the-shelf product; technical integration of local digital content into Knewton’s generic knowledge graph to achieve optimal functioning required an immense investment of time and resources by national publishers. Another key reason was Knewton’s blackboxing of its proprietary back-end. For-profit companies such as Knewton tend not to share confidential information about their underlying analytical engines, machine learning and big scale data processing (Perrotta & Selwyn, 2020). This hindered Malmberg to inform Dutch public schools on how Bingel’s personalized content recommendations could be accounted for and how algorithms led to which recommendations based on what input. Undoubtly, customers (schools, teachers and students) incentivized Malmberg/Sanoma’s shift towards developing its own adaptive platform, pushed by the implementation in 2018 of the GDPR, requiring compliance with the European privacy law concerning the use of personal data. Moreover, Knewton’s proprietary blackboxed data infrastructure also limited Sanoma/Malmberg’s control over Bingel data, obstructing both its opportunities to render data in actionable format for schools (e.g., through the design of teacher dashboards) and its likely ambitions to assetize this data for own purposes (e.g., developing future learning materials; system optimization). In other words, Knewton’s control of the adaptive learning platform worked to the detriment of Sanoma’s own business model.

A final factor probably informing the choice of legacy publishers such as Malmberg/Sanoma to develop ALPs in-house, was the increasing availability of flexible and cheap cloud storage coupled onto computing and analytics services provided by the Big Tech cloud services: GCP, MA and AWS. Altough it is currently unknown how the infrastructural back-end of Bingel’s new adaptivity engine is organized, it's no secret that Sanoma started with a cloud-first strategy utilizing the Amazon Web Services (AWS) platform for migrating Sanoma’s digital systems and applications to the cloud (Accenture, 2022). The power of Big Tech companies to offer a cloud architecture to directly and seamlessly integrate the needs of third-party developers and platform providers with the needs of schools has enhanced their nodal position in the ecosystem. As Veale (2022) concludes: ‘Schools are slowly becoming extremely reliant on a few large companies’ entire technological stacks in order to operate. In turn, these stacks are reshaping what schooling is and could be, and exercising unaccountable control over students, teachers, administrators and content providers alike’ (p. 76). Because companies like Google and Amazon can expand their data-power across sectors and across continents, their machine learning and analytics capabilities can take disproportionate advantage over sectoral and local actors. When global commercial tech companies provide the basic hardware, software, and data infrastructures for the educational sector, this shifts the power dynamics between local-national public sectors and global market actors.

Bingel’s position as a local data-feeder of a transnational data-driven ecosystem built on a global infrastructure, poignantly illustrates its growing complexity and intransparency. Through the processes of platformization and infrastructuralization, Bingel and its (national) owner Malmberg were driven by the decision-making of its (transnational, European) owner Sanoma and its capture into a (global) digital infrastructure. In this gradual transformation, teachers, schools and the educational sector lost sight of how their data are being processed as part of a global data economy; their personalized data flows function as input into a black-boxed system of asseticized services. As such, Bingel’s evolution provides insight into how trusted and easy-to-use local front ends are stacked in proprietary transnational and global back-ends—complex structures whose operative power is wielded through hierarchical and interdependent layers that are rendered invisible to schools, teachers and students. Relatedly, the story illustrates a strategic repositioning of publishing houses at global scale which manifests as a ‘privatization of data infrastructure and data, [enclosing] innovation and new knowledge about how we learn, turning public goods into private assets’ (Sellar & Hogan, 2019, p. 2).

5. Platformization and infrastructuralization as sphere transgressions

As we have seen in the previous sections, the platformization of education facilitates the transformation of schools’ public data flows into corporate assets of edtech companies, while infrastructuralization disproportionally benefits the power of global big tech companies; operating in tandem, both processes, we argue, may help undermine the mission of public schooling. Schools or even entire national sectors of education, which stand to gain from the benefits of data-based products, still have few insights into how their data is being collected and used towards the benefits of platform companies. Once they are part of the Big Tech digital ecosystem, schools cannot tap into their educational data for their own benefits as ‘EdTech companies do not return the data to schools in a usable format to improve education unless sold as a data analytics product’ (Day et al., 2022, n.p.). The local case of Bingel and Malmberg epitomizes a much larger process of sphere transgressions—across public/corporate boundaries as well as across local/national and transnational/global borders. We will explore the effect of each of these transgressions in the paragraphs below, followed by an explanation of how these transgressions can be countered at the level of legislation.

Generating and collecting learning data is an activity that is bound to take place in a local context, within the institutionalized sphere of a national education system. The Dutch school system, which has been provided for by local and national (legacy) publishers has traditionally been staked in the mutual exchange of input and output between schools and legacy publishers. At the same time, performative norms and evaluative standards in primary education were set through the interaction between learners and teachers on the one hand, and national professional organizations on the other. Although legacy publishers always acted as commercial intermediaries, their interest in data remained ‘close to home’: educational materials were developed in the same local/national environments as standards for learning. With the advent of platformized adaptive learning environments, standardization is taken to the systemic level: large data collections culled from many (anonymous) students across nations serve as input for the development of graphs that, in turn, inform personalized learning trajectories. As Davies et al. (2022) argue, this process appears to be the ‘deinstitutionalization of education—bypassing schools and universities, quality control and national, institutional and democratic standard setting’ but in fact it is more of a ‘reinstitutionalisation of education around private platforms’ (p. 88). Normsetting and standardization in education is increasingly outsourced to algorithms; since these are intellectual property owned and operated by multinational edtech companies, public monitoring gets very difficult, if not impossible. Binding users through technical standards or contracts, tech companies control the development and functionality of hardware or software (Veale, 2022, p. 68) and, compared to traditional state and public actors, are increasingly ‘better positioned to influence processes of sociotechnical change in schools in ways that have lasting outcomes’ (Selwyn et al., 2023, p. 176).

Second, platformization and particularly infrastructuralization show how Big Tech systems materialize and how they ‘territorialize’ learners’ information across geographical and sectoral boundaries. Cloud storage is often bundled with machine learning and analytics services, so the role of cloud services as a critical information infrastructure has become crucial: ‘Intersecting capital development, data sovereignty and governance, the geopolitics of the cloud is still a blind spot’ (Tang, 2022, p. 2401). Enormous amounts of students’ actual learning data are stored in and processed through global cloud services; in the past, their physical storage in data centers across the world raised poignant questions of security and privacy. More recently, issues such as the territorialization of data collections vis-à-vis the way in which data are mined as global resources for the production of personalized information services, has also touched upon issues of institutional and state sovereignty. In the information realm, data flows processed through cloud services were long regarded as extraterritorial but this changed sharply with the introduction of several new laws.

The European GDPR, implemented in 2018 restricted companies’ access to sensitive data stored within institutional boundaries, prohibiting data collection from within users’ systems. In the same year, the American Cloud Act (Clarifying Lawful Overseas Use of Data Act) was implemented, allowing American government agencies to sift through foreign data stored on US companies’ servers. In other words, Dutch students’ personalized data, even if stored in European data centers operated by AWS, Google or Microsoft, can be de-territorialized for security purposes. Squeezed between American and European legislation, companies faced national scrutiny over their use of massive data flows to feed their system. In 2021 and 2022, France and Germany temporarily banned the use of ‘integrated packages’, bundling offerings of free cloud space with other tools and services, such as Google Cloud with Google Workspace, or Microsoft Azure with MS 365 integrated services. In The Netherlands, Data Protection Impact Assessments (DPIA) concerning the use of Google and Microsoft services in schools and universities have led to fierce discussions about the ‘territorizaliation’ of data flows. The smart deployment of a data infrastructure is highly dependent on the technical architecture, business models, and bundled assets provided by the Big Tech corporations. For that matter, it is fallacy to believe that GDPR privacy regulation alone can effectively secure schools data sovereignty; also because the large amounts of aggregated non-identifiable data processed for personalization of learning are not regulated by European privacy law (Komljenovic, 2020). Without acces to the underlying infrastructural layers—invisibly anchored in the tech stack—schools and software developers are unable to offer transparent data management (Taylor, 2022).

6. Conclusion

Sphere transgressions across public and corporate boundaries clearly intersect with those across local/national and transnational/global boundaries. Personalized student data—information about students’ learning processes as well as cognitive and personal data—become valuable contributions to, and precious assets in, a globalized process of automated education. The layers of stacked application services on top of infrastructural services (including machine learning and AI) has rendered the blackboxed dynamics of the cloud-based classroom even more opaque. As Rieder (2022,) argues, these ‘large technical systems’ are increasingly complex amalgams of hardware, software and data collections, which transversality and interrelatedness have ‘major repercussions for how these companies function and compete, affecting the products and services that infiltrate users’ lives’ (p. 4).

How to remedy these sphere transgressions at the (local) level of schools, the (national) level of publishers, and the (transnational) level of governing institutions? For schools, it is important to critically scrutinize their systemic dependency on the blackboxed dynamics of dataprocessing tools. Demanding transparency from tech companies is never an easy task, so that is why schools should join forces to operate as a sector to enter a dialogue with tech providers. How are data flows navigated back and forth between the various layers of the stack? Which data from students are extracted by which tools, where are they stored, by whom are they processed, and to what ends? How are the standards for measurement and learning curves constructed, on the basis of whose data? And what is the potential for bias and inequality that creeps into in the automated calibration of these curves? Questions like these need to be raised by professional educators as a critical counterweight to the closed, proprietary logics of the integrative software- and hardware packages that are currently dominating the edtech market. Such development renders the need for deeplevel scrutiny even more urgent. As Landwehr et al argue: ‘Note that today people do not even have access to a transparent overview of how their private data is trapped, transferred, sold and aggregated. Therefore, as a prerequisite, these data pathways need to be visible for the user and the public regulators’ (Landwehr et al., 2021, p. 75).

Offering more transparency is also one of the main conditions that should be required from legacy publishers and edtech providers who function as powerful intermediaries between global tech companies and local schools. Traditionally, publishers have always benefitted from close collaboration with schools and the national educational sector, particularly in The Netherlands. Their ability to remain competitive in a world dominated by Big Tech companies still to a large extent depends on their ability to foster trusted bonds with local schools and professionals. Therefore, national publishers and edtech providers should side with schools in an effort to be as transparent as possible about the architecture of integrated systems. By the same token, they could demand full disclosure from the infrastructural systems on which they rely for the seamless functioning of their products. Powerful edtech companies such as Pearson, Wiley, and Sanoma/Malmberg could help schools by upholding rules of transparency, for instance, by making clear how they monitor algorithmic systems to avoid bias and error (Sellar & Hogan, 2019).

Finally, at the level of governing institutions, we need to look beyond the above mentioned transparency demands from schools and publishers, which are obviously inadequate to remedy the rising power of very large online platforms (VLOPs) in the public sphere. Particularly in Europe, schools and school systems, as part of a national educational sector, are historically strong public institutions empowered by professional organizations supported by governments. Collectively, and in close collaboration with national edtech companies, they could pursue technical alternatives to data governance structures at local/national levels, such as the formation of data trusts (Zhao, 2022) that ‘separate[e] control over data from the provision of data-driven services’ (Taylor, 2022, p. 282). Realizing such technological alternatives, however, is no easy task—not at the local/national level, and certainly not at the global level where Big Tech platforms reign over data governance. Public institutions are commonly ill-equipped and underfunded, lacking the necessary technical expertize to face such critical challenges. As Rieder and Hofmann (2020) suggest, one potential remedy at this global level of data governance could be to organize ‘centers of expertize tasked with building the capacity to produce relevant knowledge about platforms’ (n.p.). Again, we emphasize that such upgrading of expertise would require at least a concerted sectoral effort to join forces, most likely backed up by governments who stimulate public sector engagement rather than leave education to market forces. In other words, the transgression from the public institutional sphere to the private-corporate sphere may be the most difficult one to overcome.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Both or work has been funded by the Dutch Research Council, yet under different Grant Numbers: VI.Veni.201C.011 and SPI.2021.001.

Notes on contributors

Niels Kerssens

Niels Kerssens is assistant professor in the Department of Media and Cultural Studies at Utrecht University, the Netherlands. His research investigates the implications of platformization on education as a public good.

José van Dijck

José van Dijck is a distinguished university professor in media and digital societies at Utrecht University (The Netherlands). In 2021 she was awarded the Spinoza Prize, the highest award for lifetime academic achievement, by the Netherlands Research Council (NWO), which provided funding for this research.

Notes

1 Please note that the distinction between LMS and DLPs becomes increasingly blurry due to the increased clustering of learning and managerial services in various educational platforms.

2 Snappet is used by 2800 schools (45% of all primary schools in the Netherlands) (Molenaar, 2021).