Enclouding education

Cloud computing companies provide stacks of infrastructure, platform, and artificial intelligence services in industries and sectors like education. Photo by Crawford Jolly on Unsplash

Cloud computing underpins the contemporary digital landscape of online services and platforms. In education, cloud systems make it possible for many educational technologies to function and scale, and are increasingly present in institutions as back-end infrastructure for running digital services and managing data. While research on the ‘platformization’, ‘datafication’, and ‘automation’ of education has been growing in recent years, ‘the cloud’ has received less attention, despite underpinning many of those transformations.

The cloud promises, of course, many benefits. But like the platforms and data systems that cloud operators support, we also need to better understand its social, technical, political and economic dimensions as a route to considering the implications of the cloud in education. Wary of prematurely claiming the ‘cloudification’ of education, in this post I want to think through possible priorities for research along four lines:

1. corporate cloud enclosure of public education
2. enhanced extraction of economic and data ‘rent’ from the education sector by cloud operators
3. expanding infrastructural power over the edtech ecosystem
4. extension of capacities of automation and anticipation in education practice, policy and governance

Rather than taking up an advocacy position against cloud infrastructures or platforms, future research could take up analytical positions to understand more fully the kinds of social, economic, technical and institutional transformations that the cloud enables within education systems. Drawing from other recent relevant research, this (longish) work-in-progress provides a sketch towards such studies of the cloud in education.

Penetrating clouds

In technical terms, the cloud is a particular configuration of computing, networking, and data storage and analytics technologies, delivered over the internet without requiring management by the cloud-user. The cloud consists of hardware and software that are available on-demand and on a ‘pay-as-you-go’, ‘plug-and-play’ basis, offering customers the benefit of outsourcing management, maintenance, scaling and upgrading of critical systems, as well as saving on IT infrastructure costs. The cloud operates an ‘as-a-service’ model, providing Infrastructure-as-a-Service (IaaS), Platform-as-a-Service (PaaS), and Software-as-a-Service (SaaS) depending on clients’ needs.  

The cloud is also highly commercialized and privatized. The main players in the cloud infrastructure and platform services market are multinational US-based technology companies, such as Amazon Web Services (AWS), Google Cloud, Microsoft Azure, IBM Cloud, Oracle Cloud Infrastructure, and Salesforce Cloud, as well as Alibaba Cloud and Tencent Cloud in China. With material locations in massive data centres distributed around the world, many of these serve a range of organizations across enterprise and industry sectors, with some increasingly pushing into government and public sector areas such as health, policing and justice, welfare, and education.

The entry of proprietorial clouds into public services is raising significant concerns over governance, regulation, and the concentration of techno-political power. As Jose van Dijck has recently argued, clouds are both highly privatized and deeply penetrating public infrastructures, with little public control.

The architecture of cloud services forms a blueprint for data storage, analytics, and distribution; control over cloud architecture increasingly informs the governance of societal functions and sectors. Amazon Web Services, Google Cloud, and Microsoft Azure dominate this layer, and while states and civil society actors become increasingly dependent on them, public control over their governance is dwindling.

Notably, in Europe there is considerable political resistance to corporate cloud operators on the basis of concerns over digital sovereignty. The French government has announced investment in national cloud operators to ‘regain our technological sovereignty in the Cloud’, while in Denmark Google cloud-based services for schools have been effectively banned due to data transfer and security concerns.

Education-as-a-Service

Building on the above characteristics and controversies of cloud computing, the first priority around the cloud in education is to understand it as a significant shift in underlying computing arrangements, which may affect a vast range of educational practices. Cloud services appear in education in two main forms. First, the cloud operates as computing and data infrastructure for education platforms to operate upon. Many existing edtech firms are pivoting to the cloud, and corporate clouds underlie many other third-party education platforms and software.

AWS is the major player in this approach, as many other education platforms rely on AWS for computing and data services, such as server space, data storage, and analytics. Global edu-businesses and big edtech platforms including Pearson, the Coursera online learning platform, the 2U online program manager, the language learning app Duolingo, the home tutoring app Byju’s, the school app ClassDojo, and many more, all sit on AWS. AWS also hosts its own cloud-based education service, the AWS Educate online learning program, and actively supports third party edtech startups with AWS training, facilities and discounts. Google’s cloud-based Classroom also integrates with a vast Marketplace of thousands of third-party education platforms, as the cloud facilitates elasticity, connectivity, and interoperability between different platform operators.

Second, the cloud operates in education as institutional infrastructure. Rather than running their own in-house IT systems, educational institutions including schools, colleges and universities increasingly outsource infrastructure services and management to cloud operators. Whole school districts have migrated to cloud systems to enable greater collaboration, access and data flow.

Many of the major cloud companies offer specific cloud-based packages to the education sector, ultimately providing the ‘full stack’ of back-end infrastructure and education platform products. Salesforce, for example, offers full migration of institutions’ legacy systems to its Education Cloud, and as part of that provides the cloud-based ‘Education Data Architecture’ for a ‘360 degree view’ of the student.  MS Azure for Education includes ‘virtual desktop’ and ‘app virtualization that runs in the cloud’ to support online and remote teaching and learning, and ‘flexible migration’ to ‘the cloud with managed service’.

Likewise, Google Cloud for education provides ‘scalable infrastructure’ for educational institutions, including its Workspace suite for digital learning, its Classroom online learning platform, ‘virtual desktop infrastructure’, ‘data warehouse optimization’, ‘serverless apps for edtech’, ‘smart analytics and AI’ and ‘intelligent learning tutors powered by AI’ to support ‘personalized learning’. AWS for Education also offers ‘Desktop-as-a-Service’ solutions and ‘personal cloud desktops’ for distance learning, ‘no-cost online learning modules on cloud computing’, and ‘virtualized app streaming’ for ‘anytime access’, as well as full institutional migration to the AWS cloud.

For some commentators, the ‘as-a-service’ model of the cloud should be applied to education. ‘Education-as-a-Service’ is likened to the flexible pricing and on-demand delivery of the cloud. In this sense, the technical and economic model of the cloud has become a template for reimagining education as ‘unbundled’ into modularized ‘plug-and-play’ components. More prosaically, AWS describes the ‘big idea that all schools will move to the cloud’ as ‘absolutely right and proper’, part of its attempt to secure widespread institutional migration to cloud infrastructure.

But this marketing discourse glosses over major issues concerning the penetration of public education by proprietary clouds and the platforms with which they are vertically integrated. Commenting on Google cloud-based platforms for schools, Jose van Dijck notes that ‘the dependence of schools on proprietary information systems effectively funnels pupils’ data, generated in a public context, into a proprietary data flow controlled by one corporation’s platforms’.

In these ways, cloud operators have taken up powerful infrastructural positions in the sector: they host and enable digital education companies, services and platforms, as well as underlying and empowering institutions’ digital services, all while inspiring imaginaries of a cloud-like reorganization of education itself and enclosing or ‘locking-in’ public education in privatized infrastructure systems. This reflects how cloud operators have expanded across industries and sectors, and the transformations cloud computing has wrought on the global digital economy.

Cloud economics

The cloud is not simply a technical accomplishment. Understanding the cloud in public service areas such as education also means situating it in the shifting dynamics of the digital economy. Education platforms themselves could not operate as they do without the cloud and its distinctive business models.

Many contemporary tech business models are ‘based on turning all social interactions and economic transactions into “services” that are mediated by corporate platforms’, Jathan Sadowski has argued, thereby ‘concentrating control over infrastructure and economic value in a small number of large hands’. He adds,

… tech companies increasingly describe themselves as providing “X as a service.” But what this business model really means is that they enjoy all the rights of owning an asset while you pay for the limited privilege of access. In other words, we are now forced to deal with an explosion of landlords in our daily life — constantly paying rent, both in terms of money and data, for all of the different tools and services we use.

In the platform economy, value is generated from organizations paying on-demand fees and subscriptions for services, which Sadowski defines as ‘economic rent’, and is also amassed from the extraction of ‘data rent’ in the shape of valuable user information that can be monetized, for example by creating derivative products or upgrades.

The cloud is the techno-economic infrastructure underpinning this colossal concentration of tech power. In a recent article on cloud infrastructure, Devika Narayan argues that transformations in underlying computing arrangements are shaping the growth of platform-based companies—specifically that cloud computing arrangements are ‘setting the foundational sociotechnical infrastructure’ for ‘platform capitalism’. Noting the intense competition for cloud market share between big tech companies including Amazon, Microsoft and Google, as well as those companies building on top of their cloud services, she calls the cloud an ‘infrastructural force of 21st-century corporate expansion’.

So, the second priority is to address how the dynamics of corporate cloud expansion apply to education. There are clear issues with the commercialization and privatization of education to consider here. Public education systems are increasingly underpinned by private infrastructures, bringing about concerns over the erosion of public values by private interests and profit-seeking motives. As cloud computing arrangements have become constitutive of the platform economy, education is now thoroughly enmeshed in those new economic shifts, including the vast concentration of power by big tech firms and their new modes of value acquisition. Institutional users of cloud services are paying both economic and data rents for cloud-based services. So too are other third-party edtech platforms.

This is changing the economic landscape of education, with major multinational computing companies now acting as infrastructural forces in education systems in multiple different ways. The ‘network effect’ of increasing educational subscriptions to clouds, by institutions and platforms alike, means education is entangled in what Devika Narayan calls ‘new modes and strategies of accumulation and corporate expansion’, as well as being a key target of ‘the aggressive market-making activities of the cloud computing heavyweights—Amazon, Google, and Microsoft’.

Connective ecosystems

A third area of research on the cloud in education might focus on its connective architectures—the way specific programs and protocols connect various platforms and services together into networked ecosystems, facilitated by cloud infrastructures. Devika Narayan, for example, talks of ‘modularized architectures’ of different software applications that are connected together by application programming interfaces (APIs). She refers to APIs as ‘boundary resources’ that integrate different software applications in cloud systems, ‘resulting in a do-it-yourself approach to computing infrastructure’.

Technically, APIs are a mundane instance of software code allowing applications to connect, interoperate, and exchange data and functionalities. But in a new article, Fernando van der Vlist, Anne Helmond, Marcus Burkhardt and Tatjana Seitz argue APIs have a much more significant role. ‘APIs have become the core elements of digital infrastructure, underpinning today’s platform economy and society’, they argue, suggesting research and regulation should ‘not only focus on the market dominance of platform companies but also on their “data dominance”—specifically, how platform companies use APIs to share data or integrate their services with third parties’. API specifications ‘govern and control the possibilities for the exchange of data and services between software systems and organizations’, they add, and are a ‘major source of infrastructural power’.

Cloud operators therefore rely on API specifications to enable platform developers to build on and interoperate with the cloud infrastructure. In that sense the design of APIs sets the rules by which other third party platforms can integrate into, communicate, and exchange data with cloud operators, ‘in exchange for infrastructural control’ as van der Vlist and colleagues put it.

Moreover, as Kean Birch and Kelly Bronson argue, APIs may not only function as boundary resources in modularized ‘digital ecosystems’ of interconnected platforms and infrastructures. They also function as ‘boundary assets’ with calculable techno-economic value for their owners and controllers. These boundary assets, they argue, ‘not only enable integration across boundaries’, but ‘the modular relations they constitute end up having significant value for Big Tech firms whose valuation is based on the capitalization of future earnings derived from the users in and of their ecosystems’. Understood as assets that generate value from constructing modular relations across platform ecosystems, cloud APIs thus reinforce the dominant techno-economic business model and cloud economics of big tech firms.

APIs play a significant function in the contemporary educational ecosystem of cloud-based digital education platforms. Cloud operators such as Google and AWS, which are highly active in education, are mobilizing APIs to integrate an array of education platforms into their cloud infrastructures. The cloud-based Google Classroom, for example, integrates with a huge variety of third party educational platforms through a specific Classroom API and single-sign-on access, facilitating significant cross-platform interoperability across the edtech landscape and making Classroom itself a central gatekeeper to other education platforms.

Similarly, AWS has deployed APIs to enable other third party developers to integrate its voice and face recognition technologies. These include an API to permit other developers to integrate Alexa-based voice interfaces into educational applications, such as learning management systems, student information systems, classroom management tools and online course platforms. AWS also provides an Amazon Rekognition API enabling developers to add image and video analysis to their applications, including facial recognition and facial analysis, which has been integrated into exam proctoring software.

As van der Vlist and coauthors argue, APIs govern the development of third party services and platforms. Beyond being technical objects, API design sets kinds of policies that ‘directly influence, often in subtle ways, what can and cannot be built, sustained, or thrive in the ecosystem’, and ‘provide centralized and unidirectional hierarchical control over large numbers of apps and services—and the developers who build and maintain them’. And as Birch and Bronson put it, ‘big tech’ proprietors ‘can set the terms of engagement through contractual arrangements…, representing privately-made rules and standards’ that function as ‘gatekeeping’ devices for modular integrations and relations.

Yet very little is known about the ways APIs govern the development, functionalities and relations of third party education platforms with cloud arrangements, or about the exchanges of data that occur once a platform integrates into and interoperates with a cloud infrastructure via an API.

AIOps

The fourth area of research on the cloud in education could focus on the specific new capacities that the cloud affords for other education platforms and services. In the book Cloud Ethics, Louise Amoore approaches cloud computing in terms of its algorithmic powers of perception. The cloud, she argues, ‘is a bundle of experimental algorithmic techniques acting in and through data’, and ‘contemporary cloud computing is about rendering perceptible and actionable (almost seeing) that which would be beyond the threshold of human vision’.

The promise of the cloud is that masses of heterogeneous data processed using machine learning algorithms can generate unprecedented insights and predictions to be acted upon by reducing vast complexity to single outputs, often by automated means. The cloud, Amoore argues, generates objects of ‘attention’ from the combination and analysis of massive datasets; objects of attention that might then become targets of intervention.

In this sense, the cloud needs to be considered as the active infrastructure that makes possible applications of machine learning, predictive analytics, and the bundle of technologies known as artificial intelligence. The cloud enables new kinds of ‘AIOps’ (Artificial Intelligence for IT Operations, as tech consultancy Gartner describes it) to be built into industries and sectors such as education.

Indicatively, AWS is actively seeking to embed its ‘out-of-the-box AI services‘ and machine learning facilities in other edtech platforms, inciting customers to ‘add Amazon API-driven ML services to your education software’, such as image and video analysis, text-to-speech, speech-to-text, translation, and natural language processing. Google has recently begun adding AI functionality to Google Classroom, supported by its cloud facilities, and proposed applying its large language model LaMDA as a conversational agent within the Workspace suite of platforms widely used in education too.

Educational institutions subscribing to the cloud may also be transformed by these new AIOps capacities. Cloud prioprietors propose that institutions using their cloud facilities can access enhanced data functionality, machine learning and AI for ‘digital transformation’. As part of its ‘Student Success Services’, Google has launched a ‘new Google Cloud artificial intelligence powered learning platform’. The platform is described as ‘a suite of applications and APIs that can be integrated into an institution’s existing infrastructure’, including ‘auto-generated and personalized recommendations for courses based on their prior learning’, and an ‘interactive tutor’ that ‘uses APIs to present a chat-based experience for students, incorporating AI-generated learning activities’.

AWS similarly claims its institutional cloud users can ‘Share data seamlessly across platforms to get a comprehensive view of student performance and uncover insights’, and ‘enhance learning with powerful tools including, artificial intelligence (AI), machine learning (ML), and voice-recognition’. It has published an entire guide on becoming ‘a data-informed institution’ focused on ‘how institutions are using data and cloud technology to inform digital transformation’. Oracle, too, has launched Oracle Student Cloud, which it claims is ‘powered by artificial intelligence (AI)’ and ‘consumer-market design elements’ to ‘dramatically change the student experience at every stage of the lifecycle’.

As these examples indicate, the cloud can power education platforms with new capacities for datafication, prediction, automation, and other AIOps. It is a sociotechnical foundation for what Kalervo Gulson and Kevin Witzenberger have termed ‘automated anticipatory governance’, whereby AI enacts forms of prediction and pre-emption, particularly through integrating AI in edtech platforms via APIs. The cloud and platform APIs may therefore function to govern a widening array of processes and practices in education by introducing new capacities of corporate AIOps into edtech platforms and institutional information systems.

Cloud power

The cloud is a largely invisible, background presence in education, despite playing an increasingly significant role in many technical and institutional processes and practices. As recent relevant scholarship on the cloud has indicated, cloud computing arrangements are significantly affecting and reshaping a range of industries and sectors. The cloud represents an expansion of corporate big tech power into sectors like education, introducing new economic models, platform ecosystem arrangements, and AIOps capacities of automated governance.

At its most extreme the cloud will even underpin plans to build out new virtual reality educational experiences and institutions in the ‘metaverse’, which (if realized) will both be a huge economic bonus for cloud operators and for cloud-based metaverse platform providers.  

Much more significant are the more mundane instances of cloud computing in education, which promise many benefits, but also pose important questions for research. Areas of investigation include institutional autonomy and control over data and infrastructure, challenges to student consent and ‘opt-out’ of cloud storage, potential for breaches and leaks, the possibility of long-term lock-ins, automation of educational processes, and the redirection of educational funds from in-house systems to outsourced cloud provision and maintenance.

It may be too soon to claim the cloudification of education, but it’s clear many aspects of education are now being enclouded to a significant degree, with longer-term effects which remain to be seen. While remaining sceptical of cloud-inspired imaginaries of Education-as-a-Service or cloud-based ‘metaversities’, education research should nonetheless consider what powers the cloud and its operators could exert on the sector, and how to respond.

How Amazon operates in education

The global technology business Amazon is increasing its involvement in education at international scope and scale through complex technical and business practices. Photo by Noel Broda on Unsplash

Ben Williamson, Kalervo N. Gulson, Carlo Perrotta and Kevin Witzenberger

The global ‘big tech’ company Amazon is increasing its reach and power across a range of industries and sectors, including education. In a new paper for the special symposium ‘Platform Studies in Education’ in Harvard Educational Review, we conceptualize Amazon as a ‘state-like corporation’ influencing education through a ‘connective architecture’ of cloud computing, infrastructure and platform technologies. Like its retail and delivery logistics business it is operating at international scope and scale, and, congruent with Amazon’s growing influence across industries and sectors, possesses the power to reshape a wide range of educational practices and processes.

Our starting point is that education increasingly involves major technology companies, such as Google, Microsoft, and Amazon playing active roles as new kinds of networked governance actors. Infrastructures of test-based accountability and governance in education have long involved technical and statistical organizations. However, contemporary education governance is increasingly ‘data-driven’, using advanced technologies to collect and process huge quantities of digital information about student achievement and school and system performance.

In this context, new digitalized and datafied processes of education governance now involve multinational technology businesses offering infrastructure, platforms and data interoperability services. These connective architectures can affect the ways information is generated and used for institutional decision making, and also introduce new technical affordances into school practices, such as new platform-based learning, API-enabled integrations for increased interoperability, and advanced computing and data processing functionality from cloud infrastructures.

Our analysis focuses on Amazon, specifically its cloud computing subsidiary Amazon Web Services (AWS). Despite significant public, media, and regulatory attention to many of Amazon’s other activities and business practices, its activities in education remain only hazily documented or understood. AWS, we argue, enacts five distinctive operations in education.

Inscribing

The first part of our examination of AWS identifies how its corporate strategy underpins and infuses its objectives for education—a process we call inscribing to refer to the ways technology companies impress their business models on to the education sector. AWS is Amazon’s main profit engine, generating more than 60% of the corporation’s operating profits. Typifying the technoeconomic business model of big tech, it functions as a ‘landlord’ hosting industry, government, state and public sector operations on the cloud, while generating value from the ‘rent’ paid for on-demand access to cutting-edge cloud services, data processing, machine learning and artificial intelligence functionalities.

The ways this process of inscribing the business model on education takes place is evident in commercial marketing and discourse. AWS seeks to establish itself as an essential technical substrate of teaching, learning and administration, promoting its capacity to improve ‘virtual education’, ‘on-demand learning’ and ‘personalized learning’, and to support ‘digital transformation’ through ‘cloud-powered’ services like ‘campus automation’, ‘data analytics platforms’ and ‘artificial intelligence’. These promotional inscriptions paint a seductive picture of ‘pay-as-you-go’ educational improvement and seamless ‘plug-and-play’ transformation.

Beyond being discursive, these transformations require very specific kinds of contractual relations for cloud access, pay-as-you-go plans, and data agreements as per the AWS business model. AWS thus discursively inscribes and materially enacts its business model within education, impressing the techno-economic model of cloud tenancy, pay-as-you-go subscription rents, and computational outsourcing on to the education sector—potentially affecting some of the core functions of education in its pursuit of valuable rent and data extraction. Through this strategy, AWS is fast becoming a key cloud landlord for the education sector, governing the ways schools, colleges and edtech companies can access and use cloud services and digital data, while promoting a transformational vision of education in which its business interests might thrive.

Habituating

The second architectural operation of AWS is its techniques for accustoming users to the functionality of the cloud. We term this habituating users to AWS, or synchronizing human skills to the cloud. It does so through AWS Educate, an educational skills program designed to develop teachers and students’ competencies in cloud computing and readiness for ‘cloud careers’. AWS Educate seeks to establish a positive educational discourse of ‘the cloud’, whereby educators and students are encouraged to develop their skills with AWS services and tools for future personal success, thereby connecting hundreds of thousands of students, educators and institutions and accustoming current and future users to the AWS architecture.

With stated aims to reach 29 millions learners worldwide by 2025, key features of AWS Educate include Cloud Career Pathways and Badges, with dedicated technical courses and credentials aligned to industry job roles like cloud computing engineer and data scientist.  These credentials are underpinned by the Cloud Competency Framework, a global standard used to create, assess, and measure AWS Educate cloud programs informed by the latest labour market data on in-demand jobs. This strategy also serves the goal of increasing user conversions and further AWS adoption and expansion, advancing the business aim of converting user engagement into habitual long-term users as a route to future revenue streams.

In short, through its habituating operations, AWS promotes a normative vision of education as electronic micro-bundles of competency training and credentials, twinned with the habituation of users to its infrastructure. While serving its own revenue maximization prospects, AWS Educate challenges public education values of cultivating informed citizenship with values prioritizing a privatized and platformized education dedicated to the instrumentalist development of a future digital workforce.

Interfacing

The third operation enacted by AWS in education is interfacing. AWS provides new kinds of technical interfaces between educational institutions, intermediary partners, and the AWS infrastructure. This is exemplified by Amazon’s Alexa, a conversational interface, or voice assistant, that sits between users and AWS, and which AWS has begun promoting for integration into other educational applications. Its interfacing operations are achieved by the Alexa Education Skills Kit, a set of standards allowing Alexa to be embedded in third party products and services. We argue it illustrates how application programming interfaces (APIs) act as a connective tissue between powerful global data infrastructures, the digital education platform industry, and educational institutions.

For example, universities can develop their own Alexa Skills in the shape of institutionally branded voice interfaces for students to access coursework, grades and performance data; educators can embed Alexa in classes as voice-enabled quizzes and automated ‘study partners’; and institutions are encouraged to include Alexa Skills in ‘smart campus’ plans.  In these ways, the Alexa Skills Kit provides a set of new AWS-enabled, automated interfaces between institutions, staff and students, mediating an increasing array of institutional relations via the AWS cloud and the automated capacities of Alexa.

The Alexa Education Skills Kit is one of many APIs AWS provides for the educational sector to access fast, scalable, reliable, and inexpensive data storage infrastructures and cloud computing capacities. The integration of automated voice assistants through the Education Skills Kit provides educational institutions a gateway into the core functionality of AWS. These interfaces depend upon the automated collection and analysis of voice data on campuses, its automated analysis in the AWS cloud, and the production of automated feedback, so generating a cascade of automation within institutions that have synchronized their operations with AWS. It normalizes ideals of automation in education, including the extensive data collection and student monitoring that such automation entails. Through its interfacing operations, we therefore argue, AWS and Alexa are advancing cascading logics of automation further into everyday educational routines.

Platforming

Cloud computing establishes the social and technical arrangements that enable other technology companies to build and scale platforms. Amazon has developed an explicit market strategy in education by hosting—or platforming—the wider global industry of education technology on the AWS Cloud, specifically by providing the server hosting, data storage and analytics applications necessary for third parties to build and operate education platforms. Its AWS Imagine conference highlights its aspirations to host a huge range of edtech products and other services, and to guide how the industry imagines the future of education.

The role of AWS in platforming the edtech industry includes back-end server hosting and data storage as well as active involvement in startup development. Many of the globe’s largest and most highly capitalized edtech companies and education businesses are integrated into AWS. AWS support for the edtech industry encompasses data centre and network architecture to ensure that clients can scale their platform, along with data security and other AWS services including content delivery, database, AI, machine learning, and digital end user engagement services. This complete package enables edtech companies to deliver efficient computing, storage, scale, and reliability, and advanced features like data analytics and other AI services. It promotes the ways its ‘out-of-the-box AI services‘ can be embedded in edtech products for purposes of prediction and personalization.

As such, through its platforming operations, AWS acts as an integral albeit largely invisible cloud presence in the back-end of a growing array of edtech companies. The business model of AWS, and the detailed contractual agreements that startups must sign to access AWS services, construct new kinds of dependencies and technical lock-ins, whereby the functionalities offered by third-party education platform companies can only exist according to the contractual rules and the cloud capacities and constraints of AWS. This puts AWS into a powerful position as a catalyst and accelerator of ‘digital transformation’ in education, ultimately responsible for re-tooling the industry for expanded scale, computational power, and data analytics functionality.

Re-infrastructuring

The final operation we detail is re-infrastructuring, referring to the migration of an educational institution’s digital infrastructure to AWS. It does so through AWS Migration services, and by providing institutions with a suite of data analytics, AI and machine learning functionalities. AWS promises that by ‘using the AWS Cloud, schools and districts can get a comprehensive picture of student performance by connecting products and services so they seamlessly share data across platforms’. AWS also promotes Machine Learning for Education to ‘identify at-risk students and target interventions’ and to ‘improve teacher efficiency and impact with personalized content and AI-enabled teaching assistants and tutors’. 

This seamless introduction of AI and automation is enabled by the formation of ‘data lakes’—a repository that hosts multiple types of data for machine learning analysis and visualization in the cloud. The process of ‘architecting a data lake‘ involves the deployment of multiple AWS products and functionalities, including those for pulling data seamlessly from student information and learning management systems, and for handling the ‘machine learning workload’ of analysis. AWS promotes full infrastructure migration to the cloud in terms of making everything from students and staff to estates and operational processes more intelligible from data, and thereby more amenable to targeted action or intervention.

Through cloud migration and data lake architecting, schools and universities are outsourcing a growing range of educational and administrative operations. This ultimately reflects a fresh hierarchical stratification of education, with AWS and its cloud firmly on top, followed by a sprawling ecology of edtech companies that mediate between AWS and the clients at the bottom: the schools and universities that form the data lakes from which AWS derives value. Yet, despite being highly consequential, these infrastructural rearrangements remain opaque, hidden in proprietorial ‘black boxes’, potentially resistant to autonomous institutional decisions, and extremely expensive and challenging to reverse.

‘Big tech’ and ‘state-like corporations’

One key implication we detail in the paper is the growing role of multinational ‘big tech’ companies in education, and the complex ways they are advancing longstanding reform efforts to privatize and commercialize public education, albeit through new techno-economic business models and practices. Social scientific and legal scholarship on private platforms and infrastructures has begun to contend with their growing social, technical and economic power, particularly their implications for key functions and processes traditionally considered the responsibility of state agencies or public sector organizations. As a corporate cloud company, Amazon is attempting to create market dominance and even monopoly power across a multitude of sectors and industries, raising sharp political and legal questions over the appropriate regulatory or antitrust measures to be taken.

Part of this competition is also for infrastructural dominance in education. The expansion of AWS signifies how the governance of the public sector and its institutions is becoming increasingly dependent on the standards and conditions set by multinational big tech corporations like Amazon and Google. Amazon is gathering significant power as what Marion Fourcade and Jeff Gordon term a ‘state-like corporation’. As a corporation with state-like powers, AWS can use its technical and economic capacity to influence diverse education systems and contexts, at international scale, and potentially to fulfil governance roles conventionally reserved for state departments and ministries of education.

As such, the continuing expansion of AWS into education, through the connective architecture we outline in the paper, might substitute existing models of governance and policy implementation with programmable rules and computer scripts for action that are enacted by software directly within schools and colleges rather than mandated from afar by policy prescriptions and proscriptions. As a state-like corporation with international reach and market ambitions, AWS is exceeding the jurisdictional authority of policy centres to potentially become the default digital architecture for governing education globally.

The full paper is available (paywalled) at Harvard Educational Review, or freely available in manuscript form.

New biological realities in the genetics of education

New research with a sample of more than 3 million individuals claims to have identified thousands of genetic differences linked with education. Photo by Rob Curran on Unsplash

The recent publication of a study of the genetics of educational attainment is once again raising questions and controversies about the potential use of biological information to inform education policy and practice. In the BioEduDataSci project, funded by the Leverhulme Trust, we have been collecting and examining a very large sample of texts and commentary to understand the development of the new genetics of education and its potential consequences. Such studies are characterized by being highly data-scientific, focused on identifying minute molecular differences, and highly controversial both ethically and scientifically. This research has a high profile in the media, and draws criticism from a variety of scientific and political perspectives.

The new study, conducted and published by the Social Science Genetic Association Consortium (SSGAC), is no different. It is the fourth in a series of studies linking DNA differences to educational outcomes. The first educational attainment study (EA1) examined a sample of around 100,000 people, the second (EA2) a sample of 300,000, and the third (EA3) featured a sample in excess of 1 million. EA4, however, features a sample of more than 3 million genotyped individuals, and has identified more than 3,000 tiny genetic variants—known as single nucleotide polymorphisms (SNPs)—that are said to be associated with years spent in education. The paper, published in Nature Genetics, is extremely technical, but its key findings have been usefully summarized by one of its main investigators, and a follow-up çommentary paper by one of the authors stating simply that ‘genes matter when it comes to educational performance and social outcomes’.

Like previous projects and publications linking DNA to educational outcomes, the EA4 study certainly raises serious ethical challenges. These include the lack of representation in the data analysed, the potential to reinforce existing racial categories and discriminatory outcomes, produce negative self-fulfilling prophecies, and possible appropriation by right-wing conservatives and ideologically-motivated scientists to make racialized, hard-hereditarian arguments about IQ and social stratification. The horrific history of eugenics in education, particularly through the intelligence testing movement, has left a dark legacy that means current studies of the genetics of education come under particularly intense scrutiny, not least because of the persistence of discriminatory practices and policies grounded in genetic theories of difference. The term ‘genetics of education’, widely used in popular accounts of recent research, may itself problematically harden assumptions about the association of genes with educational achievement or even IQ. 

Rather than rehearse those issues here, however, I want to focus on one specific scientific dispute in relation to genetic educational attainment studies. The post draws from our wider research detailing the organizational, conceptual, methodological and technical systems and structures underpinning the new genetics of education, and exploring the knowledge claims and proposals for intervention emerging from such studies. What we are interested in this post is not so much developing an external critique of educational genetics, but tracking some of its internal conflicts and their implications.

New genetics of education

Over the last 15 years, new studies of the genetics of education have appeared from two overlapping fields of scientific inquiry. Behaviour genetics is a branch of psychology examining the genetic bases of human behaviours and traits, or how genotypes influence phenotypes (like cognitive ability). It has recently adopted high-tech and data-driven genomics methods to study the ‘molecular genetic architecture’ of ‘educationally-relevant traits’ and phenotypes, including cognitive ability and intelligence.

Meanwhile, social science genomics, or sociogenomics, represents the interdisciplinary combination of genomics with certain branches of sociology, economics, and political science. It is interested in the biological structures and mechanisms that interact with environmental factors to produce socioeconomic outcomes. Educational attainment is taken by sociogenomics as a key socioeconomic outcome that is related to other outcomes such as occupation, social status, wealth, and health.

Where behaviour genetics and sociogenomics overlap is in their methodologies and techniques of analysis. They both use highly data-intensive research infrastructure, such as biodata repositories and data mining software, and methods that can identify vastly complex associations between thousands of minute SNP variations and their correlation with educational outcomes or relevant traits.

They use methods such as genome-wide association studies (GWAS), processing enormous quantities of SNP data, and calculating ‘polygenic scores’—a quantitative sum of all SNP variant data—and apply them to education. Thus, educational attainment studies such as those undertaken by the SSGAC use GWAS methods and SNP analysis to produce polygenic scores for education, which can then be used to predict the attainment of independent samples. Similar approaches have been taken to a range of education-relevant phenomena, such as intelligence and learning. 

The SSGAC’s huge-sample educational attainment studies are taken as gold standard models for investigating the genetics of education by both behaviour genetics and sociogenomics researchers. Their findings underpin the arguments in Robert Plomin’s controversial book Blueprint, where he proposes that DNA data could used as the basis for personalized ‘precision education’, are explored in detail by Katherine Paige Harden in The Genetic Lottery to underpin her argument for educational reforms, and animate a wide range of other sociogenomics studies (although the SSGAC reports no practice or policy implications from the EA4 study due to being only ‘weakly predictive’).

Not all scientists agree, however, that multimillion-sample genomics studies offer any useful insights into the genetics underpinning education, even less that they might inform how education itself is organized.  

Heritability problems

Perhaps surprisingly, one of the most consistent and vocal critics of data-driven genomics studies of education is a leading behaviour geneticist, Eric Turkheimer. Turkheimer certainly believes in the heritability of human behaviours—that is, that a certain portion of behaviour is influenced by genetics rather than being entirely environmentally shaped. He wrote the ‘three laws’ of behaviour genetics affirming as much. A fourth law was added in 2015 (by authors from the SSGAC) stating that ‘a typical human behavioural trait is associated with very many genetic variants, each of which accounts for a very small percentage of the behavioural variability’.

Polygenicity among hundreds of thousands or even millions of SNPs has become a defining law of social and behavioural genomics studies, including those focused on the genetic heritability of educational behaviours and outcomes. Leading scientists claim these methods promise to ‘open the black box of heritability’ and finally reveal the pathways from DNA to social outcomes such as educational achievement.

Turkheimer, however, has become highly critical of the methodological turn to genomics when studying complex behaviours and outcomes, including in the SSGAC EA studies, as his useful recent tweet-response to EA4 indicates. The basis of the critique is the so-called ‘missing heritability problem’. Turkheimer argues in a new paper with Lucas Matthews that the concept of ‘heritability’—an ‘estimate of the proportion of phenotypic [behavioural] variance that is statistically associated with genetic differences’—has changed with shifting methodologies for its measurement.

Of course, heritability estimates are historically constructed and contested. Basically, earlier forms of behaviour genetics utilizing quantitative genetics methods, such as twin studies, identified DNA to play a large influence on any behavioural trait, often in the region of 50-70%. For example, in relation to education, quantitative behaviour genetics found around 50% of variation in intelligence, as measured by IQ tests, was heritable.

In contrast, Matthews and Turkheimer argue that recent high-tech, data-intensive genomics methods, such as genome-wide association studies (GWAS), have hugely increased computational power but reduced explanatory power. For example, ‘cutting-edge GWAS have recently estimated that only 10% of variance in IQ is statistically associated with differences in DNA’. They point out that studies of educational attainment from the SSGAC are therefore scientifically underwhelming, as they account for somewhere in the region of 12-15% of variance, despite the huge samples and computational power put to the analysis. This gap is what’s known as the missing heritability problem.

The issue for Matthews and Turkheimer is that most solutions to missing heritability appear to be focused on throwing even bigger samples and more processing power at the problem. They describe this as ‘dissolving the numerical gap’ between traditional quantitative and molecular computational kinds of heritability estimates, but argue ‘resolving the numerical discrepancies between alternative kinds of heritability will do little to advance scientific explanation and understanding of behavior genetics’. They note that ‘most writing on the topic expresses optimism that this day will soon come as researchers collect larger datasets and develop more sophisticated statistical genetic models of heritability’.

By contrast, Matthews and Turkheimer ‘argue that framing the missing heritability problem in this way—as a relatively straightforward quantitative challenge of reconciling conflicting kinds of heritability—underappreciates the severe explanatory and methodological problems impeding scientific examination and understanding of heritability’.

More urgent than closing the numerical gap, for them, is the persistent ‘prediction gap’, or the challenge of making accurate and reliable prediction from DNA to behaviour, and, even more so, the ‘mechanism gap’, which refers to the gap in explaining the specific mechanisms linking molecular genotypes to behavioral phenotypes. There remains, they suggest, a ‘black box’ of hidden mechanisms that simply solving the numerical gap will never discover.

Graphic detailing the numerical, prediction, and mechanism gaps in behaviour genetics, from Matthews and Turkheimer (2022)

These gaps in prediction and explanation of mechanisms are especially acute for studies of the genetics of education. They pose a challenge to claims that genetic data could be used—in the not so distant future—to open the ‘black box of heritability’, or even inform educational policy or practice in schools:

the putatively causal relationship between … SNPs and differences in educational outcomes is entirely opaque, other than the very general assertion that many of the SNPs are close to genes that are expressed in neural tissue. Until scientists have identified, described, and substantiated causal-mechanical etiologies that would explain why countless SNPs are correlated with behavioral outcomes like IQ and educational attainment, then what we call the mechanism gap of the missing heritability problem remains a daunting and persistent scientific challenge. … Highly complex human behavioral traits and outcomes such as intelligence and educational attainment are farthest from dissolution: the numerical gaps, predictions gaps, and mechanism gaps for these cases may never be resolved.

The paper highlights two important issues confronting the new genetics of education. First, despite significant investment in scientific infrastructure, data analytics technologies, and high-profile publications, educational genetics studies remain a long way from opening the ‘black box’ of the specific genetic mechanisms that underpin educational outcomes. And second, it highlights how educational genetics studies are not just ethically controversial and subject to external critique, but scientifically controversial and internally contested too.

Making biodatafied realities

Another key part of Matthews and Turkheimer’s argument is that the heritability estimates produced by quantitative genetics are of a very different kind than the heritability models produced by molecular genomics methods such as GWAS and SNP analysis. This is not just a matter of quantitative innovation, then, but a qualitatively different mode of investigation which produces a very different kind of knowledge.

In this sense, Matthews and Turkheimer seem close to suggesting that the computational infrastructure of genomics makes a significant difference to the knowledge that is produced. This is an argument familiar in critical data and infrastructure studies, where it is assumed data infrastructures are far from merely neutral interfaces to access factual reality, but instead represent ‘expressions of knowledge/power, shaping what questions can be asked, how they are asked, how they are answered, how the answers are deployed, and who can ask them’.

Likewise, the epidemiologist Cecile Janssens has recently argued that polygenic scores, in particular, emerged as a ‘pragmatic solution’ to the statistical problem of calculating very large SNP associations in genomics. Technically, a polygenic score is calculated using particular data-mining software applications, computing formats, algorithms and statistical standards created by specialists in high-tech genomics research laboratories. As Janssens suggests, this level of technical mediation in the construction of polygenic scores matters.

Polygenic scores, Janssens argues, ‘do not “exist” in the same way’ as other measurable biological processes such as blood pressure, but only as ‘algorithms’, ‘models’, or ‘simplifications of reality’. Her concern is that polygenic scores, as pragmatic solutions to a statistical problem, might create a new ‘biological reality’ and be used as the basis for certain forms of intervention, despite being only simplified models.

Janssens’ concern, like that of Matthews and Turkheimer, reflects important critiques of genomics in science and technology studies. The historian of bioinformatics Hallam Stevens, for example, argues that ‘these algorithms and data structures do not merely store, manage, and circulate biological data; rather, they play an active role in shaping it into knowledge’.

Polygenic scores and heritability estimates produced through informatic forms of biology, then, may generate very different conceptions of what constitutes ‘biological reality’. As Joan Fujimura and Ramya Rajagopalan have argued elsewhere, data-intensive genomics ‘is ultimately a statistical exercise that depends on the analytic software itself and the information that goes into the statistical software’. As such, they emphasize, when a GWAS or SNP analysis reveals complex associations, ‘these underlying patterns are known only through the data and data producing technologies of the geneticists’.

If Matthews and Turkheimer are correct, then it seems like the kind of heritability of educational attainment that recent GWAS, SNP and polygenic score studies have uncovered is a different kind of heritability that can be known only through the data and technologies of educational genetics research.

As a pragmatic solution to a computing problem, polygenic scores are changing the ways that DNA is understood to affect social and behavioural outcomes. They are based on a kind of heritability that exists primarily as a computational artefact–one which, despite its weak predictivity and absence of explanatory power, is nonetheless attracting significant media and public attention as a way to understand the molecular bases of educational outcomes.

Even while overall effect sizes of such studies may remain modest, they may already be shifting public and professional discourse towards a more biological perspective, infusing educational debates with the vocabulary of genotypes, heritability, phenotypes, and genetic architectures, despite the persistent explanatory gap in the underlying science.

The social life of bio-edu-data science

What we can suggest here, then, is that new biological knowledge and even biological realities are being created through the use of data-intensive genomics technologies and methods in educational genetics research. As Matthews and Turkheimer argue, a different kind of heritability emerges from the complex scientific and data infrastructures of GWAS and polygenic scoring. For Janssens, polygenic scores only exist as algorithms, not as embodied substance. This challenges assertions that once the polygenic ‘genetic architecture’ underpinning educational outcomes is objectively known through big biodata analysis, it may be possible to design educational interventions on the basis of such knowledge.

Rather, new knowledge about the genetics of education is generated through distinctive computational systems, software, and methods that all leave their mark on understandings of the genetic substrates of educational outcomes. What emerges from such studies, as Matthews, Turkheimer and Janssens suggest, are new bio-edu realities that have been produced through computers, data processing, and particular statistical applications, rather than simply unmediated insights into the objective molecular substrates that underpin students’ educational achievements.

But these new realities can be consequential despite being scientifically contested or only weakly explanatory. They may animate enthusiasm for ideas about genetically-informed schooling, and could potentially lead to a hardening of biological explanations, sometimes dangerously motivated by racism, for the complex social, political, cultural and economic factors that shape students’ achievement in schools.

For these reasons, in the BioEduDataSci project, we’re tracking the ‘social life’ of the new genetics of education. This means trying to understand the social contexts and conditions of new knowledge production, the reception of such new knowledge, and its social, political and ethical implications as such knowledge circulates in the media and in public, often by being translated and interpreted and sometimes turned to harmful results.

The new genetics of education is not yet a settled science, and if critical voices–even within fields like behaviour genetics, as well as its critics–are correct, may never be. Nonetheless, the new genetics of education remains a fast-moving science in the making, surfacing complex issues and problems that need addressing and debating among many stakeholders across the biological and social sciences, policy and practitioner sites, before any emerging findings are considered as insights for implementation.

UPDATE: A few days after posting this, the EA3 educational attainment study was cited to justify an act of horrific racist violence resulting in many deaths in Buffalo, USA. This has animated urgent calls among the genomics community for scientists of such research to take far more responsibility for their study findings – or perhaps not even publish it at all given its dangerous consequences – generating counterarguments about scientific freedom.

Google magic

Google has announced a new ‘magical’ upgrade to its Classroom platform. Photo by Ixxlhey 🇲🇻 on Unsplash

Google has announced the latest in a sequence of upgrades to its popular online learning platform Google Classroom, and it’s all about injecting more artificial intelligence into schools. Classroom was popular with teachers all around the world before Covid-19, but has experienced enormous growth since 2020. Google’s announcement of new AI functionalities reveals its continued plans for expansion across the education sector, and beyond, and its technical and economic power to reshape learning through automation.

Autopedagogy

The company publicized its new AI feature for Classroom, called Practice Sets, in a marketing blog with an explanatory video on the Google for Education site. The basic functionality of Practice Sets is powered by ‘adaptive learning technology’, which it said ‘will give teachers the time and tools to better support their students — from more interactive lessons to faster and more personal feedback.’

The marketing video is instructive of the particular imaginary of schooling that Google is seeking to operationalize with Practice Sets. The adaptive learning technology, it claims, will provide ‘one-to-one feedback,’ providing a ‘round the clock tutor for each student.’ It ‘identifies relevant learning skills’ while students are using the feature within Classroom assignments, then ‘finds relevant hints and resources’ which appear as ‘recommended video and content cards for each learning skill.’ Practice Sets also features ‘autograding’ functionality, giving teachers ‘visibility into their [students’] thinking,’ and ‘automated insights’ into ‘class-wide trends’ to provide a ‘quick view of class performance,’ as well as ‘actionable insights’ they can use to improve their teaching.     

In an accompanying blogpost outlining its approach to adaptive learning technology, the head of Google for Education said ‘applying recent AI advances’ to adaptive learning ‘opens up a whole new set of possibilities to transform the future of school into a personal learning experience.’ He added, ‘Adaptive learning technology saves teachers time and provides data to help them understand students’ learning processes and patterns.’

These marketing materials are presented in a highly persuasive way. They tap into contemporary problems of schooling such as providing adequate support and feedback to students. They also promote the idea that education is about ‘learning skills,’ resonant with contemporary education policy preoccupations with skills and their value.

But the Google marketing is also highly technosolutionist. It proposes that datafied forms of surveillance and automation are ideally suited to solving the problems of schooling. Google positions Practice Sets as a kind of always-on, on-demand classroom assistant, able to execute auto-pedagogical interventions depending on a constant trace of a student’s data signals. It’s likely just one step in Google’s roll-out of AI in its education platforms. As one favourable industry review suggested, ‘Now that Google is adding AI capabilities to Google Classroom, expect the search giant to add even more automation to its online learning platform going forward.’

The appearance of Practice Sets exemplifies the ways automation is becoming an increasing presence in schools. Neil Selwyn, Thomas Hillman, Annika Bergviken Rensfeldt and Carlo Perrotta argue that automation is not materializing in the spectacular guise of robot educators, but as much more mundane services and feature upgrades.

Education — as with most areas of contemporary life — is becoming steadily infused with small acts of technology-based automation. These automations are intrinsic to the software, apps, systems, platforms and digital devices that pervade contemporary education. … [W]e are now teaching, studying and working in highly-automated and digitally directed educational environments.

While Practice Sets certainly fits the mould of a mundane instance of micro-automation, its significance is its potential scale. Indeed, Google claimed it could achieve the promise of adaptive learning ‘at unprecedented scale.’ This is because Google Classroom has penetrated into hundreds of thousands of classrooms worldwide, reaching more than 150 million students.

Practice Sets is currently available only in beta, but will in months to come be rolled out as a feature upgrade to premium Workspace customers. A year ago, Google said it was reframing Classroom as a learning management system, not just a platform for learning from home during Covid disruptions, and set out a ‘roadmap’ for its future development. Practice Sets represents the latest step on that roadmap, one that involves becoming the global learning management infrastructure for schools and increasing the presence of autopedagogical assistants in the school classroom, with potential for intervention with millions of students.

Its ambitions for scalability exceed the Classroom alone, however. It is also positioning itself as a ‘learning company,’ dedicated to ‘Helping everyone in the world learn anything in the world’, whether at school, at work, or in everyday life itself. It’s not a huge jump to assume that Google’s expansive ambitions as a learning company will gradually see it extend automation well beyond the school.

Technomagic

What is striking about the Practice Sets announcement is the way it presents adaptive learning technology. For teachers, it will ‘supercharge teaching content,’ and ‘when students receive individualized, in-the-moment support, the results can be magical.’ Early users of the beta service are reported to have described Practice Sets as ‘Google magic.’ This emphasis on magical results and Google magic is typical of technology marketing discourse. (As an aside Google Magic is also the name of its Gmail spam filter.)

As MC Elish and danah boyd have previously argued, ‘magic’ is frequently invoked in tech marketing materials, especially in relation to AI, while minimizing attention to the methods, human labour and resources required to produce a particular effect:

When AI proponents and businesses produce artefacts and performances to trigger cultural imaginaries in their effort to generate a market and research framework for the advancement of AI, they are justifying and enabling a cultural logic that prioritizes what is technically feasible over what is socially desirable. … [W]hat’s typically at stake in the hype surrounding Big Data and AI is not the pursuit of knowledge, but the potential to achieve a performative celebration of snake oil.

AI is of course not magic. The discourse and imaginary of magical AI obscures the complex social, economic, technical and political efforts involved in its development, hides its internal mechanisms, and disguises the wider effects of such systems.

Invoking ‘Google magic’ therefore erases the underlying business interests of Google in education and the internal dynamics of its data-driven AI developments. Google has a distinctive business model, which is imprinted on everything it does, including educational platforms like Classroom. It’s a business plan that is laser-focused on generating value from data and applying automation and AI across all its activities.

And it’s a controversial business model to apply to education. Dan Krutka, Ryan Smits and Troy Willhelm argue that ‘Google extracts personal data from students, skirts laws intended to protect them, targets them for profits, obfuscates the company’s intent in their Terms of Service, recommends harmful information, and distorts students’ knowledge.’

Google magic also obscures the underlying technical functionality of Practice Sets. One clue appears in the marketing materials: ‘With recent AI advances in language models and video understanding, we can now apply adaptive learning technology to almost any type of class assignment or lesson.’ At last year’s Google developer conference, CEO Sundar Pichai trailled the company’s plans to employ language models in education. These ambitions would appear to extend far beyond the functionality of Practice Sets, to include artificial conversational agents capable of conducting real-time dialogue with students.

The presentation was a glossy marketing event, but behind the scenes huge Google teams are working on language models with proposed applications in sectors like education. So-called Google magic is actually the complex social, technical and economic accomplishment of lavishly-funded R&D teams who are exploring how to do things such as ‘quantify factuality’ in order to automatically recommend students information from trusted sources for their studies.

The idea of Google magic also hides the internal political conflicts that have characterized Google’s recent efforts to build large language models. Language models are some of the most contentious, ethically problematic AI projects that Google has pinned its future business prospects on. Language models are trained in huge datasets compiled from the web. They can therefore contain social biases and amplify discrimination, and are also extremely energy-intensive and potentially environmnetally damaging. Google however has been mired in controversy since firing two of its prominent AI ethics researchers for highlighting these problems.

Google magic can obscure other things too. It evades discussions about whether global technology companies should possess power and authority to reshape the work of teachers and the experiences of students, at unprecedented scale. It turns political problems about schooling into seemingly simple technological solutions that tech companies are best placed to administer.

Invoking Google magic places out of sight any debate about the social or collective purposes of education, assuming that individualized tuition and efficient knowledge acquisition supported by automation is the ideal pedagogic mode. It also forecloses the possibility of other forms of information retrieval and discovery, and asserts Google as an authoritative cultural gatekeeper of knowledge.

Google magic distracts attention from the complex privacy policies and user agreements that determine how Google can extract and use the vast troves of user data generated from every click on the interface. It disguises the complex thickets of algorithms that make decisions about individual students, characterizing them simply as innocent and objective ways of delivering ‘factuality’ to students.

The idea of AI technology as magic also hides the fact that automation constitutes a profound experiment in how schools operate, with potentially serious effects on teacher labour and student learning and development that remain unknown. It may even draw a curtain across persistent questions about the legality of its commercial data mining operations in education.

Misdirection

Google has of course produced a lot of useful technical services, many of which may be welcome in schools. Whether AI adaptive learning will actually work in schools, or if teachers will want to use it, remains to be seen. A sceptical perspective is best taken in relation to technomagical marketing claims of as-yet unproven technology interventions.

But a critical perspective is necessary too, one that draws attention to the political role of Google as a significant and growing source of influence over schooling. It has extraordinary power to affect what happens in classrooms, to monitor children through the interface, to intervene in the labour of teachers, to generate class performance data, to select content and knowledge, and to introduce AI and automation into schools through mundane feature upgrades. It markets its services as ‘magical’ but they are really technical, social, economic, and political interventions into education systems at enormous scale around the globe.

As my colleague John Potter pointed out in response to the Google marketing, magic often refers to the ‘skill of misdirection,’ a certain sleight of hand that indicates to the audience to ‘Look at this magical stuff over here… (but don’t look at what’s happening over there).’ But ‘over there’ is precisely where educational attention needs to be directed, at the technical things, even the boring things like privacy policies and user agreements, that are reshaping teaching and learning in schools. Google may be opening up exciting new directions for schooling, but it may also be misdirecting education towards a future of ever-increasing automation and corporate control of the classroom.

PISA for machine learners

A new report from the OECD explores how human skills can complement artificial intelligence. Source: OECD

The common narrative of the future of education is that artificial intelligence and robotization will transform how people work, with changing labour markets requiring schools to focus on developing new skills. This version of the future is reflected in influential ideas about the ‘Fourth Industrial Revolution’, where novel forms of ‘Education 4.0’ will produce the necessary skilled human labour to meet the needs of ‘Industry 4.0.’ Statistical calculations and predictions of multitrillion dollar ‘skills gaps’ in the new AI-driven economy have helped fortify such visions of the future, appealing to government and business interests in GDP and productivity returns.  

The Organisation of Economic Cooperation and Development (OECD) has played a considerable role in advancing ideas about education in the Fourth Industrial Revolution, particularly through its long-term Future of Education and Skills 2030 program launched in 2016.  A background report on the 2030 project showed how education systems were not responding to the ‘digital revolution’ and new Industry 4.0 demands, and presented the OECD’s case for the development of new skills, competencies and knowledge through ‘transformative change’ in education.

Defining the future skills required of the digital revolution is now being undertaken by the OECD’s Artificial Intelligence and the Future of Skills work program, a 6 year project commenced in 2019 by its Centre for Educational Research and Innovation (OECD-CERI). As described on its approval:

The motivation for the Future of Skills project comes from a conviction that policymakers need to understand what AI and robotics can do—with respect to the skills people use at work and develop during education—as one key part of understanding how they are likely to affect work and how education should change in anticipation.

Its ‘goal is to provide a way of comparing AI and robotics capabilities to human capabilities,’ and therefore to provide an evidence base for defining—and assessing—the human skills that should be taught in future education systems. In this sense, the project has potential to play a significant role in establishing the role of AI in relation to education, not least by encouraging policymakers to pursue educational reforms in anticipation of technological developments. This post offers an initial summary of the project and some implications.

‘PISA for AI’

The first AI and the Future of Skills report was published in November 2021. Over more than 300 pages, it outlines the methodological challenges of assessing AI and robotics capabilities. The point of the report is to specify what AI can and cannot do, and therefore to more precisely identify its impact on work, as a way of then defining the kinds of human skills that would be required for future social and economic progress.

OECD graphic detailing technological and educational change. Source: OECD

In the Foreword, the OECD Director of Education, Andreas Schleicher, noted that ‘In a world in which the kinds of things that are easy to teach and test have also become easy to digitise and automate, we need to think harder how education and training can complement, rather than substitute, the artificial intelligence (AI) we have created in our computers.’

The project, Schleicher added, ‘is taking the first steps towards building a “PISA for AI” that will help policy makers understand how AI connects to work and education.’

The idea of a ‘PISA for AI’ is an intriguing one. The implication here is that the OECD might not only test human learners’ cognitive skills and capabilities, as its existing PISA assessments do, or their skills for work, as PIAAC tests do. It could also test the skills and capabilities of machine learners in order to then redefine the kinds of human skills that need to be taught, all with the aim of creating ‘complementary’ skills combinations. Ongoing assessments might then be administered to ensure human-machine skills complementarities for long-term economic and social benefit.

Computing Cognition

So how does the OECD plan to develop such assessments? One part of the report, authored by academic psychologists, details the ways cognitive psychology and industrial-organisational psychology have underpinned the development of taxonomies and assessments of human skills, including cognitive abilities, social-emotional skills, collective intelligence, and skills for industry. The various chapters consider the feasibility of extending such taxonomies and tests to machine intelligence. Another section of the report then looks at the ways the capabilities of AI can be evaluated from the perspective of academic computer science.

Given the long historical interconnections of cognitive science and AI—which go all the way back to cybernetics—these chapters represent compelling evidence of how the OECD’s central priorities in education have developed through the combination of psychological and computer sciences as well as economic and government rationales. In recent years it has shifted its attention to insights from the learning sciences resulting from advances in big data analytics and AI. Similar combinations of psychological, economic, computational and government expertise were involved in the  formation of the OECD’s assessment of social and emotional skills.

In the final summarizing chapter of the report, for example, the author noted that ‘the computer science community acknowledges the intellectual foundation and extensive materials provided by psychology,’ although, because the ‘the cognitive capacities of humans and AI are different,’ further work would require ‘bringing together different types of approaches to provide a more complete assessment of AI.’

The next stage of the AIFS project will involve piloting the types of assessments described in this volume to identify how well they provide a basis for understanding current AI capabilities. This work will begin with intense feedback from small groups of computer and cognitive scientists who attempt to describe current AI capabilities with respect to the different types of assessment tasks.

The project is ambitiously bringing together expertise in theories, models, taxonomies and methodologies from the computer and psychological sciences, in order ‘to understand how humans will begin to work with AI systems that have new capabilities and how human occupations will evolve, along with the educational preparation they require.’

Additionally, the project will result in some familiar OECD instruments: international comparative assessments and indicators. It will involve the ‘creation of a set of indicators across different capabilities and different work activities to communicate the substantive implications of AI capabilities,’ and ‘add a crucial component to the OECD’s set of international comparative measures that help policy makers understand human skills.’ In many respects, the OECD appears to be pursuing the development of a novel model of human-nonhuman skills development, and building the measurement infrastructure to ensure education systems are adequately aligning both the human and machine components of the model.

The idea of a ‘PISA for AI’ is clearly a hugely demanding challenge—one the OECD doesn’t foresee delivering until 2024. Despite being some years from enactment, however, PISA for AI already raises some key implications for the future of education systems and education policy.

Human-Computer Interaction

The OECD-CERI AI and the Future of Skills project is establishing artificial intelligence as a core priority for education policymakers. Although AI is already by now part of education policy discourse, the OECD is seeking to make it central to policy calculations about the kinds of workforce skills that education systems should focus on. The project may also help strengthen the OECD’s authority in education at a time of rapid digitalization, reflecting the historical ways it has sought to adapt and maintain its position as a ‘global governing complex.’

The first implication of the project, then, is its emphasis on workplace-relevant ‘skills’ as a core concern in education systems. The OECD has played a longstanding role in the translation of education into measurable skills that can be captured and quantified through testing instruments, as a means to perform comparative assessments of education systems and policy effectiveness. The project is establishing OECD’s authoritative position to define the relevant skills that future education systems will need to inculcate in young people. It is drawing on cognitive psychology and computer science, as well as analysis of changing labour markets, to define these skills, and potentially displacing other accounts of the purposes and priorities of education as a social institution.

A second implication stems from its assumption that the future of work will be transformed by AI in the context of a Fourth Industrial Revolution. The project seems to uncritically accept a techno-optimistic imaginary of AI as an enabler of capitalist progress, despite the documented risks and dangers of algorithmic work management, automated labour, and discriminatory outcomes of AI in workplaces, and a raft of regulatory proposals related to AI. Cognitive and computer science expertise are clearly important sources for developing assessment methodologies. The risk however is the production of a PISA for AI that doesn’t ask AI to account for its decisions when they potentially lead to deleterious outcomes. Moreover, matching human skills to AI capabilities as a fresh source of productivity is unlikely to address persistent power asymmetries in workplaces–especially prevalent in the tech industry itself–or counter the use of automation as a route to efficiency savings.

Third, the project appears to assume a future in which skilled human labour and AI perform together in productive syntheses of human and machine intelligence. While the role of AI and robotics as augmentations to professional roles may have merits, it is certainly not unproblematic. Social research, philosophy and theory—as well as science fiction—has grappled with the implications of human-machine hybridity for decades, through concepts such as the ‘cyborg,’ ‘cognitive assemblages,’ ‘posthumanism,’ ‘biodigital’ hybrids, ‘thinking infrastructures,’ and ‘distributed’ or ‘extended cognition.’ The notion that skilled human labour and AI might complement each other, as long as they’re appropriately assessed and attuned to one another’s capabilities, may be appealing but probably not as straightforward as the OECD makes out. Absent, too, are considerations of the power relations between AI producers–such as the global tech firms that produce many AI-enabled applications–and the individual workers expected to complement them. 

The fourth implication is that upskilling students for a future of working with AI is likely to require extensive studying alongside AI in schools, colleges and universities too. Earlier in 2021, the OECD published a huge report promoting the transformative benefits of AI and robotics in education. While AI in education itself may hold benefits, the idea of implanting AI in classrooms, curricula, and courses is already deeply contentious. It is part of longrunning trends towards increased automation, datafication, platformization, and the embedding of educational institutions and systems in vast digital data infrastructures, often involving commercial businesses from edtech startups to global cloud operators. As such, an emphasis on future skills to work with AI is likely to result in highly contested technological transformations to sites and practices of education.

Finally, there is a key implications in terms of how the project positions students as the beneficiaries of future skills. As an organization dedicated to economic development, the OECD has long focused on education as an enabler of ‘human capital.’ It has even framed so-called ‘pandemic learning loss’ in terms of measurable human capital deficits as defined by economists. In this framing, educated or skilled learners represent future value to the economies where they will work; they are assets that governments invest in through education systems, and the OECD measures the effectiveness of those investments through its large-scale assessments.

The AI and future skills program doesn’t just focus on ‘human capital,’ however. It focuses on human-computer interaction as the basis for economic and social development. By seeking to complement human and AI capabilities, the OECD is establishing a new kind of ‘human-computer interaction capital’ as the aim of education systems. Its plan to inform policymakers about how to optimize education systems to produce skilled workers to complement AI capabilities appears to make the pursuit of HCI capital a central priority for government policy, and it potentially stands to make HCI capital into a core purpose of education. Students may be positioned as human components in these new HCI capital calculations, with their value worked out in terms of their measurable complementarity with machine learners.

Counting learning losses

‘Learning loss’ is an urgent political concern based on complex measurement systems. Photo by Nguyen Dang Hoang Nhu on Unsplash

The idea that young people have ‘lost learning’ as a result of disruptions to their education during the Covid-19 pandemic has become accepted as common knowledge. ‘Learning loss’ is the subject of numerous large-scale studies, features prominently in the media, and is driving school ‘catch-up’ policies and funding schemes in many countries. Yet for all its traction, there seems less attention to the specific but varied ways that learning loss is calculated. Learning loss matters because it has been conceptualized and counted in particular ways as an urgent educational concern, and is animating public anxiety, commercial marketing, and political action.

Clearly educational disruptions will have affected young people in complex and highly differentiated ways. My interest here is not in negating the effects of being out of school or critiquing various recovery efforts. It’s to take a first run at examining learning loss as a concept, based on particular forms of measurement and quantification, that is now driving education policy strategies and school interventions in countries around the world. Three different ways of calculating learning loss stand out. First is the longer psychometric history of statistical learning loss research, second its commercialization by the testing industry, and third the reframing of learning loss through econometric forms of analysis by economists.

The measurement systems that enumerate learning loss are, in several cases, contradictory, contested, and incompatible with one another. ‘Learning loss’ may therefore be an incoherent concept, better understood as multiple ‘learning losses’ based on their own measuring systems.   

Psychometric set-ups

Learning loss research is usually traced back more than 40 years to the influential publication of Summer Learning and the Effects of Schooling by Barbara Heyns in 1978, although the earliest small-sample study was reported in 1906. The book reported on a major statistical study of the cognitive development of 3000 children while not in school over the summer, using the summer holiday as a ‘natural experimental situation’ for psychometric analysis. It found that children from lower socioeconomic groups tend to learn less during the summer, or even experience a measurable loss in achievement.

These initial findings have seemingly been confirmed by subsequent studies, which have generally supported two major conclusions: (1) the achievement gap by family SES traces substantially to unequal learning opportunities in children’s home and community environments; and (2) the experience of schooling tends to offset the unequalizing press of children’s out-of-school learning environments. Since the very beginning of learning loss studies, then, the emphasis has been on the deployment of psychometric tests of the cognitive development of children not in school, the lower achievement of low-SES students in particular, and the compensatory role that schools play in mitigating the unequalizing effects of low-SES family, home and community settings.

However, even researchers formerly sympathetic to the concept of learning loss have begun challenging some of these findings and their underlying methodologies.  In 2019, the learning loss researcher Paul T. von Hippell expressed serious doubt about the reliability and replicability of such studies. He identified serious flaws in learning loss tests, lack of replicability of classic findings, and considerable contradiction with other well-founded research on educational inequalities.

Perhaps most urgently, he noted that a significant change in psychometric test scoring methods—from paper and pen surveys to ‘a more computationally intensive method known as item response theory’ (IRT) in the mid-1980s—completely reversed the original findings of the early 1980s. With IRT, learning loss seemed to fade away. The original psychometric method ‘shaped classic findings on summer learning loss’, but the newer item-response theory method produced a very different ‘mental image of summer learning’.

Moreover, noted von Hippel, even modern tests using the same IRT method produced contradictory results. He reported on a comparison of the Measures of Academic Progress (MAP) test developed by the testing organization Northwest Evaluation Association (NWEA), and a test developed for the Early Childhood Longitudinal Study. The latter found that ‘summer learning loss is trivial’, but the NWEA MAP test reported that ‘summer learning loss is much more serious’. So learning loss, then, appears at least in part to be an artefact of the particular psychometric-set-up constructed to measure it, with results that appear contradictory. This is not just a historical problem with underdeveloped psychometric instruments, but persists in the computerized IRT systems that were deployed to measure learning loss as the Covid-19 pandemic set in during 2020.

Commercializing learning loss

Here it is important to note that NWEA is among the most visible of testing organizations producing data about learning loss during the pandemic. Even before the onset of Covid-19 disruptions, NWEA was using data on millions of US students who had taken a MAP Growth assessment to measure summer learning loss. Subsequently, the NWEA MAP Growth test has been a major source of data about learning loss in the US, alongside various assessments and meta-analyses from the likes of commercial testing companies Illuminate, Curriculum Associates, and Renaissance and the consultancy McKinsey and Company.

Peter Greene has called these tests ‘fake science’, arguing that ‘virtually all the numbers being used to “compute” learning loss are made up’. In part that is because the tests only measure reading and numeracy, so don’t count for anything else we might think as ‘learning’, and in part because the early-wave results were primarily projections based on recalculating past data from completely different pre-pandemic contexts. Despite their limitations as systems for measuring learning, the cumulative results of learning loss tests have led to widespread media coverage, parental alarm, and well-funded policy interventions. In the US, for example, states are spending approximately $6.5 billion addressing learning loss.

Learning loss results based on Renaissance Star reading and numeracy assessments for the Department for Education

In England, meanwhile, the Department for Education commissioned the commercial assessment company Renaissance Learning and the Education Policy Institute to produce a national study of learning loss. Utilizing data from reading and mathematics assessments of over a million pupils who took a Renaissance Star test in autumn 2020, the findings were then published by the Department for Education as an official government document. An update report in 2021, published on the same government webpage, linked the Renaissance Star results to the National Pupil Database. This arrangement exemplifies both the ways commercial testing companies have generated business from measuring learning loss, and their capacity to shape and inform government knowledge of the problem–as well as the persistent use of reading and numeracy results as proximal evidence of deficiencies in learning.

Moreover, learning loss has become a commercial opportunity not just for testing companies delivering the tests, but for the wider edtech and educational resources industry seeking to market learning ‘catch-up’ solutions to schools and families. ‘The marketing of learning loss’, Akil Bello has argued, ‘has been fairly effective in getting money allocated that will almost certainly end up benefiting the industry that coined the phrase. Ostensibly, learning loss is a term that sprung from educational research that identified and quantified an effect of pandemic-related disruptions on schools and learning. In actuality, it’s the result of campaigns by test publishers and Wall Street consultants’.

While not entirely true—learning loss has a longer academic history as we’ve seen—it seems accurate to say the concept has been actively reframed from its initial usage in the identification of summer loss. Rather than relying on psychometric instruments to assess cognitive development, it has now been narrowed to reading and numeracy assessments. What was once a paper and pen psychometric survey in the 1980s has now become a commercial industry in computerized testing and the production of policy-influencing data. But this is not the only reframing that learning loss has experienced, as the measurements produced by the assessment industry have been paralleled by the development of alternative measurements by economists working for large international organizations.

Economic hysteresis

While early learning loss studies were based in psychometric research in localized school district settings, and the assessment industry has focused on national-level results in reading and numeracy, other recent large-scale studies of learning loss have begun taking a more econometric approach, at national and even global scales, derived from the disciplinary apparatus of economics and labour market analysis.

Influential international organizations such as the OECD and World Bank, for example, have promoted and published econometric research calculating and simulating the economic impacts of learning loss. They framed learning loss as predicted skills deficits caused by reduced time in school, which would result in weaker workforce capacity, reduced income for individuals, overall ‘human capital’ deficiencies for nations, and thereby reduced gross domestic product. The World Bank team calculated this would costs the global economy $11 trillion, while the economists writing for the OECD predicted ‘the impact could optimistically be 1.5% lower GDP throughout the remainder of the century and proportionately even lower if education systems are slow to return to prior levels of performance. These losses will be permanent unless the schools return to better performance levels than those in 2019’.

These gloomy econometric calculations are based on particular economic concepts and practices. As another OECD publication framed it, learning loss represents a kind of ‘hysteresis effect’ usually studied by labour economists as a measure of the long-term, persistent economic impacts of unemployment or other events in the economy. As such, framing education in terms of hysteresis in economics assumes learning loss to be a causal determinant of long-term economic loss, and that mitigating this problem should be a major policy preoccupation for governments seeking to upskill human capital for long-term GDP growth. Christian Ydesen has recently noted that the OECD calculations about human capital deficits caused by learning loss are already directly influencing national policymakers and shaping education policies.

It’s obvious enough why the huge multitrillion dollar deficit projections of the World Bank and OECD would alarm governments and galvanize remedial policy interventions in education. But the question remains how these massive numbers were produced. My following notes on this are motivated by talks at the excellent recent conference Quantifying the World, especially a keynote presentation by the economic historian Mary Morgan. Morgan examined ‘umbrella concepts’ used by economists, such as ‘poverty’, ‘development’ and ‘national income’, and the ways each incorporates a set of disparate elements, data sets, and measurement systems.

The production of numerical measurements, Morgan argued, is what gives these umbrella concepts their power, particularly to be used for political action. Poverty, for example, has to be assembled from a wide range of measurements into a ‘group data set’. Or, as Morgan has written elsewhere, ‘the data on population growth of a society consist of individuals, who can be counted in a simple aggregate whole’, but for economists ‘will more likely be found in data series divided by occupational classes, or age cohorts, or regional spaces’. Her interest is in ‘the kinds of measuring systems involved in the construction of the group data set’.

Figures published by the OECD on the economic impacts of learning loss on G20 countries

Learning loss, perhaps, can be considered an umbrella concept that depends on the construction of a group data set, while that group data set too relies on a particular measuring system that aligns disparate data into the ‘whole’. For example, then, if we look at the OECD report ‘The Economic Impacts of Learning Loss’, it is based on a wide range of elements, data sets and measuring systems. Its authors are Eric Hanushek and Ludger Woessmann, both economists and fellows of the conservative, free market public policy think tank the Hoover Institution based at Stanford University. The projections in the report of 1.5-3% lower GDP for the rest of the century represent the ‘group data set’ in their analysis. But this consists of disparate data sets, which include: estimates of hours per day spent learning; full days of learning lost by country; assessments of the association between skills learned and occupational income; correlational analyses of educational attainment and income; effects of lost time in school on development of cognitive skills; potential deficits in development of socio-emotional skills; and how all these are reflected in standardized test scores.

It’s instructive looking at some excerpts from the report:

Consistent with the attention on learning loss, the analysis here focuses on the impact of greater cognitive skills as measured by standard tests on a student’s future labour-market opportunities. …  A rough rule of thumb, found from comparisons of learning on tests designed to track performance over time, is that students on average learn about one third of a standard deviation per school year. Accordingly, for example, the loss of one third of a school year of learning would correspond to about 11% of a standard deviation of lost test results (i.e., 1/3 x 1/3). … In order to understand the economic losses from school closures, this analysis uses the estimated relationship between standard deviations in test scores and individual incomes … based on data from OECD’s Survey of Adult Skills (PIAAC), the so-called “Adult PISA” conducted by the OECD between 2011 and 2015, which surveyed the literacy and numeracy skills of a representative sample of the population aged 16 to 65. It then relates labour-market incomes to test scores (and other factors) across the 32 mostly high-income countries that participated in the PIAAC survey.

So as we can see, the way learning loss is constructed as an umbrella concept and a whole data set by the economists working for the OECD involves the aggregation of many disparate factors, measures and econometric measurement practices. They include past OECD data, as well as basic assumptions about learning as being synonymous with ‘cognitive skills’ and objectively measurable through standardized tests, and a host of specific measuring systems. Data projections are constructed from all these elements to project the economic costs of learning loss for individual G20 countries, and then calculated together as ‘aggregate losses in GDP across G20 nations’ using the World Development Indicators database from the World Bank as the base source for the report’s high-level predictions.

It is on the basis of this ‘whole’ calculation of learning loss—framed in terms of economic hysteresis as a long-term threat to GDP—that policymakers and politicians have begun to take action. ‘How we slice up the economic world, count and refuse to count, or aggregate, are contingent and evolving historical conventions’, argues Marion Fourcade. ‘Change the convention … and the picture of economic reality changes, too—sometimes dramatically’. While there may well be other ways of assessing and categorizing learning loss, it is the specific econometric assembly of statistical practices, conventions, assumptions, and big numbers that has made learning loss into part of ‘economic reality’ and into a powerful catalyst of political intervention.

Counting the costs of learning loss calculations

As a final reflection, I want to think along with Mary Morgan’s presentation on umbrella concepts for a moment longer. As the three examples I’ve sketchily outlined here indicate, learning loss can’t be understood as a ‘whole’ without disaggregating it into its disparate elements and the various measurement practices and conventions they rely on. I’ve counted only three ways of measuring learning loss here—the original psychometric studies; testing companies’ assessments of reading and numeracy; and econometric calculations of ‘hysteresis effects’ in the economy—but even these are made of multiple parts, and are based on longer histories of measurement that are contested, incompatible with one another, sometimes contradictory, and incoherent when bundled together.

As Morgan said at the Quantifying the World conference, ‘the difficulties—in choosing what elements exactly to measure, in valuing those elements, and in combining numbers for those many elements crowded under these umbrella terms—raise questions about the representing power of the numbers, and so their integrity as good measurements’.

Similar difficulties in combining the numbers that constitute learning loss might also raise questions about their power to represent the complex effects of Covid disruptions on students, and their integrity to produce meaningful knowledge for government. As my very preliminary notes above suggest there is no such thing as learning loss, but multiple conceptual ‘learning losses’ based on their own measurement systems. There are social lives behind the methods of learning loss.

Regardless of the incoherence of the concept, learning loss will continue to exert effects on educational policies, school practices and students. It will buoy up industries, and continue to be the subject of research programs in diverse disciplines and across different sites of knowledge production, from universities to think tanks, consultancies, and international testing organizations. Learning loss may come at considerable cost to education, despite its contradictions and incoherence, by diverting pedagogic attention to ‘catch-up’ programs, allocating funds to external agencies, and attracting political bodies to focus on mitigation measures above other educational priorities.

Nudging assets

The acquisition of learning management system Blackboard has opened up opportunities for the new company to generate value from integrating data and nudging student behaviour. Photo by Annie Spratt on Unsplash

The acquisition of the global education platform Blackboard by Anthology has brought the mundane Learning Management System back to attention. While full details of the deal remain to be seen, and it won’t be closed until the end of the year, it surfaces two important and interlocking issues. One is the increasing centrality of huge data integrations to the plans of education technology vendors, and the second is the seeming attractiveness of the data-driven approach to edtech financiers.

Primarily, the acquisition of Blackboard by Anthology centres on business interests, according to edtech consultant Phil Hill, who notes that Blackboard’s owners have been seeking to sell the company for three years. The purpose of the deal, Hill argues, ‘is a revenue growth opportunity driven by cross-selling, international growth, and the opportunities to combine products and create new value, particularly at the data level.’ This approach, Hill further suggests, makes sense on the ‘supply side’ for vendors and investors who see value in combining data and integrating systems, if less so on the ‘demand side’ of universities and schools, whose primary concerns are with usability.

There are two things going on here worth questioning a little further. First, what exactly are Blackboard/Anthology hoping to achieve by combining data, and second, why is this attractive to investors? Based on some recent company blog posts from Blackboard, the answer to the first question appears to be about the capacity for ‘nudging’ students towards better outcomes through ‘personalized experiences’ based on data analytics, and the second question might be addressed by understanding those data as ‘assets’ with expected future earnings power for their owners. This post is an initial attempt to explore those issues and their interrelationship.

Nudging

One of the key features of the Blackboard/Anthology announcement was that it would enable much greater integration of the existing software systems of the two companies, including Learning Management System, community engagement, student success, student information system and enterprise resource planning. ‘Combining the two companies will create the most comprehensive and modern EdTech ecosystem at a global scale for education’, the CEO, president and chair of Blackboard, Bill Ballhause wrote in a company blog. ‘It will enable us to break down data silos, and surface deeper insights about the learner so we can deliver unmatched personalized experiences across the full learner lifecycle and drive better outcomes’.

The idea of breaking down ‘data silos’ and integrating data systems is part of a longer Blackboard strategy on making the most of cloud computing for cross-platform interoperability. Blackboard migrated most of its services to Amazon Web Services starting in 2015, with reportedly significant effects on how it could make use of the data collected by its LMS. ‘Our new analytics offering, Blackboard Data, is a good example where we are leveraging AWS technologies to build a platform that provides data-driven insight across all our solutions’, Blackboard reported in 2017. These insights will now be generated across the entire Blackboard/Anthology portfolio, raising data privacy and protection implications that Blackboard was quick to address just a day after the announced acquisition.

Beyond data privacy issues, though, the stated purpose of integrating data is to enact ‘Blackboard’s vision of personalizing experiences’. Writing earlier in the summer, Blackboard’s CEO Bill Ballhaus set out the company’s longer-term vision for personalizing learning experiences. Drawing on examples of online shopping, healthcare and entertainment, Ballhaus argued that a ‘critical mass of data powers proactive nudges’ based on highly granular personal data profiles. Education, however, had not yet ‘kept pace with the shift to customized experiences that other industries achieved’. This, he said, had now changed with the disruptions of the previous year.

‘The massive shift to online learning driven by the COVID-19 global pandemic enabled continuity of education in the near term, while opening the door for education to move forward on a journey toward more personalized experiences’, Ballhaus argued. ‘We’ve had our sights set on the future for the past few years and have the ability to securely harness data, with robust privacy protections, from across our ecosystem of EdTech solutions with the specific intent of enabling personalized experiences to drive improved outcomes’.

The discourse of ‘nudges’ as the central technique of personalized learning runs throughout this vision. ‘Students need nudges’ to reach better outcomes, Ballhaus continued, with ‘the 25 billion weekly interactions in our learning management and virtual classroom systems’ enabling Blackboard to operationalize such a nudge-based approach to personalized learning.

By emphasizing student nudges fuelled by masses of data as the basis of personalized learning, Blackboard has tapped into the logics of the psychological field of behavioural economics and its political uptake in the form of behavioural governance. Mark Whitehead and coauthors describe how behavioural governance has proliferated across public policy in many countries in recent years—especially the UK and US—through the application of nudge strategies. This has been amplified by digital ‘hypernudge’ techniques based on personal data profiles, which, as Karen Yeung argues, ‘are extremely powerful and potent due to their networked, continuously updated, dynamic and pervasive nature’.

So, the business plan behind the Blackboard/Anthology merger appears to be to enact a form of behavioural governance in digital education, operationalizing personalized hypernudges within the architectures of vast edtech ecosystems. While such a form of ‘machine behaviourism’ has existed in imaginary form for some years, it may now materialize in the seemingly mundane machinery of the learning management systems used by institutions across the globe. And that potential capacity for nudging also appears to be the source of expected future value for financial backers.

Assets

While the Blackboard/Anthology deal has been presented by the two companies as a merger, and interpreted by most as an acquisition of the former by the latter, in reality this is a deal between their financial backers and owners. Anthology is majority owned by Veritas Capital (a private equity firm investing in products and services to government and commercial customers), with Leeds Equity Partners (a private equity firm focused on investments in the Knowledge Industries) as a minority owner, while Blackboard is owned by Providence Equity Partners (a global private equity investment firm focused on media, communications, education, software and services investments). Veritas is providing new funding and retaining majority shareholder status, with both Leeds and Providence as minority shareholders following the acquisition.

The exact value of the deal remains unknown—Phil Hill has suggested it may be in the region of $3bn—but clearly these three private equity firms see prospects for value creation in the future. To interpret this, we need to understand some of the logic of investment. Recent economic sociology work can help here, particularly the concepts of capitalization and assetization.

As Fabian Muniesa and colleagues phrase it, capitalization refers to the processes and practices involved in ‘valuing something’ in terms of ‘the expected future monetary return from investing in it’. Capitalization, they continue, ‘characterizes the reasoning of the banker, the financier and the entrepreneur’, and calculating future expected returns is central to any form of investment. Capitalization then also depends on seeing something as an asset with future value, or making it into one. Kean Birch and Muniesa define an asset as any resource controlled by its owner as a source of expected future benefits, and ‘assetization’ thus as the processes involved in making that resource into a future revenue stream. Transforming something into an asset is therefore central to capitalization. 

Capitalization and assetization may be useful concepts for exploring the Blackboard/Anthology deal. Clearly, Veritas, Leeds and Providence as owners and shareholders are seeking future value from their assets. Their entire business is capitalizing on the assets they hold investments in, in expectation of return on investment. In part, the platforms that Blackboard and Anthology will combine are the assets. It is expected that more customers will purchase from them through cross-selling compatible products (e.g. by integrating Blackboard LMS with Anthology student information systems and making them interoperable for ease of use).

But given the prominence in the deal announcement and other posts of ‘breaking down data silos’ and ‘the possibilities of delivering personalized experiences fueled by data through our combination’, it seems likely that there is a process of assetizing the data themselves going on here. If the platforms and services themselves have future value, that is dependent upon the 25 billion weekly interactions of users as a new source of value creation. How are data made valuable?

In a recent study, Birch and coauthors highlight how ‘Big Tech’ companies transform personal digital data into assets with future earnings power both for the companies and their investors. They transform personal data into assets to generate future revenue streams. And Birch and colleagues argue that this assetization of user data occurs through the ‘transformation of personal data into user metrics that are measurable and legible to Big Tech and other political-economic actors (e.g., investors)’. In similar ways, then, the new Big EdTech company emerging from the combination of Blackboard and Anthology aims to transform student data into measurable and legible forms for their investors. 25 billion weekly interactions leave traces which can be made valuable.

As Janja Komljenovic has recently argued, ‘the digital traces that students and staff leave behind when interacting with digital platforms’ can be ‘made valuable by processing data into intelligence for either improving an existing product or service, or creating a new one, selling data-based products (such as learning analytics or other data intelligence on students), various automated matching services, automated tailored advertising, exposure to the audience, and so on’. The value comes not from the data themselves, but ‘from their predictive power and inducing behaviour in others’. In other words, as Komljenovic elaborates, ‘what becomes valuable in digital education is power over the direction of student and staff teaching, learning and work patterns. It is first about the power over calculating predictions and thus performing future, and second, about tailoring experience and nudging behaviour’.

In this particular sense, then, we can see how the objective of ‘nudging’ students through data-fuelled personalized experience may be a core part of the assetization process involved in the merger of Blackboard and Anthology. The platforms and services themselves as marketable products for institutions to pay for, or the weekly 25 billion data points of interaction with them, are not the only sources of expected value. Instead, the predictive capacity to shape education by personalizing experience and nudging student behaviours appears to be the key to unlocking future revenue streams.

Assetizing the nudge and nudging the asset in Big EdTech

The Blackboard/Anthology deal seems to foreground two complementary trends in the edtech sector. The first is that the ‘nudge’ has become the source of expected future value to asset owners. Personalized learning via digital nudges is clearly a core part of the expected value that Blackboard will return its new private equity owners and shareholders. This is assetizing the nudge.

The second is that student data have become the focus of the nudge, with digital nudges expected to increase student outcomes. In this sense, the masses of student data held by Blackboard/Anthology are being transformed into assets too. And if we understand those data to produce ‘data subjects’ or informational identities of a student, then we might conceivably think of students themselves as assets with value that can be increased through predictive nudging. This is nudging the asset, although it’s too early to see quite how this will work out in practice at the new company or in the institutions that use its services,

Maybe later details on the deal will help us clarify the precise ways assetization and nudging complement one another in an emerging environment of Big EdTech deals and integrations. It is important for critical edtech research to get up-close to these developments at the intersections of nudging and assetization, as practical techniques of behavioural governance and capitalization, even in the most mundane places like the LMS.

New biological data and knowledge in education

Research centres and laboratories have begun conducting studies to record and respond to the biological aspects of learning. Photo by Petri Heiskanen on Unsplash

Novel sources of data about the biological processes involved in learning are beginning to surface in research on education. As the sciences of the human body have been transformed by advances in computing power and data analysis, researchers have begun explaining learning and outcomes such as school attainment and achievement in terms of their embodied underpinnings. These new approaches, however, are generating controversy, and demand up-close social science analysis to understand what processes of knowledge production are involved, as well as how they are being received in public, academic and political debates.

Late last year, the Leverhulme Trust awarded us a research project grant to study the rise of data-intensive biology in education. As we now kick off the project, I’m really pleased to be working with a great interdisciplinary team that includes Jessica Pykett, a social and political geographer at Birmingham University; Martyn Pickersgill, a sociologist of science and medicine at Edinburgh; and Dimitra Kotouza, a political sociologist joining us at Edinburgh straight from an excellent previous project on the policy networks, data practices and market-making involved in addressing the ‘mental health crisis’ in UK higher education.

The project focuses on three domains of data-intensive biology in education:

• the emergence of ‘big data’ genetics in the shape of ‘genome-wide association studies’ utilizing molecular techniques and bioinformatics technologies including biobanks, microarray chips, and laboratory robot scanners to identify complex ‘polygenic patterns’ associated with educational outcomes
• neurotechnology development in the brain sciences, such as wearable electroencephalography (EEG) headsets, neuro-imaging, and brain-computer interfaces with neurofeedback capacities, and their application in school-based experiments
• rapid advances in the development and utilization of ‘affect-aware’ artificial intelligence technologies, such as voice interfaces and facial emotion detection for interactive, personalized learning, that are informed by knowledge and practice in the psychological and cognitive sciences

We are planning to track these developments and their connections with cognate advances in the learning sciences, AI in education, and recent proposals around ‘learning engineering’ and ‘precision education’. Across this range of activities, we see a concerted effort to employ data-scientific technologies, methodologies and practices to record biological data related to learning and education, and in some cases to develop responses or interventions based on it. We’re only just starting the project with the full team in place, but a couple of very recent developments help exemplify why we consider the project important and timely.

Controversy over the genetics of education

On the very same day our Leverhulme Trust grant arrived, 6 September, The New Yorker published a 10,000 word article entitled ‘Can Progressives Be Convinced That Genetics Matters?’ Primarily a long-form profile of the psychology professor Paige Harden, the article describes the long and controversial history of behaviour genetics, a field in which Harden has become a leading voice—as signified by the forthcoming publication of her book The Genetic Lottery: Why DNA Matters for Social Equality.

The main thrust of the article is about Harden’s attempts to develop a ‘middle ground’ between right wing genetic determinists and left wing progressives. She is described in the piece as a ‘left hereditarian’ who acknowledges the role played by biology in social outcomes such as educational attainment, but also the inseparability of such outcomes from social and environmental factors (‘gene x environment bidirectionality’). The article is primarily focused on the politics of behaviour genetics, which has long been a major field of controversy even within the scientific disciplines of genetics due to its ‘ugly history’ in eugenics and scientific racism.

Judging from reactions on Twitter among genetics researchers and educators, these are problems—both disciplinary and political—which are more complex and intractable than either the article or the science lets on. Concerns remain, despite optimistic hopes of a ‘middle ground’, that new molecular behaviour genetics insights will be mobilized and reframed by ideologically-motivated groups to reinforce dangerous genetically-reductionist notions of race, gender and class.

The New Yorker profile also notes that recent developments in genome-wide association studies (GWAS) have begun producing significant findings about the connections between genes and educational outcomes. These are ‘big data’ endeavours using samples of over a million subjects and complex bioinformatics infrastructures of data analysis, and are part of a burgeoning field known as ‘sociogenomics’. Again, many of these sociogenomics studies appear informed by the left hereditarian perspective—seeing complex, biological polygenic patterns related to educational outcomes as operating bidirectionally with environmental factors, and arguing that genetically-informed knowledge can lead to better, social justice-oriented outcomes.

But educational GWAS research and polygenic scoring informed by a sociogenomics paradigm is not itself a settled science. As I began illustrating in some recent preparatory research for this project, the scientific apparatus of a data-intensive, bioinformatics-driven approach to education remains in development, is producing very different forms of interpretation, and is leading to disagreement over its pedagogic and policy implications. Even from within the field, a behaviour genetics approach to education based on big data analysis remains a fraught enterprise. Outside the field, it is prone to being appropriated to support ideological right-wing positions and as fuel to attack so-called ‘progressives’ and their ‘environmental determinism’.

The controversy over behaviour genetics and education is not new, as Aaron Panofsky has shown. As part of a long-running series of critical studies and publications on behaviour genetics, he analyzed its involvement in promoting ideas about genetically-informed education reform. Focusing in particular in the work of behaviour geneticist Robert Plomin, Panofsky notes that his vision of genetically-informed education utilizing high-tech molecular genomics technologies represents a form of ‘precision education’ modelled on ‘precision medicine’ in the biomedical field. In precision medicine, doctors ‘could use genetic and biomarker information to divide individuals into distinct diagnosis and treatment categories’. A precision education approach would ostensibly use similar information to support ‘personalization’ according to students’ ‘different genetic learning predispositions’.

According to Panofsky, however, precision medicine ‘represents an approach to health and healing very much in line with our neoliberal political times’. It focuses, he argues, ‘toward “me medicine” that seeks to improve health through high-tech, expensive, privatized, individualized, and decontextualized intervention and away from “we medicine” that aims to improve health and illness in the broad public through focusing on widely available interventions and targeting health’s social determinants’.

Thus, for Panofsky, Plomin’s vision of precision education raises the risk that ‘while genetically personalized education is represented as a tool to help educate everyone, it represents more of a “me” approach than a “we” approach’. He argues it risks deflecting attention away from other educational problems and their social determinants–such as school funding, policy instability, workforce quality and labour relations, and especially underlying inequalities and poverty–by focusing instead on the identification of individuals’ biological traits and the cultivation of ‘each individual’s genetic potential’.

Overall, The New Yorker article helps illustrate the controversies that genetics research in education may continue to generate in coming years. It also shows how advances in data-intensive bioinformatics technologies and sociogenomics theorizing are already beginning to play a role in knowledge production on educational outcomes. As the high-profile publication of Harden’s The Genetic Lottery indicates, these advances and arguments are likely to continue, albeit perhaps in different forms and with different motivations. Robert Plomin’s team, for example, argues that ‘molecular genetic research, particularly recent cutting-edge advances in DNA-based methods, has furthered our knowledge and understanding of cognitive ability, academic performance and their association’, and will ‘help the field of education to move towards a more comprehensive, biologically oriented model of individual differences in cognitive ability and learning’.

A key part of our project will involve tracking these unfolding developments in biologically oriented education, their historical threads, technical and methodological practices, and their ethics and controversies.

Engineering student-AI empathy

The second development is related to ‘affect-aware’ technologies to gauge and respond to student emotional states. Recently, the National Science Foundation awarded almost US$20m to a new research institute called the National AI Institute for Student-AI Teaming (iSAT), as part of its huge National AI Research Institutes program.  One of three AI Institutes dedicated to education, iSAT is focused on ‘turning AI from intelligent tools to social, collaborative partners in the classroom’. According to its entry on the NSF grants database, it spans the ‘computing, learning, cognitive and affective sciences’ and ‘advances multimodal processing, natural language understanding, affective computing, and knowledge representation’ for ‘AI-enabled pedagogies’.

The iSAT vision of ‘student-AI teaming’—a form of human-machine collaborative learning—is based on ‘train[ing] our AI on diverse speech patterns, facial expressions, eye movements and gestures from real-world classrooms’. To this end it has recruited two school districts, totalling around 5000 students, to train its AI on their speech, gestures, facial and eye movements. The existing publications of iSAT are instructive of its planned outcomes. They include ‘interactive robot tutors’, ‘embodied multimodal agents’, and an ‘emotionally responsive avatar with dynamic facial expressions’.

The last of those iSAT examples, the ‘emotionally responsive avatar’, is based on the application of ‘emotion AI’ technology from Affectiva, an MIT Affective Computing lab commercial spin-out. The lead investigator of iSAT was formerly based at the lab, and has an extensive publication record focused on such technologies as ‘affect-aware autotutors’ and ‘emotion learning analytics’. In this sense, iSAT represents the advance of a particular branch of learning analytics and AI in education, supported by federal science funding and the approval of the leading US science agency.

Emotion AI-based approaches in education, like molecular behaviour genetics, are deeply controversial. Andrew McStay describes emotion AI as ‘automated industrial psychology’ and a form of ‘empathic media’ that takes ‘autonomic biological signals’ captured through biosensors as proxies for a variety of human affective processes and behaviours. Empathic media, he argues, aims to make ‘emotional life machine-readable, and to control, engineer, reshape and modulate human behaviour’. This biologization and industrialization of the emotions for data capture by computers therefore raises major issues of privacy and human rights. Luke Stark and Jesse Hoey have argued that ‘The ethics of affect/emotion recognition, and more broadly of so-called “digital phenotyping” ought to play a larger role in current debates around the political, ethical and social dimensions of artificial intelligence systems’. Google, IBM and Microsoft have recently begun rolling back plans for emotion sensing technologies following internal reviews by their AI ethics boards.

Over the last few years, several examples have emerged of education technology applications utilizing emotion AI-based approaches. They generally tend to provoke considerable concern and even condemnation, as part of broader public, media, industry and political debates about the role of AI in societies. Given that such technologies are already currently the subject of considerable public and political contestation, it is notable then that similar biosensor technologies are being generously supported as cutting edge AI developments with direct application in educational settings. While iSAT certainly has detailed ethical safeguards in place, some broader sociological issues remain outstanding.

The first is about the apparatus of data production involved in such efforts. iSAT employs Affectiva facial vision technology, which is itself based on the taxonomy of ‘basic emotions’ and the ‘facial action coding system’ developed in the 1970s by the psychologist Paul Ekman and colleagues. As researchers including McStay, Stark and Hoey have well documented, basic emotions and facial coding are highly contested as seemingly ‘universalist’ and mechanistic measures of the diversity of human emotional life. So iSAT is bringing highly controversial psychological techniques to bear on the analysis of student affect, in the shape of biosensor-enabled automated AI teaching partners. There remains an important social science story to tell here about the long historical development of this apparatus of affect measurement, its enrolment into educational knowledge production, and its eventual receipt of multimillion dollar federal funding.  

The second concerns the implications of engineering ‘empathic’ partnerships between students and AI through so-called ‘student-AI teaming’. This requires the student being made machine-readable as a biological transmitter of signals, and a subject of empathic attention from automated interactive robot tutors. Significant issues remain to explore here too about human-machine emotional relations and the consequences for young people of their emotions being read as training data to create empathic educational media.

In the research we are planning, we aim to trace the development of such apparatuses and practices of emotion detection in education, and their consequences in terms of how students are perceived, measured, understood, and then treated as objects of concern or intervention by empathic automatons.

Bio-edu-data science

Overall, what these examples indicate is how advances in AI, data, sensor technologies, and education have merged with scientific research in learning, cognitive, and biological sciences to fixate on students’ bodies as signal-transmitters of learning and its embodied substrates. While the apparatus of affective computing at iSAT tracks external biological signals from faces, eyes, speech and gestures as traces of affect, learning and cognition, the apparatus of bioinformatics is intended to record observations at the molecular level.

The bioinformatics apparatus of genetics, and the biosensor apparatus of emotion learning analytics are beginning to play significant parts in how processes of learning, cognition and affect, as well as outcomes such as attainment and achievement, are known and understood. New biologized knowledge, produced through complex technical apparatuses by new experts of both the data and life sciences, is being treated as increasingly authoritative, despite varied controversies over its validity and its political and ethical consequences. This new biologically-informed science finds traces of learning and its outcomes in polygenic patterns and facial expressions, as well as in traces of other embodied processes.

In our ongoing research, then, we are trying to document some of the key discourses, lab practices, apparatuses, and ethical and political implications and controversies of an emerging bio-edu-data science. Bio-edu-data science casts its gaze on to students’ bodies, and even through the skin to molecular dynamics and traces of autonomic biological processes. We’ll be reporting back on this work as we go.

Breaking open the black box of edtech brokers

Mathias Decuypere and Ben Williamson

Education technology brokers build new connections between the private edtech industry and state schools. Photo by Charles Deluvio on Unsplash

A new kind of organization has appeared on the education technology landscape. Education technology ‘brokers’ are organizations that operate between the commercial edtech industry and state schools, providing guidance and evidence on edtech procurement and implementation. Staffed by new experts of evaluation and decision-making, they act as connective agencies to influence schools’ edtech purchasing and use, as well as to shape the market prospects of the commercial edtech companies they represent or host. As a new type of ‘evidence intermediary’, these brokerage organizations and experts possess the professional knowledge and skills to mobilize data, platform technologies, and evidence-making methods to provide proof of ‘what works’, demonstrate edtech ‘impact’, and provide practical guidance to school decision-makers about edtech procurement.

Although brokers represent a novel point of connection between the edtech market and state systems of schooling, little is known about the aims or practical techniques of these organizations, or their concrete effects on schools. Edtech brokers are ‘black boxes’ that need opening up for greater attention by researchers and educators.

We are delighted to have received an award from a global research partnership between KU Leuven and the University of Edinburgh, which is funding a 4-year full-time PhD studentship to research the rise of edtech brokers with Mathias Decuypere (KU Leuven) and Ben Williamson (Edinburgh). The project will examine and conceptualize edtech brokerage as part of a transnational policy agenda to embed edtech in education, the operations of brokers in specific national contexts, and their practical influences on schools. The research will build on and advance our shared interests in digital platforms and edtech markets in education, as well as our broader concerns with data-intensive governance and digitalized knowledge production.

We have already identified a wide range of brokers to examine. One illustrative case is the Edtech Genome Project, ‘a sector-wide effort to discover what works where, and why’, developed by the Edtech Evidence Exchange in the US with partnership support from the Chan Zuckerberg Initiative, Carnegie Corporation, and Strada Education Network. It is building a digital ‘Exchange’ platform to enable ‘decision-makers to access data and analysis about edtech implementations’ with a view to both ‘increase’ student learning and save schools billions of dollars on ‘poor’ edtech spending. 

To a significant extent, we anticipate edtech brokers such as the Exchange becoming highly influential platform and market actors in education, across a range of contexts, in coming years.   

Post-Covid catalytic change agencies

In the context of the Covid-19 educational emergency, the role, significance and position of educational brokers have already grown: they are able to marshal their knowledge and expertise to advise schools on the most impactful edtech to address issues such as so-called ‘learning loss’ or ‘catch-up’ requirements.

These evolutions are not radically new. They reflect the increasing participation of private technology companies as sources of policy influence (supported by external consultancies, think tanks and international organizations) in education systems worldwide; and the rise in new types of ‘evidence’ production, including ‘what works’ centers and ‘impact’ programs, and the related emergence of new kinds of professional roles for evaluation and evidence experts in education.

However, especially during the Covid-19 emergency, brokers have begun asserting their expertise and professionality to support schools’ post-pandemic recovery, and creating practical programs and platforms to achieve that aim. Not only are edtech brokers positioning themselves as experts in evaluation and evidence, or as connective nodes between private companies and public education; they act as catalytic change agencies advising schools on the appropriate institutional pathways and product purchases to make for digital transformation.

Ambassadors and engines

We have initially identified two types of brokers:

1. Ambassador brokers represent either a single technology provider or a selected sample of industry actors. They provide sales, support and training for specific vendor products, including global technology suppliers such as Google and Microsoft, acting as supporting intermediaries for the expansion of their platforms and services into schools.
2. Search engine brokers function as public portals presenting selected evidence of edtech quality and impact to shape edtech procurement decisions in schools. They function as searchable databases of ‘social proof’ of ‘what works’ in the ‘edtech impact marketplace’, enabling school staff to access product comparisons and evaluative review materials.

Edtech brokers represent significant changes in the ways state education is organized. Both types of brokers operate as or through platforms that offer (part of) their services through digital means, exemplifying as well as catalyzing fast-paced digital transformations in education systems.

Ed-tech brokering is furthermore a global phenomenon, with initiatives variously funded by international organizations, philanthropies, national government agencies, and associations of private companies, representing concerted transnational and multisector reformatory ambitions to embed edtech in schools.

Edtech brokers all draw on ‘evidence’ and ‘scientific evaluations’, making it accessible and attractive for decision makers in schools. They are thereby shifting the sources of professional knowledge that inform schools’ decisions towards particular evaluative criteria of quality, impact, or ‘what works’. More particularly, ed-tech brokers are emblematic of the rise to power of new types of professionals and new forms of expertise in education.

Overall, we approach brokers as new intermediary actors in state education that are shifting the cognitive frames by which educators and school leaders think and act in relation to edtech. Brokers not only generally guide users’ decision-making processes and cognition; they equally contribute to structure particular forms of education and make specific forms of education visible, knowable, thinkable, and, ultimately, actionable.

The social lives of brokers

This project examines the transnational expansion of edtech brokering as a new organizational type and a new form of professionality in education, and provides an up-close empirical examination of its practical work and concrete effects, by opening the ‘black box of edtech brokering’ in different national contexts. We will utilize a social topology framework to study the policy ecosystem, platform interfaces, and data practices of edtech brokers, as well as their effects on school users of these services.

Exploring the fast-developing intermediary role of edtech brokers is crucial both for academic purposes as well as for the educational field itself: brokers are assembling the knowledge, expertise and platforms through which post-pandemic education will be defined. Because of their very position as connective intermediaries between specific schools and the edtech corporate world, brokers translate both the objectives of edtech companies and educational institutions into shared and context-specific aims. In doing so, they reformat, redo, restructure, and reconceive what education is or could be about.

Moving from the transnational level of edtech brokering as an emerging phenomenon to the ‘social lives’ of edtech brokers in action, the project will drill down to their influence on decision-making in schools in comparative national contexts. In countries such as Belgium and the UK, we have already observed how both ambassador and search engine brokers are actively seeking to influence the uptake and use of edtech in schools. The project will commence autumn 2021, with fieldwork to be carried out in Belgium and the UK.

Valuing futures

Ben Williamson

Education technology investors are imagining new visions of the future of education while calculating the market valuation of their investment portfolios. Photo by Lukas Blazek on Unsplash

The future of education in universities is currently being reimagined by a range of organizations including businesses, technology startups, sector agencies, and financial firms. In particular, new ways of imagining the future of education are now tangled up with financial investments in education technology markets. Speculative visions and valuations of a particular ‘desirable’ form of education in the future are being pursued and coordinated across both policy and finance.

Visions and valuations

Edtech investing has grown enormously over the last year or so of the pandemic. This funding, as Janja Komljenovic argues, is based on hopes of prospective returns from the asset value of edtech, and also determines what kinds of educational programs and approaches are made possible. It funds unique digital forms of education, investing speculatively in new models of teaching and learning to enable them to become durable and, ideally, profitable for both the investor and investee.

We’ve recently seen, for instance, the online learning platform Coursera go public and reach a multibillion dollar valuation based on its reach to tens of millions of students online. New kinds of investment funds have emerged to accelerate edtech market growth, such as special purpose acquisition companies (SPACs) that exist to raise funds to purchase edtech companies, scale them up quick and return value to both the SPAC and its investor, plus new kinds of education-focused equity funds and portfolio-based edtech index investing that select a ‘basket’ of high-value edtech companies for investors to invest in.

The result of all this investment activity has been the production of some spectacular valuation claims about the returns available from edtech. The global edtech market intelligence agency HolonIQ calculated venture capital investment in edtech at $16bn last year alone, predicting a total edtech market worth $400bn by 2025.

But, HolonIQ said, this isn’t just funding seeking a financial return—it’s ‘funding backing a vision to transform how the world learns’. These edtech investments tend to centre on a particular shared vision of how the future of education could or should be, and on particular products and companies that promise to be able to materialize that future while generating shareholder value. To this end, it just announced three ‘prototype scenarios‘ for the future of higher education, ‘differentiated by market structure’, as a way of developing consensus about desirable imaginaries and market opportunities for investment. The scenarios are imaginary constructs backed by quantitative market intelligence that HolonIQ has calculated with its in-house valuation platform. These are, to draw on the economic sociologist Jens Beckert, instruments of ‘fictional expectations’ that investment organizations craft to showcase their convictions and hopes, supported by specific devices of financial speculation that provide a more ‘calculative preview of the future’.

The aim of such instruments of expectation here is to stimuate speculative investments in new forms of education, and stabilize them as durable models for prospective future returns. The vision and the valuation of educational futures are intricately connected, and as Keri Facer recently noted, speculative investment of this kind is about making ‘bets’ on certain ‘valued’ educational futures while ‘shorting’ or foreclosing other possible futures for education.

What bets are being made? These bets are being made, for example, on the vision contained in the 2021 Global Learning Landscape report and infographic from HolonIQ. The landscape is a taxonomy of 1,250 edtech companies that HolonIQ has assessed in terms of their market penetration, product innovation, and financial prospects. As a fictional expectation inscribed in material form, the purpose of the infographic here is both to attract investors—for whom HolonIQ provides bespoke venture capital services—and to attract educational customers to ‘invest’ in institutional digital innovation through procuring from these selected services.

A persuasive vision or fictional expectation of the future of education is contained and transmitted in this infographic. As an instrument of expectation it emphasizes companies and products promising data-driven teaching and learning and analytics; online platforms such as MOOCs, online program management and other forms of public-private platform partnerships; AI in education, smart learning environments and personalized learning; workforce development and career matching apps, and other forms of student skills measurement and employability profiling. The infographic distills both an imaginative educational vision and a speculative investment valuation of the digital future of teaching and learning.

Education reimagined

The vision and valuation of educational futures are currently being joined together powerfully in the UK by an ongoing partnership between Jisc—the HE sector non-profit digital agency—and Emerge Education, a London-based edtech investment company. Jisc and Emerge have recently produced a series of visionary reports and strategy documents dedicated to Reimagining Learning and Teaching towards a vision of higher education in 2030. Together, the reports function as instruments of expectation with the intention of producing conviction in others that the imaginaries they project are desirable and attainable.

All the reports, written by Emerge with Jisc input, focus on the central fictional expectation of ‘digital transformation’ or ‘rebooting’ HE through partnerships with edtech startups, for example, in teaching, assessment, well-being, revenue diversification, and employability. They have produced an ‘edtech hotlist’ of companies to deliver those transformations, and created a ‘Step Up’ programme of partnerships between startups and universities to actively materialize the imaginary they’re pursuing.

The Jisc-Emerge partnership highlights how investment and policy are being coordinated towards a shared aim with expected value for HE institutions and for edtech companies and their investors at the same time. Exemplifying how investors’ fictional expectations catalyse real-world actions, this valued vision of HE in 2030 appears across the partnership’s reports, and especially in the main report also supported by UUK and Advance HE.

The report offers a vision of revolutionary digital acceleration, university adaptation and reimagining as digital organizations, characterized by personalized learning experience driven by artificial intelligence and adaptive learning systems that are modified automatically and dynamically. Universities are told to invest in their digital estates, learning infrastructure, personalized and adaptive learning, and AI. The sector is urged to adopt new data standards for the exchange of learner data, new micro-credentials, forms of assessment and well-being analytics.

The vision of learning and teaching ‘reimagined’ here, with the approval of Jisc, UUK and Advance HE, is highly congruent with the investment strategy of Emerge itself, with its emphasis on investing in a portfolio of ‘companies building the future of learning and work’. The fictional expectations and investment imaginary of Emerge have therefore been inscribed both into policy-facing documents and into its own strategic portfolio of investments.

Portfolio futures

So what this indicates is how edtech investment has become highly significant to how the future of teaching and learning is being imagined and materialized. Education futures are being imagined in parallel with market calculations and speculative investments, inscribed in graphical scenarios and calculative previews as instruments of expectation. Investment portfolios are being fused to policy imaginaries of education by way of shared fictional expectations that coordinate both policy and investment towards the same aims. Certain possible futures are being funded into existence or to scale.

Investment organizations are not just funding fortunate companies, but actively shaping how the future of education is imagined, narrated, invested in, and made into seemingly actionable strategies for institutions. By coordinating both policy and investment portfolios towards shared objectives, they’re valuing and betting on visions of digital transformation that promise prospective investment returns while devaluing and shorting alternative imaginaries of possible HE futures. This begs the question of how other futures of education can be produced, negotiated dialogically by educators, and invested in as a collective portfolio of counter-imaginaries of teaching and learning.