Automating what? Scholastic products and instructional automation in virtual schooling

ABSTRACT

Schools produce multiple products and digitization articulates with them in different ways. In this paper we expand the frame for analyzing instructional automation by examining its implications for three scholastic products – embodied learning, grades and test scores, and the narratives that connect the two. We draw on data from interviews with 47 teachers in four full-time virtual elementary and secondary schools in the US to argue that at present most of the actual work of automation in virtual schools is focused on the production of marks and grades, and that the narrative products of digitization efforts – routinization discourses, potentials discourse, and ‘live’ teaching discourse, play key roles in shaping how we understand the connections between those products and student learning.

KEYWORDS:

• Virtual schooling
• educational products
• instructional automation
• technology discourses
• live teaching
• routinization

Much of the research on educational technologies focuses on the automation of instructional tasks or the uses of digital and algorithmic systems for data analytics or surveillance. Critically-oriented scholars continue their attempts to pump the breaks on the techno-optimism (of scholars and investors alike) that drives these developments, and have generated important commentaries on issues of corporate power and educational governance (Perrotta, Gulson, Williamson, & Witzenberger, 2021; Sellar & Gulson, 2021; Williamson, 2018), changes to teacher labor (Selwyn, 2020, 2022; Selwyn, Nemorin, & Johnson, 2016), and the social and cultural contexts of automation efforts (Pink, Ruckenstein, Berg, & Lupton, 2022). However, even within this tradition, there’s little attention to how automation is articulated with the traditional products of schooling. This neglect is consistent with the broader definition of ‘automation’ as the replacement of human labor with machines to produce what is assumed to be the same product or outcome: to automate is to ‘substitute mechanical power for human musculature, machine-consistency for human handiwork, and digital calculation for slow and error-prone “wetware”’ (Autor, 2015, p. 5; cf. Bagrit, 1966, p. 14). Products aren’t supposed to change. You expect to get the same car whether robots or humans do the welds. You expect kids to learn at least as well interacting with software as in a traditional classroom.

The problem is that schools generate multiple products. Automation transforms some, leaves some unchanged, and puts others under erasure. To understand these interactions we need to examine the technological reconfiguration of schooling along the whole product line. We do this, albeit in an incomplete and provisional way, by focusing on a school sector arguably at the forefront of digital re-mediation – full time virtual elementary and secondary schools. Just as the power and sophistication of digital technologies would seem to be making the automation of teaching functions viable in such settings, these schools are moving in a contrary direction: raising the number of synchronous or ‘live teaching’ hours they offer, and giving teachers more responsibility over curriculum and teaching strategies. We use this seeming paradox as a conceptual pry-bar to raise questions about the nature of school products, the narratives of instructional digitization and automation emerging in educational studies, and finally the question of what purposes synchronous ‘live teaching’ serves in an asynchronous online environment.

Method and data

The setting for this work is the full-time virtual school sector of the US public education system. There were 477 of these schools in 2019–2020, enrolling over 330,000 students, about 30% of whom were in schools run by two charter management organizations (CMOs): Stride, Inc. (formerly K12 Inc.) and Pearson Education (Molnar et al., 2021, p. 4, 18). Although this is a conceptual article, we draw for illustration on data from semi-structured interviews (DeVault & McCoy, 2006) with 47 teachers from four virtual schools run by these CMOs in two states. We treat the teachers as knowing agents who can explain how they go about their work and how that work is articulated with larger institutional contexts. The approach could be described as a kind of ‘constitutive’ analysis (Pacewitz, 2020) – a claim about the nature of the phenomenon of scholastic automation – but is perhaps better understood as a problematizing analysis that aims at ‘intensifying or interrupting existing patterns of problem formation and problem solving’, in this case in the area of instructional automation (Barnett, 2015). This approach differs somewhat from other analyses of power and ideology in educational technology developments. Understood as a project of broadening the ‘horizon of intelligibility within which problems come to matter for people’ (Warner, 2002, p. 154), problematization nonetheless serves similar political purposes by decentering traditional problem configurations that, by guiding attention toward the state-of-the-art, obscure vantages of the state-of-the-actual (Selwyn, 2010). We begin with the narrative that has been most prominent in that area for the past century.

Discourses of automated routine

Teaching has been described as one of the occupations ‘least susceptible’ to automation. It requires ‘deep expertise and experience’, involves ‘complex interactions with other people’ (McKinsey Global Institute, 2017, p. 46), and depends on ‘flexibility, judgment and common sense’ – qualities that have proved ‘vexing to automate’ (Autor, 2014, p. 136). Yet for almost a century people have been automating parts of it. The trick here is that, although some writers argue that ‘automation technologies fully substitute for human labor’ (Benanav, 2020, pp. 5–6), most take the position that ‘automation does not occur at the job level but rather at the task level’, meaning that parts of the job – ‘tasks’ – can be parceled out among machines, professionals, contingent workers, customers, and clients (Sampson, 2021, p. 122). In the case of teaching, it’s the ‘routine’ tasks that have been targeted for automation.

‘Routine’ tasks are said to be easiest to give to the machines (Acemoglu & Restrepo, 2021), inasmuch as they involve ‘a limited and well-defined set of cognitive and manual activities, those that can be accomplished by following explicit rules’ (Autor, Levy, & Murnane, 2003, p. 1280). Explicit rules can be translated into a ‘fully specified’ set of instructions that can be repeatedly executed by a machine (Acemoglu & Autor, 2010, p. 20), resulting in work that, at least in theory, gets done at greater speed with more consistency than a human could do it. The discursive continuity over time in framing instructional automation in such terms is striking. In the 1920s, Pressey wrote of automating parts of teaching – in particular testing – to

free the teacher from ‘mechanical tasks of her profession – the paperwork and routine drill – so that she may be a real teacher, not largely a clerical worker’. (Quoted in Petrina, 2004, p. 313; Brass & Lynch, 2020)

Skinner (1968) similarly argued that teaching machines could ‘free [the teacher] for the effective exercise’ of more ‘intellectual, cultural, and emotional contacts’ with students (pp. 26, 27). In the 1950s and 60s, ‘Instructional television’ proponents imagined ‘taking away the teacher’s basal expository role by assigning that to the media’, thus freeing ‘him/her to assume the role of diagnostician of, and prescriber for, individual student strengths and weaknesses within the varying disciplines’ (Berkman, 1977, pp. 102–103). Contemporary automation advocates position digital technologies as serving similar functions. Fishman and Dede (2016) suggest that ‘digitized pedagogical agents’ can enable ‘teachers to extend and deepen their own activities’ (pp. 1272–1273), and UNESCO touts a model in which ‘virtual teaching assistant … takes over the teacher’s routine task, frees up teachers’ time, enabling them to focus on student guidance and one-to-one communication’ (Pedro, Subrosa, Rivas, & Valverde, 2019, p. 13).

Phrases like ‘routine drill’ or ‘basal expository role’ imply that portions of the teacher’s work are intrinsically routine and readily reducible to explicit rules amenable to automation. This ignores the fact that such routine elements are products of earlier technologies – textbooks, worksheets, multiple-choice tests, grade ledgers, and the like. Instructional technologies don’t as a rule replace routine work. They routinize work (and sometimes create new tasks). Current automation efforts are laminated onto older technological incursions, stabilizing and reinforcing them while routinizing new task areas (Selwyn, 2020, 2022; Selwyn et al., 2016). Things like ‘automated grading … the personalization of stock content towards individual students, automated student feedback and content filtering’ (Perrotta et al., 2021, p. 8) would not have seemed like new ideas to earlier teaching machine advocates (e.g. Skinner, 1968).

More problematically, the emphasis on ‘routine’ has deflected attention from what teaching produces – teachers engage in ‘student guidance and one-to-one communication’ to produce what? The idea that teaching is riddled with routine elements assumes that the outcomes of schooling can remain unchanged as the processes generating them change. The question of what’s being produced goes unasked.

Automating what?

To understand automation efforts, we need to examine at least three kinds of products associated with both traditional and virtual schools (we ignore the new data-products being generated through educational digitization). First, there are repertoires of speaking, acting, and participating that students assemble through classroom interactions across school timespace with peers and adults. The processes of participation and identification generated by co-present schools are aspects of learning as a ‘process of transformation of participation’ (Rogoff, 1995, p. 209). Learning, in a popular formulation, involves becoming a ‘kind of person’ in relation to specific collectives or systems of activity (Lave & Wenger, 1991, p. 53). Call this kind of learning Product 1. In full-time virtual schools, automation displaces these processes. The physical school itself is a cognitive artifact: students mobilize ways of seeing, speaking, and moving, in and out of classrooms, that generate learning contexts for one another. Inside classrooms teachers orchestrate emergent orders shot through with multiple streams of sense-making, intention, feeling, and improvised adaptations. Digital instruction in its present forms cannot automate these processes and instead replaces them with radically simplified modes of interaction and participation.

Second, another school product consists of the marks, scores, and grades that represent student performance. In principle these scores reflect the quality of time spent in the embodied work described above. In practice, the relation of such texts to student actions is ambiguous. Students with the same scores can claim the same knowledge regardless of how they produced the scores or how they can use the knowledge the scores supposedly represent. Put another way, traditional brick and mortar schools are alchemy chambers for transforming embodied capacities and emergent interactions into radically standardized and simplified textual records structured around a limited number of standardized institutional inscriptions. Virtual schools are mechanisms for transforming evidence of attendance and items of student work into the same set of inscriptions. It is impossible to determine if inscriptions generated through either type of schooling actually represent the learning of specific competencies or abilities.

This differentiates schools from other sites of automation. Replace a welder with a machine and you expect the same (or better) weld. Automate the production of grades and you get the same grade but no way to know what the grade indexes, or indeed if it is anything more than a signal that allows others (such as prospective employers or college admissions officers) to gamble on the holder’s potential (Bills, 2003). Unlike online retailers that have to cough up boxes full of stuff, digital instructional systems produce boxes – scores, marks, credentials – that buyers can’t open. However, unlike learning-as-participation, such mark/score/grade boxes – Product 2 – are amenable to automation. Software can be designed to generate a test, students can automate part of their work by searching the web for answers, and software can then grade the test, check for cheating, and generate a score or mark. A score produced this way has the same institutional status as the score of a student who produced it through emergent, collective, socially and technically-mediated work in a brick and mortar setting. Unless the academic work is integrated with work or on-the-job training (e.g. Busemeyer, 2009), grades and scores are at best ‘signals’ of a person’s capabilities in out-of-school activities. And if you can automate the production of these product shells, it is in principle possible to automate the agent responsible for their production. Taken to its logical conclusion, pretty much any system that produces grades or marks will do. Proof of concept:

For more than a year, a teacher at one of Ohio’s largest online schools hired a former student to teach her classes and work with students. The students apparently had no idea their ‘teacher’ was a college dropout whose work experience consisted of stints at Burger King and in a hospital laundry. The Electronic Classroom of Tomorrow found out about the incident only after the hired student complained to the school about not being paid. (State Impact, 2012)

As long as you can’t see inside the box, and the arrangement generates the proper inscriptions – grades and attendance records – it would appear to an outsider, and to the pupils participating, to be working.

Third, schools generate accounts of the processes through which it produces Product 2. Call this Product 3. The makers and sellers of most commodities use advertising and marketing techniques to create narratives and brand images for products, but with few exceptions (for things like ‘fairtrade’, or ‘cruelty-free’ commodities) they don’t dwell the production process – indeed, many try to keep buyers from thinking about where things come from: images of CAFOs and slaughterhouses plants won’t help you sell meat, and stories about soil degradation and sweatshops won’t make that pair of jeans more appealing. But with education we do expect a narrative. Traditional schools themselves provide a story schema. As Meyer (1977) points out, ‘formalized educational systems are, in fact, theories of socialization institutionalized as rules at the collective level’ (p. 65) – in this case, theories that portray learning as an individuated process requiring the guidance of teachers in sequestered spaces under the control of state or religious agents, with grades and marks representing objective assessments of the quality of the individual student’s accomplishment. The story tells us that it’s the teacher and especially the students who do the work.

Virtual schooling and instructional automation schemes disrupt this narrative. Digitization pushes Product 1 – collective embodied learning – out of the frame and substitutes virtual or digitally/textually-mediated interactions for classroom interaction. At the same time, the use of software-structured tasks and machine-graded tests automates major elements of the production process for grades, marks, and test scores (Product 2). This creates a need for a compelling narrative (Product 3) that explains the connection between the pupil’s activity to the mark and places responsibility for marks on particular actors.

Technology as product narrative

Prominent narratives in play now portray the new technologies as working the ways their corporate producers say they intend them to. As we shall see, the ‘replacing routine tasks’ narrative remains central to corporate accounts, but here we examine the way that technology itself becomes a narrative. Much of the research literature, pro or con, seems to assume that technical capabilities demonstrated on a small scale can be implemented on a much wider scale, and that corporate agendas include a desire to maximize the use of sophisticated instructional technologies to improve education or help children learn. Corporate marketers and designers are taken at their word, with the claims of proponents serving as descriptions of corporate intent or practice (e.g. Perrotta et al., 2021; Zeide, 2020). At best, academic analyses explicate assumptions congealed in design (e.g. Selwyn, 2022), but many read like theoretical embellishments on marketing literature. ‘Automation’ has become a discursive project in the sense that demonstration or prototype versions of artifacts or digital systems function as material props for narratives of imagined potency or financial return. Whether the system or tool can do what’s claimed for it is a secondary issue at best. The US military’s heavily automated F-35 fighter, for example, promises to ‘deliver an awesome combat capability’ (Carey, 2016) but in reality has been a non-functioning disaster (Hruska, 2021). Yet the funds still flow (Cockburn, 2021), as they do for other automated systems known to be faulty and biased (e.g. Eubanks, 2017; Stop LAPD Spying Coalition, 2021). The primary aim of technological extravaganza is to secure funding and organizational expansion (Cockburn, 2021). There are non-military analogies in areas such as the ‘platform economy’, where corporations use digital technologies to make money not by generating profit (at least not in the short term) but by generating speculative investment and gaining monopolies by driving competitors out of business (Horan, 2019; Roy, 2020).

The argument is not that it’s impossible to automate core processes in such systems, nor are we questioning the sincerity or intentions of AI workers and designers (see Eynon & Young, 2021). The point is that a successful narrative product allows one to prosper even if the first order product fails or works poorly. Even if developers and instructional designers really want to improve learning for everyone (whatever that might mean), the corporations funding the work are in it to make profit, and at least for an indefinite span of time they don’t really need a workable product – such as student learning – to make money. The potentials and purported effects of the technologies are part of a hyperbolic discourse that works by misdirection.

Potentials discourse

We can use an extreme case to illustrate the point. Consider, for example, Williamson’s (2016) description of Pearson’s ‘Center for Digital Data, Analytics and Adaptive Learning’ (CDDAAL) as a site for the ‘reconfiguration of the methods by which learning is conceptualized, measured and understood’ (p. 38). Such efforts are portrayed as part of a shift to forms of governance in which experts claim authority based on the ‘methodological and technical capacity to know, asses and act upon education through data collection, aggregation and analysis’ (p. 38). While Williamson draws out a number of points that communicate a critical perspective on the governance shifts implied (and created) by Pearson’s ‘global education data infrastructure’, the paper repeats the company’s own claims regarding the Center’s activities – that CDDAAL is successfully ‘mapping and modelling the generalizable patterns of learning processes and cognitive development’ (p. 44). This repetition amplifies and legitimizes the hyperbolic marketing claims that construct what Williamson himself recognizes elsewhere as ‘imagined futures’ in education, designed to appeal to edtech investors (Williamson & Komljenovic, 2022).

The image given, both by Pearson advertising copy and Williamson’s paper, is of a strong automation effort, but the account is padded with qualifications. We see descriptions of a software platform that is designed to track individual students –

through their digital data traces in real time and to provide automated predictions of future progress … Prescriptive analytics can then be mobilized as ‘recommender systems’ for personalized pedagogic intervention … Its new data-derived models of learning and cognitive development have the potential to shape how pedagogic practitioners and policymakers understand what learning is and how to activate it through specific pedagogic resources, approaches and applications. (Williamson, 2016, p. 47, 49)

Phrases like ‘have the potential to’ are common in such articles (it’s used five times in this one), and naming ‘potentials’ may be useful. But focusing on them fetishizes the technology (Hornborg, 2001), and deflects attention from the animating logic that shapes the direction of development in instructional automation: profit (Wajcman, 2017, p. 124). Virtual schools, publicly funded ones in particular, are sources of profit for corporations like Pearson and Stride, Inc. The money comes from state funding based on school attendance. The revenue-generating potential of these organizational forms is less a matter of ‘optimizing learning’ or ‘personalizing’ curriculum than of making sure students show up, and, in the case of publicly-traded CMOs, convincing investors of the speculative value of the stock. Neither of those things depend on the software living up to its ‘potential’. This is a mode of ‘incantatory governance’ in which ‘performances, symbols and narratives appear to be just as important as the production of rules, institutions and instruments’ – or working technologies (Aykut, Morena, & Foyer, 2021, p. 3).

This is something obvious to teachers working in full-time virtual schools run by CMOs like Pearson, Inc. or Stride, Inc., both of which (for a price) provide personnel, curriculum, and software to the schools they manage. While these managed schools have to produce attendance records, grades, and scores in order to maintain funding streams, they do not have to produce ‘personalised pedagogical interventions’, ‘optimal learning’ or anything else along those lines. As a teacher explained in 2020:

Oh yeah – so, Pearson bought our school, because it made so much money … And they brag all the time about how they’ve made millions of dollars off of our … school every year. (Ms Parker)

‘Digital data traces’ and analytics do play a role in this teacher’s work, but the key analytics used by the administration have to do with the enrollments that determine the school’s revenue:

You meet with your supervisor twice a month. You go over numbers. It’s very data-driven. So how many students did you call this week? How many did you actually speak to? Where’s your grade-point at, like grade distribution at. What are you doing to help students? … But there’s no real evaluation for what you’re teaching or how you’re teaching it, because everything’s driven by Pearson. It’s more, ‘how are you talking to kids, how are you connecting with families, and making sure that they’re logging in and doing work’. (Ms Parker)

Dealing with classes of 200 or more students, the emphasis is on keeping kids on the books and generating Product 2. Pearson supplies a curriculum, but it’s ‘personalized’ only in the sense that teachers can edit it – a kind of ‘complementary labor’ (Shestakofsky, 2017) needed to make the schools run. Other forms of editing and augmentation simply consist of configuring a doable work system:

The curriculum is dictated by Pearson as a whole. Everything from the lessons to the tests to the quizzes – every assignment the kids do is all given. Whether we choose to modify any of those, or change them or remove them is up to us as a teacher … [The accrediting agency] reviews it and goes, ‘Oh, this is great … you have all of the content, you have all of the things you’re supposed to have’. But each teacher has autonomy over what they want to actually teach and what they want to remove in their course. And all three of those [other teachers’] courses removed half of the curriculum. (Ms Parker)

As long as students attend and the instructional mechanism produces grades the school will appear to be – and is, from the corporate perspective – ‘working’. The teacher’s activity need only be loosely coupled to this production process:

I’m supposed to work 7:00 to 3:30. And I’m allowed a 30 minute lunch in the middle of the day. Other than that I’m supposed to be on the computer, doing whatever. I usually get up at 7:00 and walk my dog, work out a little bit, take a shower – and maybe log in by 9.00 or 9:30, unless there’s a meeting scheduled for some reason. I do a little bit of grading. I check on all my homeroom kids. And then I start making calls … I’m usually taking a nap by 1:30, 2:00. (laughs) … I mean, I’m always available for my students. Don’t get me wrong. If I have a kid or parent that says ‘I need to talk to you at 7:30 because that’s what works for me’, I will obviously be available. And I always keep my phone on me. I definitely work after hours. (Ms Parker)

This is not a typical workday for the teachers we interviewed, but the limited autonomy of the teacher, the student 156 lessons behind in their work, the curriculum cut in half, naps in the afternoon, and so on, also represent ‘potentials’ of instructional automation mostly absent from corporate and academic discourse. If the school’s Product 2 can be automated – the grade and scores allocated to students – what students and teachers actually do can be effectively blackboxed.

Discourse lines and bottom lines: why live teaching?

Blackboxes, however, need stories.

Why, if you can run the school Ms Parker describes, and your management organization is churning out things like ‘personalized pedagogical interventions’, have the fully online schools run by the large CMOs been increasing the time during which their ‘error-prone wetware’ is in use? The corporations have been ratcheting up the number of ‘live’ hours – the hours of synchronous instruction that teachers are required to teach each day – for at least a decade: ‘When I started at School 2, it was 2 sessions, then we went to 3. And then last year was the first one that we went to 4 [45 minute classes a day]’. (Ms Owens)

Ms Rainier:

[teaches four 40-minute classes a day] They change the schedule every single solitary year … When I was hired … I think I only taught like, like 3 to 5 hours a week … I still just have a handful of kids participating … Like 180 kids – I would still just have like 3 kids participating … with [a state-tested course] I will still just get maybe 20 percent.

Some teachers suggested these increases were ways for the corporations to document attendance – a critical function to ensure the school’s state funding: ‘We have to show attendance, and we can make sure kids are attending school by actually being in class with us’ (Ms Larson). As Ms Parker’s and Ms Rainer’s accounts show, however, most students don’t attend the offered sessions.

Instead, live teaching is best understood in relation to the third form of product we described: it supplies a three-stranded production narrative. One strand is temporal. Some critiques of full-time virtual schools argue that open asynchronous systems allow students to finish courses in days or even hours (e.g. Molnar, 2014). Live teaching suggests an embodied speed limit and supports the implicit narrative that grades and marks are warranted by the synchronous interactions of teachers and pupils. A second, related narrative strand is that synchronous teaching leads to better student scores and grades.

Ms Gardener:

Their original business model was that we’d basically be graders, you know, that they could leverage it. Like you could have 300 or 400 students because these kids would just learn, but they realized that no, funny thing is kids don’t just learn from reading, you know, or looking at BrainPOP movies.

Ms Brighton:

We use a [CMO-made] curriculum. So, when they initially created it, they thought it was perfect, so students could just get on and do it, and really you don’t need a teacher, right? So they quickly learned that that wasn’t the case and that students really need their teachers.

If asynchronous arrangements make the teacher’s work hard to see, ‘live teaching’ makes it more visible, even ‘hyper-visible’ in the sense that workers are ‘deliberately spotlighted by employers as part of the service relationship’ (Poster, Crain, & Cherry, 2016, p. 10).

This is tied to a third narrative strand. Virtual school corporations target students who are structurally marginalized by traditional schools (e.g. Bottari, 2013), including students who move a great deal, who fall behind because of illness, who are constructed as disciplinary problems, or who do relatively poorly on tests. Such students provide part of the corporate rationale for increasing live teaching. Stride, Inc explains: ‘We provide more synchronous sessions for at-risk students based on data driven instruction that provides for targeted teacher intervention to assist students with lesson challenges’ (K12, Inc., 2020, p. 13). This folds narratively into the older discourses of automation as substitution-for-routine-work: technology ‘allows teachers to direct their attention towards the students who need it most, while enabling more proficient students to continue making progress on their own’ (Fishman & Dede, 2016, pp. 1272–1273). Despite this narrative, requiring synchronous instructional hours in no way guarantees the provision of targeted pedagogical supports to at-risk students. For example, Ms Caruno described her live academic support sessions with students as more focused on how to catch up on backlogs of incomplete work than on supporting their understanding of the associated concepts:

I bring them to academic support and I’m like ‘Listen. This is overwhelming that you’re this far behind in all of your classes. This is what I would do. This is how I would handle it … Okay, for the next two days just work on your fine art class. Get all of your assignments made up, and then after that day, or however long it takes you’ … I say just focus on one class and get that class caught up.

The narrative highlighting of ‘at-risk’ students as a target population provides management corporations with a reservoir of explanations for their schools’ poor performances on accountability measures. It also locates pedagogical problems in the students. While the digital infrastructure may have the potential (to use that word again) to free teachers to work more intensively with at-risk students, the teachers we interviewed described themselves as more focused on coordinating the students’ work with that infrastructure in a way that would minimally produce a corpus of work to be evaluated (to produce a Product 2). Although the systems of automated prediction and intervention that guide students’ asynchronous engagement in some online schools presuppose a normalized version of the student around which algorithms have been constructed (Witzenberger & Gulson, 2021), the schools can function so long as the students produce enough work: hence Ms Caruno’s emphasis on catching up on incompletes. Here we see the beginnings of another scholastic product, in this case one internal to the school: the paired categories of students as computer-ready and machine-readable – that is, students producing the required textual work through asynchronous activity – or as problematic due to the ‘unpredictability, inconsistency, or resistance’ implied by their failure to produce such work. In the latter case, ‘live’ teachers and students are narrated into a supposedly integrated synchronous instructional system for producing the necessary work. Automated systems, in a sense, desire automated subjects, in this case students visible only through the production of textual products slotted into predefined categories, since actual students (who might, for example, neglect to produce any texts) ‘threaten to gum up the works’ (Andrejevic, 2020, p. 2). The narrative of live teaching and the gradual increase in the number of synchronous hours provided (whether or not more students attend) repairs any damage done to the potentials discourse, while at the same time deflecting attention away from the fact that the automated instructional systems used in these schools don’t work particularly well for many categories of students. Again, this is not necessarily a problem from the corporate perspective, so long as the automated systems for producing Product 2 – the inscriptions recording attendance, scores, credits, etc. – function without interruption and the potentials discourse continues to attract speculative investment and generate corporate profit.

Conclusion

We do not intend to deny the possibility of automating aspects of teaching involved in producing what we called Product 1 – learning through participation, assembling identities in complex activity systems, and so forth. Rather, we argue that, at least in the present context, corporate schools can get by – that is, make money – by instead automating the production of Product 2 and wrapping it up in narratives that constitute a third scholastic product. Corporate developers’ need to generate profit and ensure continued speculative investment motivates the circulation of hyperbolic discourses that foreground the potentials of their technologies. When these discourses are reproduced by scholars, they obscure the role of corporate self-interest in projects of educational automation. Reconciling hyperbolic potentials discourse with on-the-ground technical capacities and priorities is an important project – we are certainly not the first to point this out. In the context of schooling, part of this reconciliation requires the recognition that changes to the ‘production process’ unavoidably result in changes to the product(s).

Even the relatively mundane uses of automation in full-time virtual schooling transform teaching and learning by eliminating the embodied, collective interactions that are core products of traditional co-present schooling. By contrast, virtual schools can easily generate the grades and marks that form the representational products of the school: asynchronous arrangements drop teachers out of the loop and automate major elements of instruction (e.g. task assignment, assessment, recording keeping). These are the critical products. The problem for owners and investors in virtual schools is not managing behavior or ensuring that students learn. It’s making money. In the US, the money comes from attendance-based state funding. To get it, the schools have to get kids enrolled and generate evidence that they attend. Part of that evidence consists of showing that the kids did recordable work – that they made grades and produced test scores. A virtual school doesn’t have to produce ‘personalized learning interventions’ or have systems that do a particularly good job of diagnosing and remediating student learning needs as long as it produces the grades and marks.

A third school product is the narrative explaining how those marks are produced. In traditional schools one could point to the temporal and spatial enclosure of the classroom and the work of the visible teacher. For virtual schools, potentials discourse and arguments for automating ‘routine’ teaching tasks supply alternative narratives that are targeted more toward the speculative investors in the schools’ parent company than the students and families who engage with the actually existing instructional systems. Drawing on two ‘modalities’ of genealogical work (Harcourt, 2022, pp. 6–7) we have tried to ‘problematize’ these narratives as well as ‘debunk’ them, unpacking the taken-for-grantedness of the idea that there are automation-ready routine teaching tasks, while also ‘unveiling the illusions’ of potentials discourse, in part by weaving the ongoing problematizations of actors such as Ms Parker into our accounts. As Barnett (2015) suggests, debunking approaches can overestimate the solidity of settlements and the pervasiveness of illusions. For Ms Parker and others the illusions of virtual schooling were already in tatters; interfolding her critique with ours is a way of turning problematization into ‘the intensification of always already difficult situations’ (Barnett, 2015). If the aim of genealogy is to inform praxis, inspire people to act (Harcourt, 2022), such articulations are a way of mapping landscapes of alliance to resist the kinds of educational practices. Absent such research with teachers (and students), researchers’ accounts risk reinforcing instead of challenging the school’s legitimizing narratives.

Acknowledgements

The authors would like to thank the anonymous reviewers for Discourse for identifying problems and ambiguities in earlier drafts and for giving us suggestions that we think have improved the article a lot.

Disclosure statement

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