Childhoods, circulating II: artificial intelligence, the loT and infrastructures

The examples in the previous section demonstrated not only how childhoods circulate but, more specifically, how those circulations are composed of both bits of

childhoods and, simultaneously, bits associated with childhoods. Dolls are a good example: as bits of childhoods, they comprise extensions of childhood experience and, indeed, as Aitken and Herman (1997) showed, are one of a number of ‘transitional’ objects that help a child make sense of themselves and the world. And as bits associated with childhoods they are - in the collective as well as the particular - some of the most enduring and ‘obvious’ objects. Yet, to repeat and extend Livingstone’s (2019) point, neither these material objects nor digital traces correspond entirely with (in her terms) ‘real’ selves or subjects. Connecting Livingstone’s argument with OOO and speculative-realist philosophies, this is - in a flat ontology - because just as with any other object or machine, human bodies can themselves not be exhausted, actualised or visualised in their depth or extent (as Deleuze and Guattari [1988] also famously argued through then concept of the Body without Organs).

Continuing these lines of thought, the purpose of this section is twofold: first, through juxtaposing a range of fairly brief examples taken from contemporary media and academic coverage about Al and the loT, to tackle one of the key debates about contemporary childhoods: datafication. Datafication is defined as the increasing accounting for. and reduction of, information about children (such as then learning) to databases, software and attendant surveillance practices, and the increasing reliance upon such data for dealing with (usually teaching) children (Finn, 2016). The second is to indicate how analyses of datafication could be supplemented with and connected to theorisations of matter and social difference. Specifically, I want to develop predominantly Foucault-inspired analyses of power, visibility and surveillance that theorise “how data can reproduce the child as a ‘data double’, shipped of all complexity, and how this dividual can be subject to visibility rather than the person” (Bradbury, 2019: 10). In part, as Gulson and Webb (2017: 9) argue, this means critiquing the

transformation of biology to a data-driven and computing-based field has seen the entangling of computing power, data and new kinds of technical infrastructures with a focus on statistics and sequencing. These entanglements are part of ‘bioinfomiatics’, a term used interchangeably with ‘computational biology’ and ‘systems biology’.

Yet, as I have already argued - through the plural notion of ‘interface’ - these infrastructures are not merely technical (in the digital sense) and nor are they confined to questions of the biological and to the geological (Gallagher, 2019; see Chapter 4) but to, as I will show in this chapter and others, the banality of everyday childhood objects like toys.

The rise of Al products - especially in education - has been seen as an intensification of efforts to ‘personalise’ learning in (particularly) neoliberal contexts (Cukurova et al., 2019; also Pykett. 2009; Perrotta and Williamson, 2018). Boiled down, Al can provide real-time support and recommendations for leanring content, speed and method, at the same time as - recursively - collecting data about children, both individually and collectively It can also - as will become clear - offer different interfaces for learning although, as Thompson and Sellars (2018) argue, the status of these technologies as comprising an event that will ‘rupture’ or ‘revolutionise’ learning is rather less certain. As part of the Energy Beyond Technology project, we developed systematic reviews of academic literature, (social) media coverage and commercial websites dedicated to datafication, Al, loT and algorithms. I focus here on the features of a subset of these technologies, in order to continue developing a conceptual framework around the circulation of stuff about childhoods that is attuned to both the digitality and materiality of these technologies.

As indicated above and in the growing literatures on datafication and Al, the education sector is seeing perhaps the most concentrated growth in digital technologies aimed either at children themselves or at generating data to support teachers in their work with children (Kumar et al., 2019; Gulson and Sellar, 2019). Many of these technologies are produced by commercial companies, prompting a range of fears and concerns about the ultimate uses of the data produced (see below) and “a redistribution of agency across socio-technical networks” that involves a range of ‘new’ actors (Williamson, 2017; Jarke and Breiter, 2019. A couple of examples will suffice.

Firstly, Kidaptive Adaptive Learning Platform (ALP) is a learning platform service for other educational companies to analyse the data produced by teachers and pupils using the platform in order to ‘improve learner engagement and outcomes’ (xTOw.kidaptive.com/). It is a cloud-hosted Big Data platform which, Kidaptive argues, makes it scalable and secure. Using a bespoke machine-learning algorithm it can be used to create personalised learning paths that are communicated to children and parents; identify learning-relevant behaviours like guessing, focusing and skipping; set appropriate challenges; track skills learnt by learner during game-play as evidence for educators; and help test preparation tluough identifying gaps in learning and through personalised learning. The ALP works tluough three tiers: Analytics (‘‘One-way event sending to ALP, which provides analytics based on event data [. .. such as] engagement and retention”); Insights (“metrics, ability insights and insights [. . . with ability to] predict test performance, estimate time on task, and more”); and Adaptivity (“support for real-túne client-side adaptivity [. . . using] a Bayesian-IRT psychometric framework to determine the opthnal next challenge”) (www.kidaptive.com/). There are many other similar platforms, sttch as ALEKS (Assessment and LEaming in Knowledge Spaces). This is an Al assessment and learning system to determine quickly and accurately knowledge of students. ALEKS periodically assesses the student as they progress through a course, providing instruction and dhected assessment to ensure learning is retained. It is available for maths, business, science and behavioural science subjects in America (uww.aleks.coni/about_aleks).

One of the most common responses to Al and the mcreasingly pervasive ‘atmospheres of progress’ (Finn, 2016) that ensue in schools has been to outline concerns not only with the rise of commercial companies as providers of platforms but the security and ultimate use of data circulating about children. Yet there are many other possibilities for explaining and conceptualising these kinds of digital technologies. On the one hand, there is a need to theorise the positioning of childhoods - and children - within an increasingly evident ‘platform society’ (Van Dijck et al., 2018). “The term refers to a society in which social and economic traffic is increasingly channelled by an (overwhelmingly corporate) global online platform ecosystem that is driven by platforms and fuelled by data” (Van Dijck et al., 2018: 4). A platform is a digital structure in which users can interact and where data about users can be collected and afforded financial value (e.g. Uber or ALEKS); a platform ecology refers to the system of platforms that predominate in any context - in countries like the United Kingdom and the United States, these include Amazon, Google, Uber, Facebook and so on.

Kumar et al. (2019) offer one of the first systematic conceptualisations of platforms that can help us understand the positioning of children within digital media such as Kidaptive and ALEKS. They argue that platform systems encourage teachers to become ‘surveillant consumers’ of digital technologies and orient them “to see student data as interchangeable with students” (Kumar et al., 2019: 1). Once again, then, platforms are dedicated to visibility, but, once again, rather than making-visible students’ own experiences or voices, these systems move beyond voice and agency towards the constitution of datasets about individual children or groups (also Kraftl, 2013a). Those datasets can then take on their own lives, both figuratively and literally far from the children with whom they originated. The vast majority of such datasets remains obscured (invisible) but is then made visible in partial ways to both teachers and students, through the carefully designed interfaces of any particular platform. Thus, not only do data selves correspond poorly with ‘real’ selves (Livingstone, 2019) bitt children themselves only ever see partial and heavily policed versions o/those data selves. As Finn (2016) and Kumar et al. (2019) argue, these forms of representative have significant effects and affects: some learners may be pleased with their data; others anxious or scared; others will ignore it. But now that data accompany children throughout then- whole childhood, into adulthood, it is becoming increasingly possible that those data will govern the opportunities (and limitations) presented to any child as much as any other considerations (Bradbury. 2019).

Kumar et al.’s (2019) analysis is particularly helpfill for the ways in which it offers additional tools for thinking and doing, after childhood. My reading of their work (and I stress that this is my interpretation) is that it offers a conceptual lexicon for thinking more about platforms in education, as follows. Data are allconsuming-. they chase teachers, even as they sleep, pressing for a response. This affords, perhaps more than anything, a sense in which data are out of the total control of not only teachers but also the technology companies that create them. This is because the data are so complex, so massive (even for just one school) and because data are aggregated and reported back through machine-learning that can take place with little measure of human intervention. Meanwhile, where lack of information is now seen as a threat, platforms can - because of their interoperability across multiple devices and places - offer possibilities for ‘eliminating friction and flattening students into data’ (Kumar et al., 2019: 6, my emphases). This argument is, then, not only about the difference between real and digital selves but the flattening of students into ‘one-dimensional units within a uniform interface’ (ibid.: 7). Thus, the very vibrations of the interface (Ash et al., 2018) mean that, increasingly, teachers run the risk of confusing their knowledge of data with knowledge of then students. However, “[t[he platformization of the classroom does not cause teachers to monitor then' students. Rather, it reduces the friction involved in monitoring and frames monitoring as an attractive method through which teachers can perform then duties” (ibid.: 7; my emphasis). In other words, neither teachers nor students are cultural dupes - and as I know from my experience of working in many schools that use many different methods to collect and deploy data - neither do they not know their students, beyond then data. However, the platforms, then interfaces and wider professional and societal expectations and nonns around the ‘threat’ of data-lack mean that they offer a plethora of attractions for teachers. Thus, platforms are all-consuming, out of (total) control of humans, may eliminate friction and flatten students, and may seduce and attract teachers (and others) through their interfaces.

Certainly, of the above concepts, the central idea that digital technologies and data are out of control is one that has gained widespread attention. Recent cases involving networked objects - most commonly referred to as the Internet of Things (loT) - have raised many critiques and fears about the rights of consumers and the relative control that we all have over our data (Livingstone and Third, 2017). Some of the most high-profile examples have involved products for children. At the time of writing, the most prominent centred on Amazon’s Alexa Echo Dot, which had been collecting voice recordings from hundreds of thousands of children, despite a lack of clarity about consent for doing so (BBC, 2019a). Whilst the recordings can be deleted by parents, it is likely that many will not have been -not least because of the tune involved and because Amazon uses the recordings to personalise its services.

Although in the case of the Echo Dot there have not been any direct concerns over security, other examples highlight that the utopia of‘frictionless’ data (Kumar et al., 2019) is one that requires considerable scrutiny, and especially where the technological elements of loT are becoming increasingly invisible (Plowman, 2019). For instance, of many similar cases, a company specialising in internet-connected teddy bears (CloudPets) leaked over 800,000 user credentials in 2017-exposing the problems of hacking and smart toys (Franceschi-Bicchierai, 2017). CloudPets are smart toys which enable families to record and leave voice messages on a teddy bear, uploaded via bluetooth. Spiral Toys company left the database of credentials without firewall or password protection, exposing data which was then hacked and held to ransom (ibid.). Although the voice messages were on the exposed database, other researchers have found the messages were also stored without authentication (ibid.) and therefore easily accessible for hackers. These toys have since been bamied in several countries, including Germany, and removed from platforms such as Amazon and other prominent online selling platforms (BBC, 2018b).

Thus, a corollary of the concepts of visibility, control and friction - and sitting in the relation between them - is the ever-present threat of exposure and the potential trauma that that exposure might cause (Chapter 9). I am not sufficiently well versed in digital security to comment further; rather, and thinking again of modes of thinking and doing, after childhood, I would note again that all of these concepts are components of the broader massif cation of data and hence of the ways in which childhoods circulate. Arguably, the voice recordings held by Amazon are - even if they are left unmined - somewhat akin to the Mass Observation Archive, which collected and collated writings and other media by ‘ordinary’ British citizens between the late 1930s and the early 1950s (www.massobs.org. uk) and again from 1981 onwards (including an everyday childhoods collection from 2013 to 2014, held at www.niassobs.org.uk/about/news/192-the-everyday-childhoods-collection-is-iiow-oii-figshare). However, there are also obvious differences, not least in terms of the levels of consent obtained for the latter and the huge size of the former, which is already many orders of magnitude larger in terms of the sheer number of children involved and the details held about them. But what is also significant is the fact that although human operatives are listening to a very small proportion of Amazon’s recordings, most are - like other data collected by Al - simply either deposited in vast digital reserves or analysed through automated algorithms, enabling ever-more sophisticated forms of machinelearning, ostensibly without humans.

These databases are, then, empirical examples par excellence of one version of the idea of thinking and doing, ‘after childhood’. Not only are children still a concern (as data-producers, potential victims and paying consumers), but children are also, and simultaneously, decentred as their data are literally and metaphorically displaced from their lives, from then view and from their control (indeed, perhaps the control of any humans). Furthermore, a little like the toys discussed in Chapter 6 - although in some ways not all like them - these recordings could exceed children’s childhoods and, indeed, their own lifetimes and could be taken not only as a proxy for an entire generation of childhoods but constitute in themselves ‘generations’ of a sort (generations of childhood-data), with temporalities that interweave and interface with human generations but also exceed them. One question is, then - amongst many others - whether the massification of these recordings and the many other kinds of data produced through Al and loT constitute hyperobjects; and/or, indeed, whether, as a result, it might be possible and usefill to think of childhoods, circulating as hyperobjects - particularly, but not only, in their datafied form.

Looping back to the feminist, queer and critical race theories I introduced in Chapter 2, and with an eye to forms of physical, behavioural and emotional difference that I discuss in Chapter 7, a final set of conceptual considerations for Al and loT is ethical and political and focused around questions of inclusion. hi light of concerns about visibility, control, friction and exposure, several platforms and technologies now purport to offer users - and specifically children -opportunities to wrest back some measure of control, particularly by learning more about the workings of these media. Teens in Al is an organisation launched at the Al for Good Global Summit at the UN in May 2018 to support young people aged 12-18 to explore Al, machine-learning and data science (https://teensinai.

corn/#about_us). The organisation hosts ‘bootcamps’, ‘hackathons’, workshops and talks and other events for young people to learn how to work with Al. Summer programmes take place in London but are very expensive (a ten-day course costs £2,150 and you need to bring a computer), and while folly funded places exist, the programme remains inaccessible to many young people.

The exclusivity and inaccessibility in terms of young people being able to access codes is increasingly recognised as something impacting the algorithms and Al being developed (because particular, human-inputted codes and values underlie at least the initial development of machine-learning). As the technology is created and designed by a homogeneous group of people, the resulting technologies are biased towards then' (likely) homogeneous needs, propping up serious concerns that Al and other digital technologies could reinforce economic or social divides and, in particular, the opportunities opened out to children and young people during then lifetimes (Gulson and Webb, 2017; Bradbury, 2019). In response, AI4A11 (http://ai-4-all.org/) is an organisation which amis to provide mentorship and education so that underrepresented young people in North America can learn Al (ibid.). Underrepresented groups (such as girls, low-income students and ‘youth of color’ [Johnson, 2019]) are mentored through training programmes hosted at universities across the United States. The success or otherwise of these schemes has yet to be determined.

In a very different set of circumstances, there have been moves to design robots to support children with autism spectrum disorders to better interact and learn social skills from their teachers and therapists. The University of Luxembourg’s QTrobot is used within therapy sessions to decrease discomfort for those with autism and increase a willingness to interact (Waltz, 2018). The robot is described as a ‘cute tech-based intermediary’ (Waltz, 2018), designed to teach those with autism spectrum disorders to interact more comfortably. A video released by the lab working on the robots shows a boy being instructed by the robot to place objects in certain arrangements and congratulating the boy when he completes the task; the boy’s gaze is focused on the robot. During the research, the designers found that children directed their gaze towards the robot twice as long as towards the human; behaviours which commonly signify distress or anxiety occurred three times less with the robot than with sessions with the human.

Once again questions of interface come to the fore here, hi this case, however, the interface between children with autism spectrum and the robot raises two distinct yet potentially troubling issues that require further consideration. First, in terms of the promise of ‘inclusion’, the concern here is - as with any of the Al technologies cited above - inclusion in what? Do the robots and other technologies simply promise the presence - and visibility - of an increased range of consumers, who as the Al technologies become ever-better-attuned to the nuances of diverse human experiences can be incorporated into logics of commercialisation and datafrcation (Thompson and Sellar, 2018)? Recursively, a range of questions could be raised about the intent lying behind efforts to use technologies that might attempt to normalise, channel or control nemo-atypical humans into what may be actually fairly conventional educational and societal norms. Of comse, the ability of the robot to make a child feel comfortable or communicate could be a good thing; yet only where ‘comfort’ and forms of emotional literacy are not taken as a proxy for the subjection of children with behavioural, emotional or social differences to power-laden educational practices that may - alongside processes of datafication - mark out their capacities and life-chances as a priori ‘different’ (Gagen, 2015). Second, the presence of robots - as with other loT technologies -signals the importance of materialities that look like ‘objects’, even as in other aspects of technological development the importance of objects may recede (Plowman, 2019). In other words, the inclusive capacities of QTrobot - not least its ‘cuteness' - are afforded by its taking the form of a robot, rather than a human. The interaction between a screen and a robot may - as a result of the connectedness of both technologies - bear certain similarities that can be appreciated via Ash et al.’s conceptualisation of unit/vibration/tone. Yet robots are not smartphones or tablets, and the very particular kinds of (partial) encounter with robots-as-objects serve as a reminder of the proliferation of material forms that digital technologies can take as they become (part of) media-machines (Bryant, 2014).