Caring for robots: How care comes to matter in human-machine interfacing

Abstract

Care robots promise to assist older people in an ageing society. This article investigates the socio-material conditions of care with robots by focusing on the usually invisible practices of human-machine interfacing. I define human-machine interfacing as the activities by roboticists and others to render interaction between robots and people possible in the first place. This includes, efforts to render prototypical arrangements of care ‘robot-friendly’. In my video-assisted ethnography of human-robot interaction (HRI) experiments. I identify four types of interfacing practices, where care comes to matter: integrating the ephemeral entity that is ‘a robot’, helping it by way of mundane courtesies, making users ‘fit’ for interacting with it, and establishing corridors of interaction between the robot and people’s bodies. I show that robots do not so much care for (older) people but rather, the other way around – people need to care for robots. Hence, care robots are not simply agents of care but also objects of care, rendering necessary a symmetrical analysis of human-machine interfacing. Furthermore, these practices do not merely reflect the prototypical state of the art in robotics. Rather, they indicate a more general mode of how robots and people interface. I argue that care with robots requires us to re-consider the exclusive focus on the human and at least complement it with care for the non-human and, incidentally, the robotic, too.

Keywords

robotics care interfacing human-robot interaction intra-action milieu

In his art project ‘With robots’, Diego Trujillo Pisanty explores the question of what living with robots might look like (see Figures 1 and 2). He departs from the premise that for robots the human world is a hostile place. What appears orderly to us is nothing less than chaos for robots. Changing lighting conditions, soft objects or unpredictable human behaviour make it hard for robots to work on their own under these circumstances. While such difficulties are usually brushed aside as a merely preliminary problem, Trujillo Pisanty poses the intriguing question of ‘how our homes and objects might change in order to accommodate the needs of robots’ (Trujillo Pisanty, 2011). This focus on ‘robotic needs’ stands in stark contrast to common visions of robotics. While machines are often portrayed as autonomous (Matsuzaki and Lindemann, 2016) and intelligent (Halpern et al., 2017), the images above suggest otherwise. In fact, robots need ‘robot-friendly’ environments and, as I will show, constant human support.

Figure 1.

Plate design with cut-out and machine readable markings on table (Trujillo Pisanty, 2011).

Figure 2.

Cup designs with stubs and machine readable markings (Trujillo Pisanty, 2011).

Trujillo Pisanty’s art provides me with an entry point for analysing human-robot interaction (HRI) experiments in a European care robotics project. I will show how a prototypical home environment needs to be adapted to robots in order to demonstrate assistive services for older people. This focus is informed by an analysis of human-machine interfacing. I am not primarily interested in the interactions between robots and human users, but rather in the manifold efforts by which roboticists and others interface robots, people, everyday objects and their surroundings. These often-mundane practices enable phenomena of interaction in the first place. Focusing on such efforts, I argue that care comes to matter in a specific form or, rather, with a specific target: the robot itself. The robot comes to the fore not as a self-acting, ‘autonomous’ agent of care, but rather as a distributed, intricate and fragile object of care. Robots do not so much care for people as people have to care for robots.

Amplifying care in human-machine interfacing provides a critical entry point to the dominant narratives of robots taking over care work. It shows the additional labour that needs to be invested in installing and maintaining care with robots. It reminds us of the omnipresence of care practices in technoscience (Puig de la Bellacasa, 2011). Such a perspective allows us to account for the increasingly intimate relationship between humans and ‘intelligent’ machines beyond narratives of technological determinism and anthropocentrism. It works to question overly optimistic as well as pessimistic views regarding the possibility of robots caring for older people. Here, in line with Trujillo Pisanty’s speculative proposition, I argue that practices of caring for robots do not merely reflect an oddity of the technology’s prototypes. Rather, they indicate a more general mode of how robots and people interface. Carving out the intricacies of how people and their surroundings need to adapt to robotic machines will be crucial in taking stock of the costs or at least the additional labour associated with an increasing robotization of everyday life.

Scholars have shown how certain ideas about care, ageing and older people have been inscribed into robotic technology. Such inscriptions take effect when implemented within socio-technical arrangements of care. I propose a symmetrical analysis of human-machine interfacing as a key pathway to enquire into the reciprocal ways in which humans and machines are rendered available for one another. Such a framework further develops STS scholars’ recent interest in care directed at technological objects, which will provide the focus of this article’s analysis. I present four different types of interfacing practices: integrating robot(ic)s, rendering the apartment robot-friendly, fitting users and calibrating corridors of interaction. An analysis of human-machine interfacing not only sheds light on roboticists’ practices in prototypical home environments but also might serve as a probe for investigating emerging entanglements between humans and machines more generally.

‘I don’t know whether this should be recorded yet.’ Amplifying practices of human-machine interfacing in a European care robotics project

The analysis of this article is based on ethnographic data collected during two field visits to a three-year European care robotics project at the end of the project’s timeframe. The project was funded under the European Commission’s (EC) Seventh Framework Programme and ran from 2012 until 2015. Its aim was to develop and demonstrate prototypes at two pilot sites, one in Northern and another in Southern Europe. To preserve anonymity, I omit details, and all names of humans and non-humans referred to have been replaced with fictitious names.

The project promised to develop and implement three different robot models able to provide both domestic help and services outside of the home, such as reminding the user of their medication or doing their shopping. The crucial selling point of the project was to offer an integrated solution to assisted living by deploying multiple robots together. During my research, however, researchers tested only one platform, which was restricted to home use – a platform in robotics designates a singular system of hardware on which a robot’s software and operating system is running.

I cannot go into detail on the robot’s appearance, but it involved a service robot with human-like features. The robot had a recognizable head and torso, and the head featured eye-like lights that would blink in different colours indicating different states of the system (e.g. ready to interact or not ready to interact). It also had a ‘grappler’, a mechanical arm that allowed the robot to pick up things and carry them to the user. There were two main interfaces to communicate with the robot: It had a tablet fastened to its side, which people could pick up and use to assign certain tasks to the robot, and it was equipped with a speech interface.

The project’s consortium consisted of twelve partners, including: the operators of the two test facilities, a municipality, a public assisted-living provider, a service robotics company, researchers from engineering and computer science, a gerontological partner and a company that specialized in User Experience (UX). The last of these facilitated my field access. This also restricted my access, since the company was only involved in the setup and implementation of the tests in the Northern facility. I actively participated in the project by taking over some of the tasks with which the UX company was charged. Among these tasks was the monitoring and briefing of the HRI experiments during the so-called ‘pre-tests’. These consisted of a series of test runs over a period of two weeks, where the researchers had to integrate the robot’s different modules on-site. I will go into more detail on this below, but the main objective of such pre-tests was to get an integrated robot running with human users.

This is my entry point to the project, which, at the time, was in its so-called ‘second experimental’ loop, when the development work of the previous two years would be put to the test in a homelike test apartment. This facility was situated within a care home and was advertised as ‘one of the world’s largest and most homelike test beds’ (test apartment website). On its website, the apartment is positioned as a speedy passage point for innovation in healthcare settings and promises to bring together innovative developers and prospective users. The apartment was operated by an innovation network that included the local university and the public housing company that ran the care facility. The network also received funding from the European Union and featured numerous regional and national companies involved in healthcare technology. The apartment was furnished to emulate a homelike environment. At the same time, it was a host for experiments involving researchers, technicians and their devices. Roughly, the apartment consisted of four parts relevant for the tests: a large living room in the centre, a bedroom, a kitchen and a control room, from where researchers would monitor and, if needed, intervene into the robot system during the experiments. Most of the test runs’ action took place in the living room, where the user was seated in always the same chair.

Practices of human-machine interfacing distribute across the apartment and among different actors. As my research was to register how roboticists and others would prepare, recap and re-adjust the experiment along the way, I opted for a video-assisted ethnographic approach. I draw on field notes, interview transcripts, project documents and video feed. I set up cameras in the apartment to record roboticists’ activities, for example, to see what was happening in the control room during test runs. While video is well established as a form of ethnographic research to understand socio-material practice (Pink, 2010; Suchman and Trigg, 1991), I used it in this study as an auxiliary tool to accompany and reflect on my ethnographic observations. Against the lure of video data as representing the world ‘out there’, I follow Law (2004) in arguing that any method plays an active part in data’s production. In my case, video data allowed me to register and amplify the distributed nature of human-machine interfacing. Interestingly, this clashed with some of the assumptions of my fellow researchers in the field.

To understand this it is important to point out the ‘vernacular’ role of video in robotics (Tuma, 2019). Video is a crucial tool for observing and evaluating HRI (Veling and McGinn, 2021). For this, the test apartment featured an extensive camera setup in each room, directly streaming to the control room. However, this camera setup was specifically geared towards observing the interactions between the robot system and the test subjects. My own setup of cameras differed crucially from this in that it also recorded when experiments were not running and in places where no HRI took place (e.g. the control room). Despite approval through consent forms, one researcher had reservations about this. After explaining to him that I am interested in the setup of the HRI experiments, he replied ‘I don’t know whether this should be recorded yet’. In the mind of the roboticist, there was nothing to see outside of testing and beyond the interaction between robot and test subject. This points to the epistemic politics of video in robotics, which is mostly used to communicate results to colleagues or the outside world (Bischof, 2017: 257ff). Of course, the fact that I used video differently does not mean that I am able to look behind the veil of roboticists’ staging efforts, showing things as they ‘really’ are. Rather, it means that the invisibility of human-machine interfacing is embedded into the epistemic structures and practices of robotics (Alač, 2009), which makes it all the more worthwhile to amplify and examine them more closely.

The politics of care robots and its critique

In the context of European research and development funding, robotics in general and care robots in particular have become prominent topics (European Commission [EC], 2021). Here, the project of using robots in elder care is infused with a number of dominant narratives and assumptions. Most importantly, service robots for elder care have become tightly connected to the concern of demographic ageing and an ostensible nursing crisis (Maibaum et al., 2021). It is hoped that care robots will provide or at least assist in the care of a growing older population. They are expected to help people in activities of daily living like shopping or communication with relatives, which should allow them to remain in their homes for longer and prevent ‘unnecessary hospitalization’ (European Innovation Partnership on Active and Healthy Ageing [EIPAHA], 2012: 4) thus saving on public spending and rendering the provision of care more efficient (EC, 2009: 73). It is this concurrence between economic arguments and the ostensible wish of older people to live at home that legitimizes many autonomy-enabling innovation projects in recent years (López Gómez, 2015; Neven, 2015), including care robots.

The discourse on care robots heavily depends on assumptions around these machines’ autonomy. Traditionally, robots have been used in the automation of industrial production (Fleck et al., 1990). The autonomy of such machines was only possible because their operating environments were highly controlled, that is, robots operated in factory cages separated from human workers. Such machines were precise and fast but could not deal with the slightest changes in their environment. By contrast, care robots need to operate in close proximity to humans and in much less controlled environments, such as a home. Care robots are expected to ‘adapt to people’s needs and not the other way round’ (Heeren, 2013: 5). This adaptability is a crucial component in how a discipline that used to develop industrial machines now positions itself with respect to elder care; robotics is assumed to be a ‘universal tool’ (Bischof, 2017: 162; my translation) that can be readily ‘plugged into’ different contexts. Finally, care robots embody the silent helper that should be at all times available yet invisible (Atanasoski and Vora, 2019; Suchman, 2007: 217ff; Treusch, 2015).

STS scholars have critically engaged with these narratives. They have traced the genealogical trajectories and politics of care robots, in which demographic change has been (re-)framed as an opportunity for entrepreneurs and technologists. Care robots ought not only ‘solve’ an impending nursing crisis but also create new jobs and stimulate the economy (Maibaum et al., 2021; Neven and Peine, 2017). These works point to the political transformations, especially in the context of European innovation policy, that have helped this discourse to thrive. They take issue with the claim that robots would have direct effects on society, such as rendering care more efficient and less costly (Neven and Leeson, 2015; Šabanović, 2010). By contrast, a growing body of work has investigated the situated, socio-material ways in which robots and people interact, both in care practice (Pfadenhauer and Dukat, 2015) and the laboratory (Alač, 2009). Studies in this vein show the additional human labour by roboticists and caregivers that render care robots both operable and ‘social’ in the first place.

Scholars have investigated configurations of care, ageing and older people in robotics. For instance, Neven (2011) has shown how robotics assumes older people to be technologically illiterate and reluctant about digital technology. Such stereotypes, he argues, may lead to resistance by older people towards robots, since they equate old age with frailty and deficiency (Neven, 2015). Furthermore, roboticists assume care to be fragmented, that is, that it consists of distinct tasks that can be automated individually and integrated by robotic technology (Vallès-Peris and Domènech, 2020: 165). This is based on a capability approach in robotics, which matches particular activities of daily living with certain engineerable tasks that robots are able to perform (Lipp, 2019: 107ff). Additionally, elder care is taken as an attractive testbed for piloting robots ‘in the wild’ (Šabanović et al., 2006). However, configuring care as ‘wild’ presumes predictability as a positive value (Vallès-Peris and Domènech, 2020: 168), which means that care practices should be rendered rational and calculable. Dualisms between rational and irrational, human and machine, care and technology also surface within care discourses on robots. For instance, Pols and Moser (2009) have criticized the distinction between ‘cold’ technology and ‘warm’ care common in ethical debates about robotics. They argue that, in fact, the entanglements between machines and humans cut across such dualisms (see also Lapum et al., 2012). For example, (robot) technology itself can be the mediator of care, as Pols (2012) shows in her study on tele-care devices.

Finally, scholars have pointed to the reconfiguration and installation of care arrangements as a result of digital care technologies. Here, López Gómez (2015) shows that older people’s autonomy living with tele-care devices is not merely an effect of technology but rather denotes an attribute of fragile, socio-technical living arrangements. He argues that such arrangements are in constant need of care in order to ensure and improve autonomy. Furthermore, Neven (2015) illustrates the profound reconfigurations that can result from deploying autonomy-enabling technology in the homes of older people. He argues that the assumed wish of older people to live at home renders invisible actual, possibly undesirable changes in their living environment. Finally, Sánchez-Criado et al. (2014) connect literature on user configuration and care arrangements by investigating installation practices of tele-care devices in older people’s home. They point to the ‘precarious infrastructure of usership’ (Sánchez-Criado et al., 2014: 712) that comes with the installation of tele-care devices. They show that installation practices are another important arena where people, machines and arrangements of care become entangled with one another.

From interaction to interfacing

To this body of literature, I contribute an analysis of human-machine interfacing, which focuses on the ways by which people, machines and their shared environments are rendered available for one another. I will develop such a framework with the help of Barad’s notion of intra-action and Simondon’s concept of the associated milieu (Lipp, 2022). Such a framework seeks to shift attention from interactions to the coordinative efforts of roboticists and others in engendering phenomena of HRI.

I start off my theoretical deliberation by way of a vignette. It shows the mundane mishaps and repairs involved when user and robot attempt to interact. The speech interface in this case comprised a microphone held by or fastened to the user, a laptop connected to the robot system, and speech recognition software running on that laptop. In the course of the tests, interacting via speech proved to be a highly precarious endeavour, where all kinds of bugs, mishaps and repeated breakdowns occurred. They were met by roboticists’ efforts to constantly reconfigure the user’s body and the machine’s speech apparatus in order to somehow make them interact.

Getting ‘speech’ right

The experiments are about to begin when a problem with the speech interface comes up. The robot system does not seem to react properly to the commands of the user. One of the roboticists addresses the test person: ‘Could you avoid holding the microphone too close?’ He prompts her to fasten the microphone to her pullover’s collar. That way the distance between the test person’s mouth and the microphone would remain constant. After a few more attempts to initiate the sequence, another roboticist comes from the control room and asks the test person to speak slower into the microphone. The problem persisted. This was strange since the speech interface had just worked when the user had tested it in the control room. After the tests, the roboticists explain to me that the microphone was simply connected to the wrong laptop. The reason why it worked earlier was that the laptop with the speech software on it stood next to it and recorded the voice via its own built-in microphone.

Even the most trivial dis-/connections can lead to its breakdown. This is often met with considerable frustration by roboticists who repeatedly fail to make interaction happen: ‘How are we supposed to evaluate human-robot interaction when the robot does not interact?’ one of them asks. While we have grown accustomed to use the term ‘interaction’ to designate relations between digital technology and humans (Suchman, 2007: 34), these experiments show that the conditions for such phenomena are located neither within ‘the human’ nor inside ‘the robot’ and its interfaces (Alač et al., 2011). Rather, a myriad of often seemingly trivial entities like a USB port, a laptop or an inconspicuous microphone decide the fate of whether interaction between two entities happens or not. As a result, the entities only take shape in relation to one another as either recalcitrant objects or bodies to be disciplined. Hence, to borrow from Barad (2003: 815), we are not talking about interaction but rather about the intra-active becoming of these elements. They are emerging in the constant process of bringing-into-relation. In other words, it is through practices of human-machine interfacing that robots and human beings, along with mundane objects and surrounding participants, come into being as interacting (or not).

This has important consequences for the development of the notion of human-machine interfacing. Barad’s approach helps shift attention away from ‘the human’, ‘the machine’ and ‘the interaction between them’ and towards the socio-material practices that produce those relata as interacting (or not) in the first place. Hence, the task at hand, to register and amplify the (human) care work necessary to enable relations between care robots and their environment, benefits from such a radically relational view that takes human and machine not as the source but as the mutually constituted, fragile products of their entanglement. In effect, such an approach runs counter to a traditional view in STS that sees non-human elements such as technologies or material objects as foremost a stabilising force rendering particular social or symbolic relations more durable (Latour, 1991). By contrast, the case of human-robot interaction experiments illustrates the unpredictable potentiality and persistent fragility of materializations (Lipp, 2017). Theorizing the notion of interfacing in such a way means acknowledging the open-endedness of techno-scientific practices. It means that they

iteratively reconfigure what is possible and what is impossible – possibilities do not sit still. … Possibilities aren’t narrowed in their realization; new possibilities open up as others that might have been possible are now excluded: possibilities are reconfigured and reconfiguring. (Barad, 2007: 234)

For Barad, materiality is not what stabilizes the world, but is what channels its ongoing materialization. It supplies the world with noise and unforeseen variability. The question of how and which entities will inter- or disconnect is never fully determined. Rather, the analysis can only observe certain continuities in between events. Along those lines, I do not think about interfacing as the primordial source of stability vis-á-vis a particular interconnection but rather as a practice ‘in its intra-active becoming – not a thing, but a doing, a congealing of agency’ (Barad, 2003: 822). Hence, ‘human-robot interaction’ is but a temporary snapshot in the continuous efforts of human-machine interfacing. It designates a persistent call to action – for roboticists but also for users and other participants – to continue interfacing.

Barad inspires an analysis of human-machine interfacing in that she proposes a radically relational and performative approach that allows us to challenge and look past the dualistic notion of human-and-machine interaction. Another important step in this regard is to de-centre our attention towards the distributed conditions that enable interaction between humans and machines to come into being in the first place. For this I draw on Simondon’s (2017) philosophy of open objects. For him, to construct an object, for example a robot, relies on the continuous effort of rendering those disparate elements available for one another within particular associated milieus.

Having become detachable, the technical object can be grouped with other technical objects according to such or such setup [montage]: The technical world offers an indefinite availability of groupings and connections. For what takes place is a liberation of the human reality that is crystallized in the technical object; to construct a technical object is to prepare an availability. (Simondon, 2017: 251)

With this, Simondon (2017: 248ff) argues for a detachability of the technical world, which exhibits a logic of its own and thus requires particular attention. Being detachable, technical objects become available for their recombination in the context of particular milieus. Hence, to construct a technical object means to install interconnections between objects, that is, ‘to prepare an availability’ with respect to each other (Simondon, 2017: 251). It is this detachability of elements, their separating from other elements, as well as their availability for recombination, their interconnecting with other elements, that characterizes the technical operation (Simondon, 2009). This notion of technicity should not be taken as techno-essentialism. Simondon is not after an ontology that separates technology from everything else. Rather, he argues that in order to properly understand technology, one needs to attend to its specific, procedural logic within socio-technical milieus instead of merely relegating it to human action (Simondon, 2017: 255). It is in this sense that I understand ‘robotic needs’ not as an essential quality, but as something that becomes apparent through the continuous process of rendering-available-for-one-another within a particular milieu.

With regard to an analysis of human-machine interfacing, Barad and Simondon teach us to attend to the manifold, performative and distributed ways in which phenomena of interaction are continuously reconfigured and reconfiguring. In fact, this connects my project of human-machine interfacing with the central premise of (feminist) technoscience studies, namely, that humans and machines co-constitute one another (although not necessarily in the same way; see Suchman, 2007: 268ff.). Hence, I am not interested in interaction in general, but rather in how the relata of interaction are produced and re-worked within a more or less controlled milieu. What a given relata is, what agency it may attain is only resolved within such a milieu, a complex space of intra-active interdependency. An analysis of human-machine interfacing, as I propose it here, is able to grasp the reciprocal, performative ways in which robots, people and their environments are continuously being re-worked to become available for one another. It thus focuses attention on the coordinative efforts by roboticists and others to produce and calibrate those relata. In this way, it is open to capturing both how, as a result of human-machine interfacing, robots might be able to foster caring relations with humans but also, conversely, how they require extensive attention themselves in order to be able to do so.

This second aspect will be the focus of the ensuing analysis. In line with recent scholarship in STS, I will specifically focus on how care comes to matter in practices of human-machine interfacing. This interest is in line with recent work in feminist STS that investigates ‘matters of care’ (Puig de la Bellacasa, 2011). Here, the notion of ‘care’ operates as a counterweight to the often objectified and glossy accounts that technoscience gives of itself. It seeks to re-affect that which is often presented as separated and mute. To add to this discussion, I will propose human-machine interfacing as another crucial avenue through which care comes to matter in technoscience.

Caring for robots: How care comes to matter in HRI experiments

I turn now to analyse how roboticists aim to engender phenomena of HRI by attending to their ‘ordinary technical practices’ (Vinck and Blanco, 2003: 2f.). Here, I identify four different sets of such practices: integrating robot(ic)s, rendering the apartment robot-friendly, fitting users and calibrating corridors of interaction. In contrast to the promissory visions of care with robots, robots do not navigate (prototypical) care arrangements autonomously but rather are dependent on a vast range of usually invisible, petty, decidedly human activities, which meticulously bring about a robot-friendly milieu. As a result, the experiments do not leave the involved relata untouched but rather re-work them, so they fit within this milieu.

Precarious demonstrations: Integrating robot(ic)s

As described above, the innovation project subscribes to the task of demonstrating robots in (prototypical) care settings, in this case the test apartment. During the first batch of ‘pre-tests’, the overall focus was less on staging the robots’ performance to outside audiences than on making these machines work at all, that is, performing certain tasks and maintaining interactions with tests subjects.

A thousand pieces

The first pre-test is scheduled for the morning I arrive. The test facility, designed to look like a real home, buzzes with activity. About ten people run around, type on their laptops, give each other advice, and discuss what still needs to be done. The apartment abounds with a myriad of cables, robots, screens, keyboards, computer mice and other equipment. This differs greatly from the pictures I have seen of this apartment online. The pictures there feature happy-looking older people in a tidy, comfortable home. Now, it seems, roboticists and their machines have settled in here for good.

One of them, Francis, a computer science professor at the local university and my contact person ‘on the ground’, approaches me and starts to explain. ‘The robot is a very complex system’. It is especially difficult to get all the different parts to run together. Roosje, one of Francis’s PhD students, seconds that: ‘About a thousand pieces need to work together there!’ Philipp, a master student, describes the resulting work process as follows: ‘You build your module so that it works in a certain way. You have to predict what the other modules are doing. Basically, it’s only the interface that matters, but the modules do not always work as expected’. Everybody comes with functioning components, but the components stop working when they have to work together. It is for these unexpected failures that meetings such as these are so important for the research project. ‘We can have Skype meetings but I prefer these meetings, because everybody is together in one place’, Francis says.

This introduction to the cosmos of robotics might come as a shock: There seems to be no such thing as a robot and no such thing as robotics. Instead, robots consist of components developed by research teams belonging to very different disciplines, epistemic cultures and geographical places (Bischof, 2017; Meister, 2014: 110ff). Usually, it is not the primary interest of computer scientists, engineers or HRI specialists to build an integrated system. Rather, the robot platform is merely a residual product of ‘robotics’ projects serving as an experimental milieu, in which each partner can test their respective ‘modules’. However, in the case of this project, the funder, that is, the European Commission, expects the demonstration of an integrated system, which can perform assistive tasks for older people. The project cannot merely construct ‘basic robotic components’ (EC, 2011: 72) but must test a robot system in care-like environments with users. These divergent expectations not only testify to the conflictual relationship between disciplinary and interdisciplinary research (Barry et al., 2008) but also set in motion precarious demonstrations, where integrating robots means laboriously interfacing a range of different technical components and experts during a local ‘integration week’.

Integrating robots is not a formulaic routine task but is rather highly improvised. This seems to be essential not least because, on this morning, a lot does not work as expected.

Something is terribly wrong

Most of the action takes place between the control room and the living room. Despite an hour’s worth of effort, the robot has not even moved yet. Philipp hastily runs back and forth between the two rooms, tampering with the robot platform or typing something at the main terminal. Casually leaning against a wardrobe, Francis observes the situation. ‘This is typical. If you try to demonstrate, you have hardware problems’. He explains that the system does not recognize the laser sensors or that they do not respond to the system. … It is a ‘low level software’ connection problem, he explains further. The lasers supply the system with information about the environment and this information is essential for navigation.

Suddenly, Roosje curses in the back of the control room: ‘Shit!’ Philipp runs back from one of his journeys to the living room and says, passing by, ‘I have a simple solution to this problem. Two minutes of work, but it doesn’t work’. I notice that Francis, with an absent look downwards, starts tapping nervously against the door of the wardrobe behind him. The robot, still immobile, now begins to utter sounds. I am not sure whether the system reacts to commands given by Philipp at the main terminal. Suddenly, though, Philipp seems to react to the robot’s sounds as he storms past me into the living room, singing with an ironical tone ‘Something is terribly wrong’, then calls out ‘Yes!’ The robot finally starts to move from the living room into the hallway, taking its starting position for the first pre-test that will begin in a moment.

The way that Philipp handles the ‘low-level connection problem’ suggests that he cannot simply apply ready-made knowledge but has to improvise. The promised simple solutions do not seem to work. Instead, Philipp needs to repeatedly hunker down, tamper with the physical components inside the robot platform and sit in front of the main terminal typing requests and rewriting parts of the code. Meanwhile, other problems come up. Components believed to be working suddenly malfunction and roboticists like Roosje and Philipp need to react to them in real time. Even when the problem finally seems fixed, Roosje acknowledges that she does not really know why the robot moved. It could be because Philipp has restarted it. In the end, these practices do not result in a stock of secure knowledge, but rather in a tacit understanding that some problem has been solved for now.

Analysing this through Simondon’s notion of the milieu shows that to construct a robotic object means to install and secure connections between components developed more or less independently from one another. Moreover, despite being configured according to the ‘interface’, they do not fit as expected. In other words, the relata of the relations that make the robot system cannot be presumed as stable entities in practice as they frequently resist being interfaced. This results in meticulous efforts of ad hoc tinkering, in which different relata come to the fore as recalcitrant, agentive, moving parts. The robot’s stability as an object is thus at best preliminary and in need of careful, situated adjustments (Suchman, 2007: 73f.). However, these adjustments do not only pertain to the robot itself but, as I will show, also extend into the milieu of the test-apartment.

Mundane courtesies: Rendering the apartment robot-friendly

Care robots might be more adaptive than their industrial predecessors, but they still need some level of control in the environment in order to properly function. The test apartment is the opposite of that, at least for robots. Mundane objects such as a carpet, changes in lighting or the user’s unpredictable behaviour create ‘hostile’ conditions with which robots have a hard time dealing. Despite the promises of robotics research, roboticists ‘fix’ these problems by way of mundane courtesies towards the robot. So, while care robots are usually portrayed as autonomous beings, they enjoy supporting efforts by roboticists and others, work that is usually removed from narratives about ‘intelligent’ technology (Suchman, 2007: 217).

A good example to illustrate the mundane courtesies of roboticists towards their robots is computer vision. In principle, sensors, together with image recognition software, enable the robot to perceive its environment. For sensing the robot uses a Microsoft Kinect, a camera sensor normally used in the video game industry. The notion of computer vision is actually deceptive however, because a robot does not ‘see’ like a human being. A robot does not simply recognize objects, but rather perceives its environment as a distribution of values attached to pixels in a digital image. From this, it can infer patterns, which are then translated into particular objects via machine learning algorithms as long as lighting conditions remain constant. However, in the case of the test apartment, lighting varied throughout the day, since the rear wall of the living room, where most of the test runs were conducted, was almost completely glazed. This caused the robot’s algorithm to yield completely different results, for example, mistaking something like a shadow for an object. This becomes especially problematic every time the robot ‘manipulates’ its environment, for example, fetches a bottle – a classic example of a ‘fetch-and-carry-task’, where robots ought to bring and hand over a certain object to a particular place or person. In such cases the robot must navigate to a particular place in the living room, a table, recognize the correct object (i.e., distinguish it from a control object), grasp it and bring it back to the user. In order to make it easier for the robot to identify and grasp the bottle, roboticists tidy up and cover parts of the surroundings (see Figures 3 and 4).

Figure 3.

Sideboard during the ‘pre-tests’ in June.

Figure 4.

Sideboard during the ‘realistic tests’ in August.

More of these courtesies were required when, as it often happened, the robot would move its grappler off-target, or would lose the bottle after successfully grabbing it. In such instances roboticists will come to the rescue by, for example, putting the bottle back into the robot’s grappler or displacing the bottle on the table when it is obvious that the robot is going to miss it. As research on social robots has shown, robots need coordinative efforts in order to become ‘social’ (Alač et al., 2011; Neven and Leeson, 2015). The examples presented here show that this coordination extends to mundane aspects of the robot’s environment as well. They are less about managing the attention of users than about coming to the robot’s aid. Mundane courtesy in this context takes the form of caring for an object and its technicity (Simondon, 2009). Especially during the ‘pre-tests’, where practically everyone in the team was present, this happens in plain sight. While such courtesies usually remain invisible during public demonstrations (Shapin, 1989), practices of human-machine interfacing are an acknowledged part of what goes on in the test apartment. The project team reacts to such instances with laughter or, frequently, frustration.

Beyond single objects, these interfacing practices concern the milieu of the apartment as a whole. One example of this is the apartment’s floor. Project members were required to take off their shoes before they entered the premises. Although Francis explained to me that this rule was rooted in the country’s culture where we tested, there was another reason for this. If people wanted to keep their shoes on, they had to put blue plastic bags around them. We had to prevent outside dirt from entering the apartment not because people liked a clean floor, but because it could jam the robot’s mechanics. Hence, this rule was less a cultural token gesture but rather another way of protecting the robot from the messiness of the outside world.

This ‘mess’ can be as trivial as a carpet. In the beginning of the ‘pre-tests’ in June, the apartment featured a number of them. By the time the ‘realistic tests’ were conducted in August they had all disappeared: The edges as well as the meshed surface of the carpets proved to be a problem for the robot’s wheels and, thus, impeded its mobility. Such issues were called ‘friction problems’. For instance, the robot struggled to get onto the carpet in the corridor of the apartment, which was defined as the robot’s start and ‘rest’ position. Every time it navigated to or from that position it took a few accelerations to make it. On the carpet, the robot’s mobility was limited, and it moved much slower. ‘We have to get rid of this carpet. It’s useless!’ a roboticist exclaimed during one of the ‘pre-tests’. This, again, shows the particular perspective of roboticists on supposedly messy environments such as the home. It presumes predictability, a robotic need, as the only target value that matters, as opposed to, for example, the aesthetic value of a carpet for a resident (Vallès-Peris and Domènech, 2020: 168). The robot proves to be much more compatible with the slippery, laminate floor, which this courtesy revealed. Such practices are not restricted to the prototypical arrangements of care robots but can already be observed in other cases where people use robots in their homes. For instance, Forlizzi (2007) shows that users of Roomba, a commercial cleaning robot, would pre-clean the floor so the robot could do its job better. She argues that activities such as these engender social relations between robot and user. Taken together with roboticists’ reactions to the robot’s mishaps described above, human-machine interfacing produces social relations in a post-social society (Knorr-Cetina, 1997). Contrary to the perception that experts uphold a ‘rational’ relationship with their objects, roboticists laugh about, pity or even insult the robot when it gets stuck (see also Voss, 2021). In this way, caring for robots has an affective dimension that goes beyond the intended sociality of humanoid robotics.

Finally, the user’s position within the apartment mattered in making the robot work. During all the tests that I have observed, the user had to sit in a specific chair that had a pressure sensor attached to its bottom. In order to start the test sequence, the user has to ‘wake up’ the robot, that is, utter a particular command (‘Hey, Robo!’). This initiated a sequence where the robot would locate the user and then head for that position. While the concept of the project stipulated locating the user in any position within the apartment, and while this was technically feasible in principle, the user’s position for the tests remained the same at all times. Hence, the experimental situation was strategically simplified in order to allow for a more reliable outcome. This shows that not all of the courtesies that roboticists perform are obvious. Some of them are rather subtle and remain implicit in the course of the experiments. These courtesies never result in a routinized practice or lasting changes to the spatial surroundings. They remain on the level of superficial, quick ‘fixes’ that allow for a mundane ‘laboratization of the social’ (Bischof, 2017: 214ff; my translation).

In all of these examples we witness the common narrative of care robots being turned upside down. Contrary to the claim that robots can be autonomous caregivers, the descriptions above show that these robots are rather fragile beings in constant need of care. By way of mundane courtesies roboticists become part of the action coordinating not only robots and people but also everyday objects in the environment of the test apartment. Contrary to roboticists’ ‘cult of domesticity’ (Vallès-Peris and Domènech, 2020: 165), where supposedly ‘inferior’ tasks of care are taken as available for automation, it is precisely these seemingly mundane tasks of picking up bottles, navigating over a carpet or recognising a person’s position that confront robots with extreme challenges. It reveals the immense complexity that is involved in manual labour, the ‘helping hands’ of care work (Von Bose and Treusch, 2013; my translation). Rendering the test apartment robot-friendly thus does not leave the milieu untouched but rather is a means to subtly adapt it to the needs of robotic machines.

Fit for robots: Selecting, assessing and training users

Users are crucial for interfacing humans and machines. However, as with the moving parts described above, they also cannot be assumed to be stable relata. They cannot interact with the robot just like that but have to be prepared, that is, they have to go through a procedure in which they are selected, assessed and trained. Similar practices have been discussed in the literature as the ‘configuration’ (Woolgar, 1991) or ‘installation’ of users (Sánchez-Criado et al., 2014). While configuration denotes the disciplinary activities of making the user conform to the designer’s script, those activities also rely on an infrastructure of usership, an array of tests and trainings that delimit the flexibility the user has to follow particular instructions. In this case, usership relates to a particular problem: fitting the user to the needs of the fragile robot while at the same time encouraging the user to thoroughly test the robots’ capabilities within given corridors.

This fitness is constructed in a selection procedure taking place prior to the experiments. While the system was designed to assist older people, the ‘pre-tests’ only featured young to middle-aged test subjects (from around mid-twenties to late forties). Younger users were deemed more available and reliable. They would be able to compensate for the fragility of the system. In the pre-tests those users were acquaintances, either colleagues or friends. Similar to an I-methodology (Akrich, 1995) roboticists picked those users most similar to them (young, engineer, academic) in order to safeguard the robot’s operation.

For the so-called ‘realistic’ tests in August, project members had to recruit older people to test the robot. Users were recruited by a representative of the local care facility, where the test apartment was located. The project’s experimental protocol stipulated balanced quotas of these users based on different characteristics, such as gender, age and autonomy. In order to ensure those quotas, the users needed to be assessed by way of different tests. The users’ autonomy was deemed especially crucial here. It was defined as their predisposition with assistive technologies as well as their physical and mental ability to perform activities of daily living. The project distinguishes three levels of autonomy: low, medium and high. According to the experimental protocol, test subjects should be equally distributed among those three categories. This requirement sparked a controversy among researchers, especially with regard to users on the lower end of the autonomy scale. Karl, a sociologist who worked for the UX company, argued that the experiments should represent all possible users, especially those who are severely impaired. After all, the robot could help those people lead a more independent life. Most of the other technical partners, however, opposed this. Francis feared that with such users the experiments ‘will be … extremely more likely to fail, because … the technology was not designed with those users in mind’. On the one hand, this shows the development team’s implicitly imagined prospective user – precisely not the ‘frail’ older person so frequently invoked in ageing-and-innovation discourses (Neven, 2011). On the other hand, it points to the continued work by roboticists to compensate and make up for the deficiencies of the robot itself. It designates a particular form of co-construction of technology and user (Oudshoorn and Pinch, 2005), not so much about selling the product to an audience as protecting the product from that audience.

So, roboticists select and assess potential users on the basis of their ability to interact with the system. This, however, does not yet ensure that users are fully prepared. Before they are admitted to interact with the robot, they need to be trained. This so-called ‘user training’ takes place prior to each test run and can last from 25 to over 40 minutes.

User training

Karl and the test subject sit at the kitchen table. He explains to her the different user interfaces of the robot. She can control the robot via a tablet. Karl takes the device and prompts the test person to tap on its touchscreen. He points out an area on it and explains that she can call the robot by pressing a particular button. She can also control the robot via speech input. ‘If you want, you can try’. He adds: ‘The main command is “Hey, Robo”, because Robo is the name of the robot’. He points out to the test person that most of the answers to questions by the robot are ‘Yes’ and ‘No’. ‘The commands are the same as last week’, he says.

During the user training, the interviewer explains to the user how she should control the system. Here, Karl does not simply present the robot’s functionalities, what the robot can do, but also conveys the prescribed way of using the robot, for example, by saying that most of the robot’s questions require either ‘Yes’ or ‘No’ answers. This is what Woolgar (1991) has described as ‘configuring the user’. Engineers attempt to direct and discipline users in the use of ‘their’ technologies through, for example, user manuals or other types of instructions. However, this only captures half of the interfacing process under operation here. The interviewer also repeatedly tries to take away pressure from the user and to prevent her from thinking that she could do something wrong. For example, in a user training on the next day, Karl tells another test subject that ‘You can try whatever you want. You can’t destroy anything’. Here, the job of the interviewer is not simply to introduce the user to the robot but also, as specified by the protocol, to engender ‘trust in the interviewer and test … [and to minimize] reservations in the handling of the technical devices’. Granting the user freedom in experimenting and getting to know the system is an important part of the user training. Ultimately this is a way to render the system available to the user by at least suggesting that it can, to a certain extent, adapt to the user’s behaviour. Hence, in the context of HRI experiments, human-machine interfacing renders people and machines available for one another by both disciplining and inciting users, sitting ambivalently between roboticists’ preference for control and their aspiration of building a somewhat interactive machine.

This ambivalence becomes even clearer in another controversy about so-called ‘use cards’. A use card contains a brief description of what the user can expect from a service, and a detailed account of its sequential order. The latter part bothered some of the roboticists, since it conveys a very linear structure of the interaction with the robot. While the robot needs a structured dialogue, as we will see, it is also somewhat flexible (e.g., it can recognize different modes of speaking). Some roboticists hold that the latter is not acknowledged enough in the use cards. Another issue is that users needed to hold the printed cards in their hands and often needed to sort through them before choosing a particular service. Roboticists feared that this would distract the user from the robot and thus inhibit indeterminate interaction. In the end, the use cards remained. However, the controversy shows that the training of the user involves a negotiation between more closed and open modes of interfacing differently fit users and robotic devices.

When compared to the grand vision of robots caring for older people, the procedures of selecting, assessing and training users turn those visions upside down. They point to the need for attachment to attain autonomy not only for users but also for the technology itself. At least at the prototype stages, in which I have observed them, robots reliably attach themselves only to particular older users, who are resilient enough and are able to manage the sturdy and frail object that is a robot. The users’ ability to be independent is, if anything, not the result of the interaction with robots but its prerequisite. Robots need people who are fit to use them, and this fitness is the result of a long chain of interfacing processes that, on the one hand, adapt people’s behaviour to robots’ needs but also, on the other hand, render robots available for people as they try to accommodate (albeit restricted) levels of indeterminacy in HRI. Caring for robots results in a particular co-construction of users and robots, where roboticists’ implicitly restrictive user representations come to the fore.

Corridors of interaction: Calibrating ‘speech’ and voices

The analysis above should not create the impression that users and robots do not interact at all or that their interaction is completely determined, either by roboticists or by the robotic system. Rather, robots and users together produce controlled, but not completely determined, pathways through which interactions can come into being. As an example, I will discuss the speech user interface again, which affords a meticulous calibration procedure and the adjustment of a myriad of elements, both human and non-human. From that arise what I would like to call corridors of interaction. Such corridors are (de-)stabilized by a constant tension between roboticists’ planning efforts in configuring the robot’s speech interface before and the situated speech acts of users during the HRI experiments (Suchman, 2007). Hence, to stabilize interaction involves the reconfiguration of human bodies vis-à-vis the robot system (Alač, 2009). However the ‘norm’ to which such a speech corridor is calibrated is not apparent beforehand. It needs to be established in the constant interfacing effort between roboticists, users and the machine.

The following field note illustrates roboticists’ calibrating practices in order to engender communication between user and robots.

Out of sync

During test runs, Carol and Andrea sit down on the couch in the living room right next to the chair where the test subject has to sit. In front of them on the table sits a laptop running the speech recognition software. On its screen, they monitor what the system recognizes as input by the user and its interpretation of it (e.g., a request for a particular service). They constantly compare the screen with the situation next to them.

During the test runs, interaction repeatedly stalls, often with long pauses, during which the user simply waits for a response on the part of the robot. In such situations, Carol and Andrea instruct the test subject what to say to the robot. ‘You have to say “No”!’, ‘Please try again!’ or ‘Again, once again!’ With such instructions, they also intervene into the way the test subject holds the microphone in their hand. Andrea tells them to ‘hold the microphone like this’ indicating the middle of his chest. In another instance, Carol points out that the microphone is too close to the user’s mouth. The test run continues, marked by many pauses, waits, and instructions.

After the test run is over, Andrea exclaims, ‘Everything that could go wrong, did go wrong’. Everyone in the team agrees that this test run was not good. Francis explains ‘If you say something wrong, everything goes out of sync’. And: ‘The girl said things in the wrong order [he laughs]. I mean, for us, because we never tested it that way. For example, the sentence “Hey, Robo” restarted the whole sequence, even though it should not’. Roosje adds that the robot seemed not to be prepared for the test subject’s question ‘What did you say?’

Interaction via ‘speech’ proved to be an especially hard challenge for many reasons. For example, it could be that the system failed to recognize the user’s voice or that the microphone recorded a conversation of the user with the interviewer not intended for the interaction with the robot. Some utterances also caused the system to proceed in unintended ways, such as re-starting the whole sequence. This meant that the experiments were marked by recurrent pauses and breakdowns but the reasons for those often escaped the experts. They could not look into or change the insides of the speech user interface but instead had to reconfigure the test subject’s behaviour. They tried to correct the user’s behaviour according to what they deemed the robot system needed in order to proceed. This could mean reminding them to use ‘proper’ vocabulary or to take the ‘right’ distance to the microphone.

However, this is not the whole story of the calibration procedure. First, much of the work in between test runs is invested in the technological components and their re-adjustment in response to the events during the experiments. Roboticists acknowledge that the robot was not prepared and that they ‘need to make the system more robust’, as Francis tells me after another test run with speech problems. So, roboticists see that their technical plans fail and respond to that by continuously trying to adapt them with respect to the situated actions during HRI experiments. Second, interactive machines such as social robots exhibit a certain level of non-triviality (Dickel and Lipp, 2017). To users, but also to designers, it is not immediately apparent why the system behaved like it did or why it did not respond. As I have previously illustrated, roboticists do not manage to build up a stock of robust knowledge when it comes to their system’s interactions with the environment. Hence, communicating with interactive machines is constantly stricken with ambiguity about what (non-)response or repetition on the part of the system means to the user (see Suchman, 2007: 125ff) – and to the designers. Hence, the supposed ‘right order’ of speaking with a robot is not apparent beforehand but needs to be produced through the reparative efforts described above. Another version of this was apparent earlier in the calibration procedure, before a test subject was allowed to interact with the robot, as they had to test the speech interface. For that they uttered a number of commands, after which roboticists checked whether the system had ‘accepted’ their utterances. If not, they would tell them to speak louder or slower. One particularly interesting instance occurred when a test subject, in anticipation of the robot having difficulties understanding her, spoke in a rather mechanical fashion. The system did not recognize her voice and so roboticists told her to ‘speak more naturally’, since the system was able to understand ‘normal speech’ they argued. This is an instance where care is enacted in a double sense. On the one hand, the robot system is protected against the volatility of human speech by disciplining the user. On the other hand, the robot is being cared for in a conversational sense, defended against the impression that it can only deal with a very narrow corridor of possible actions.

Hence, the separation of ‘wrong’ ways of speaking rests on practices of interfacing, which aim to render recalcitrant, disparate entities available for one another. However, this is not simply about normalization or discipline (Foucault, 1995) but about the reciprocal re-calibration of the various distributed entities vis-à-vis one another. The voice of the user is not simply disciplined in relation to a previously apparent norm. None of the participants, including the roboticists, know exactly how the system will react to certain inputs and when a corridor of interaction is (temporarily) stabilized. It is precisely because of this that such laborious, ordinary technical practices of calibration are needed: not to conform to the system but to produce that norm in the first place. Hence, the interaction between user and robot are not simply stabilized by ‘the speech interface’. Rather, corridors of interaction afford the constant and careful calibration of bodies, pieces of code, a microphone and speech acts with respect to one another. The fact that at the end of the procedure HRI ‘finally works’ does not mean that it will work in future instances. To speak with Barad, interfacing practices ‘iteratively reconfigure what is possible and what is impossible – possibilities do not sit still’ (Barad, 2003: 234). The microphone, the software, the receiver, the user’s vocal tract or posture may cease to be compliant and new efforts to interface may be needed.

Conclusion: Interfacing robots and care in a more-than-human-world

The promise of care robots is that they would take over domestic tasks for older people, allowing them to live more independently. In this promissory discourse, robots are portrayed as autonomous and faultless machines able to seamlessly integrate into our everyday lives. However, the empirical analysis of interfacing practices in a European care robotics project suggests otherwise: Rather than robots caring for people, we witness numerous instances where people must care for robots to make them work in everyday environments. This care involves an asymmetry not between the faultless machine and frail human beings but rather the other way around, between fragile robots and concerned roboticists.

Caring for robots involves an ensemble of practices, which have to deal with the precariousness of HRI. It sets in motion various efforts to stabilize that which is not yet ‘in place’. A robot can be seen as a highly distributed system with respect to its technical components and the multi-disciplinary knowledges needed to build it. Robots come into being, if at all, when these elements are rendered available for one another within locally situated milieus of experimentation. This makes integration a laborious endeavour, which entails endless tinkering, ad hoc improvisation and a kind of localized gathering of knowledge and people.

Moreover, the homelike milieu of the test apartment is a hostile terrain for robots. Even though this seems to be a perfectly ordered place, the apartment ‘kicks back’ in all sorts of ways (Barad, 1998: 112). Changing lighting conditions, carpets and dirty shoes bring the experiments to the verge of breakdown. Most importantly, these issues cannot be remedied by way of technical integration alone but require mundane precautions: removing carpets, handing over objects or taking off shoes. Shielding the robot from the intricacies of the apartment’s milieu involves the users and their bodies as well. Users, too, need to become ‘fit for robots’, able to deal with their fragile nature. Finally, these interfacing efforts produce (temporary) corridors of interaction. These involve the meticulous re-calibration of many human and technical elements vis-à-vis a standard that is not apparent beforehand to either test users or roboticists. All of these elements attain agency in the process of being interfaced in the precarious event that is HRI. The latter requires the anticipatory protection and preparation of a robot-friendly milieu that involves the robot’s technical parts, mundane objects, fit users and temporarily stable corridors of interaction.

Notwithstanding their heterogeneity and situatedness, these practices share and render visible a number of features crucial for developing a more general notion of human-machine interfacing. First, they are reciprocal. In narratives of automation, there often is a uni-directional determinism. For example, robots are believed to have direct effects on society and on ‘us’ humans (Neven and Leeson, 2015; Šabanović, 2010). Interestingly, this determinist stance is often shared among apologists and critics of robots alike. Both the fears of dehumanization (Van Est, 2014: 69; Sparrow and Sparrow, 2006) and the hopes for more efficiency and lower costs (EC, 2009: 73), in the end, assume a direct impact of robots. Care with robots is not uni-directional, but rather involves reciprocally rendering formerly disparate elements available for one another. For robots to participate in care practice, they need to be accommodated first – this means caring for robots, too.

This ‘appropriation’ is never fully apparent beforehand, because interfacing is processual. As shown in the case of calibration, the standard for a ‘right fit’ between human, machine and all of the other elements involved is not a given but rather needs to come into being intra-actively, that is, by being rendered available for one another. Speaking with Simondon, technical objects such as robots are never fully but only temporarily determined within particular milieus. However, such milieus are not stabilizing per se but rather confront the robotic object with constant noise and interferences. Hence, developing robots for care is not about simply building a robot in the narrow sense but rather about continuously interfacing and thus re-working a whole milieu of often mundane elements.

Just like those elements, practices of human-machine interfacing often remain invisible. At best, mundane courtesies, precarious demonstrations or mishaps are shrugged off or laughed at by roboticists. Even more so, the attempt at rendering them visible, that is, recording them via video, is actively resisted. None of the researchers really wanted to acknowledge the importance of their courtesies; not to me nor, of course, to their colleagues at conferences, let alone their funders. This points not only to the peculiarities of uncertain experimentation but also to a more general, epistemic politics of seemingly ‘intelligent’ technology (Suchman, 2007: 217). This is all the more dangerous, since the benefit of automation will crucially be decided by how healthcare, lifestyles and homes need to adapt to those machines, in order to make them work at all. There are, however, first attempts in the community to acknowledge the fragility of robots. For instance, Lammer et al. (2014) have conducted HRI experiments in which they configure their robot so that if it does not succeed in a task it asks the user for help. These authors argue that rendering visible the dependence of the robot on human help actively engages older people in their care, increasing their sense of agency and wellbeing while also enhancing the acceptability of care with robots (Lammer et al., 2011).

Returning to the introduction’s artwork, a focus on robotic needs poses profound questions about the nature of care, as well as the costs associated with its ever-increasing technologization. Amplifying practices of human-machine interfacing provides a critical lever with which to question both overly optimistic and pessimistic views on the outlook of having robots care for older people. For robotics and similar ‘intelligent’ technologies, their ostensible autonomy and smartness is crucial to their promissory lure. Thus, carving out their constitutively frail and affective being as well as their dependency on neglected caring practices provides a crucial counterweight to their promissory power (Lipp, 2019; Neven and Peine, 2017). Attending to the ways in which not only roboticists but also potential prospective users of robot technology have to adapt their environment to host robotic objects (Forlizzi, 2007) provides insights into the additional care work underlying processes of automation (Treusch, 2017). This additional care work might not be exclusively directed at humans but also include non-humans like computer screens and, possibly, robots. This aspect is often criticized for deflecting attention away from the body of the care receiver ‘losing human experience’ (Hülsken-Giesler, 2017: 163). Care technologies, like robots, are thus often portrayed as alienating care from its ‘human’ core (Sparrow and Sparrow, 2006). However, such critiques assume a dichotomy between care and technology that remains inattentive to the constitutive role of technology in care practices (Lapum et al., 2012; Mol et al., 2010). Moreover, it neglects the affective qualities that are enacted by the users of such technologies with such technologies (Pols and Moser, 2009). Hence, accounting for care directed at the non-human allows us to question the overly technocentric as well as anthropocentric assumptions that underscore much of (Western) sensibilities vis-à-vis robotic caregivers.

Finally, conceiving of care in a de-centred, ‘more than human world’ (Puig de la Bellacasa, 2017) prompts us to re-think established paradigms and distributions of labour inscribed into and enacted with care robots. In the end, the central issue of such a perspective is neither to uncritically herald the robotization of care work nor to wish for a human-centred, robot-free future of care. Rather, this perspective opens up a new playing field for enquiring into, engaging with, and shaping the social, technological, political and material-discursive processes of interfacing robotics and care (Lipp, 2019). From the case presented, we can learn that robotics is not merely the modernist, rationalizing demon but rather itself in need of careful attention and support. Conversely, care is not solely about humanness but needs to be re-directed (partly) towards caring for the non-human and, more specifically, the robotic as well.

Footnotes

Acknowledgements

I would like to thank three anonymous reviewers for their extensive and insightful feedback, which helped me substantially in developing the argument that I present in this article. I am grateful to Pat Treusch, Arne Maibaum, Vera Gallistl and Marc Strotmann for their constructive critique and encouragement. Finally, I thank Lucy Suchman, collaborating editor of this journal, for additional comments and a smooth review process.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Benjamin Lipp

Author biography

Benjamin Lipp studied sociology, economics and criminology at the Ludwig Maximilian University of Munich. In 2019, he received his doctorate from the Technical University of Munich for his investigation of care robotics and European innovation politics. Since October 2021, he has led a Marie Curie postdoctoral project on neuromodulation and chronic pain at the University of Hamburg and Cornell University. In his research he combines Science and Technology Studies, sociology, and new materialist thinking to theorize the human-machine interface, especially in arrangements of care.

References

Akrich

(1995) User representations: Practices, methods and sociology. In: Rip

Schot

Misa

(eds) Managing Technology in Society. The Approach of Constructive Technology Assessment. London: Pinter, 167–184.

Alač

(2009) Moving android: On social robots and body-in-interaction. Social Studies of Science 39(4): 491–528.

Alač

Movellan

Tanaka

(2011) When a robot is social: Spatial arrangements and multimodal semiotic engagement in the practice of social robotics. Social Studies of Science 41(6): 893–926.

Atanasoski

Vora

(2019) Surrogate Humanity: Race, Robots, and the Politics of Technological Futures. Durham: Duke University Press.

Barad

(1998) Getting real: Technoscientific practices and the materialization of reality. Differences 10(2): 88–128.

Barad

(2003) Posthumanist performativity: Toward an understanding of how matter comes to matter. Journal of Women in Culture and Society 28(3): 801–831.

Barad

(2007) Meeting the Universe Halfway: Quantum Physics and the Entanglement of Matter and Meaning. Durham: Duke University Press.

Barry

Born

Weszkalnys

(2008) Logics of interdisciplinarity. Economy and Society 37(1): 20–49.

Bischof

(2017) Soziale Maschinen bauen: Epistemische Praktiken der Sozialrobotik. Bielefeld: Transcript.

10.

Dickel

Lipp

(2016) Systemtheorie und Technikkritik. In: Möller K and Siri J (eds) Systemtheorie und Gesellschaftskritik. Perspektiven der Kritischen Systemtheorie. Bielefeld: Transcript, 151–168.

11.

European Commission (EC) (2009) FP7 Cooperation Work Programme. Information and Communication Technologies. Updated Work Programme 2009 and Work Programme 2010. Brussels: European Commission.

12.

European Commission (EC) (2011) ICT – Information and Communication Technologies. Updated Work Programme 2011 and Work Programme 2012. Brussels: European Commission.

13.

European Commission (EC) (2021) Digital Economy and Society Index (DESI) 2021. Brussels: European Commission. Available at: https://ec.europa.eu/newsroom/dae/redirection/document/80563 (accessed 14 March 2022).

14.

European Innovation Partnership on Active and Healthy Ageing (EIPAHA) (2012) Action Plan on ‘Development of Interoperable Independent Living Solutions, Including Guidelines for Business Models’. Brussels: EIPAHA.

15.

Fleck

Webster

Williams

(1990) Dynamics of information technology implementation: A reassessment of paradigms and trajectories of development. Futures 22(6): 618–640.

16.

Forlizzi

(2007) How robotic products become social products: An ethnographic study of cleaning in the home. In: Breazeal

(ed.) Proceedings of the ACMIEEE international conference on human-robot interaction. New York: ACM (Association for Computing Machinery), 129–136. ACM.

17.

Foucault

(1995) Discipline and Punish: The Birth of the Prison. New York: Vintage Books.

18.

Halpern

Mitchell

Geoghegan

(2017) The smartness mandate: Notes toward a critique. Grey Room 68: 106–129.

19.

Heeren

(2013) Where the machines go to school. BI Research 43.2013/2014: 4–9.

20.

Hülsken-Giesler

(2017) Mimesis in nursing practice: The hermeneutical potential of the body to understand patients’ lived experiences. In: Foth

Holmes

Hülsken-Giesler

, et al (eds) Critical Approaches in Nursing Theory and Nursing Research. Implications for Nursing Practice. Göttingen: V&R (Vandenhoeck & Ruprecht) Unipress, 151–168.

21.

Knorr-Cetina

(1997) Sociality with objects: Social relations in postsocial knowledge societies. Theory Culture & Society 14(4): 1–30.

22.

Lammer

Huber

Weiss

, et al. (2014) Mutual care: How older adults react when they should help their care robot. In: Kibble

(ed) Proceedings of the 3rd international symposium on new frontiers in human–robot interaction. London: Routledge, 1–8.

23.

Lammer

Huber

Zagler

, et al. (2011) Mutual-care: Users will love their imperfect social assistive robots. In: Mutlu

Bartneck

Ham

, et al (eds) Work-in-progress proceedings of the international conference on social robotics. Amsterdam: ICSR (International Conference on Social Robotics), 24–25.

24.

Lapum

Fredericks

Beanlands

, et al. (2012) A cyborg ontology in health care: Traversing into the liminal space between technology and person-centred practice. Nursing Philosophy 13(4): 276–288.

25.

Latour

(1991) Technology is society made durable. In: Law

(ed.) A Sociology of Monsters. Essays on Power, Technology and Domination. London and New York: Routledge, 103–131.

26.

Law

(2004) After Method: Mess in Social Science Research. London: Routledge.

27.

Lipp

(2017) Analytik des interfacing: Zur materialität technologischer verschaltung in prototypischen milieus robotisierter pflege. Behemoth 10(1): 107–129.

28.

Lipp

(2019) Interfacing RobotCare: On the techno-politics of innovation. PhD Thesis, Technical University of Munich, Germany.

29.

Lipp

(2022) Interfaching the Human/Machine. Distinktion: Journal of Social Theory: 1–19 (no issue or volume available)

30.

López Gómez

(2015) Little arrangements that matter. Rethinking autonomy-enabling innovations for later life. Technological Forecasting and Social Change 93: 91–101.

31.

Maibaum

Bischof

Hergesell

, et al. (2021) A critique of robotics in health care. AI & Society. Epub ahead of print 16 April 2021. DOI: 10.1007/s00146-021-01206-z

32.

Matsuzaki

Lindemann

(2016) The autonomy-safety-paradox of service robotics in Europe and Japan: A comparative analysis. AI & Society 31(4): 501–517.

33.

Meister

(2014) When is a robot really social? An outline of the robot sociologicus. In: Pfadenhauer

Böhle

(eds) Of social robots and artificial companions: Contributions from the social sciences. STI Studies 10(1): 107–134.

34.

Mol

Moser

Pols

(2010) Care: Putting practice into theory. In: Mol

Moser

Pols

(eds) Care in Practice: On Tinkering in Clinics, Homes and Farms. Bielefeld: Transcript, 7–25.

35.

Neven

(2011) Representations of the Old and Ageing in the Design of the New and Emerging: Assessing the Design of Ambient Intelligence Technologies for Older People. Enschede: University of Twente.

36.

Neven

(2015) By and means? Questioning the link between gerontechnological innovation and older people’s wish to live at home. Technological Forecasting and Social Change 93(April): 32–43.

37.

Neven

Leeson

(2015) Beyond determinism: Understanding actual use of social robots by older people. In: Prendergast

Garattini

(eds) Aging and the Digital Life Course. New York and Oxford: Berghahn, 84–105.

38.

Neven

Peine

(2017) From triple win to triple sin: How a problematic future discourse is shaping the way people age with technology. Societies 7(3): 26.

39.

Oudshoorn

Pinch

(2005) Introduction: How users and non-users matter. In: Oudshoorn

Pinch

(eds) How Users Matter. The Co-construction of Users and Technology. Cambridge: MIT Press, 1–25.

40.

Pfadenhauer

Dukat

(2015) Robot caregiver or robot-supported caregiving? International Journal of Social Robotics 7(3): 393–406.

41.

Pink

(2010) Video in ethnographic research. In: Pink

(ed.) Doing Visual Ethnography. Images, Media and Representation in Research. Los Angeles: SAGE Publications, 96–116.

42.

Pols

(2012) Care at a Distance: On the Closeness of Technology. Amsterdam: Amsterdam University Press.

43.

Pols

Moser

(2009) Cold technologies versus warm care? On affective and social relations with and through care technologies. Alter 3(2): 159–178.

44.

Puig de la Bellacasa

(2011) Matters of care in technoscience: Assembling neglected things. Social Studies of Science 41(1): 85–106.

45.

Puig de la Bellacasa

(2017) Matters of Care: Speculative Ethics in More Than Human Worlds. Minneapolis: University of Minnesota Press.

46.

Šabanović

(2010) Robots in society, society in robots: Mutual shaping of society and technology as a framework for social robot design. International Journal of Social Robotics 2(4): 439–450.

47.

Šabanović

Michalowski

Simmons

(2006) Robots in the wild: Observing human-robot social interaction outside the lab. In: Institute of Electrical and Electronics Engineers (ed.) The 9th IEEE International Workshop on Advanced Motion Control. IEEE International Workshop on Advanced Motion Control. Istanbul: Piscataway, 576–581.

48.

Sánchez-Criado

López

Roberts

, et al. (2014) Installing telecare, installing users. Science Technology & Human Values 39(5): 694–719.

49.

Shapin

(1989) The invisible technician. American Scientist 77(6): 554–563.

50.

Simondon

(2009) Technical mentality. Parrhesia (7): 17–27.

51.

Simondon

(2017) On the Mode of Existence of Technical Objects. Minneapolis: University of Minnesota Press.

52.

Sparrow

(2006) In the hands of machines? The future of aged care. Minds and Machines 16(2): 141–161.

53.

Suchman

(2007) Human-Machine Reconfigurations: Plans and Situated Actions. Cambridge: Cambridge University Press.

54.

Suchman

Trigg

(1991) Understanding practice: Video as a medium for reflection and design. In: Greenbaum

(ed.) Design at Work. Cooperative Design of Computer Systems. Hillsdale: Erlbaum, 65–90.

55.

Treusch

(2015) Robotic Companionship: The Making of Anthropomatic Kitchen Robots in Queer Feminist Technoscience Perspective. Dissertation. Linköping: TEMA (Department of Thematic Studies.

56.

Treusch

(2017) The art of failure in robotics: Queering the (un)making of success and failure in the companion robot laboratory. Catalysts 3(2): 1–27.

57.

Trujillo Pisanty

(2011) With robots. Available at: www.trujillodiego.com/work/withrobots.html (accessed 27 August 2020).

58.

Tuma

(2019) Vernacular video analysis – Communicative construction of evidence in police work. Society Register 3(1): 137–162.

59.

Vallès-Peris

Domènech

(2020) Roboticists’ imaginaries of robots for care: The radical imaginary as a tool for an ethical discussion. Engineering Studies 12(3): 157–176.

60.

Van Est

(2014) Intimate Technology: The Battle for Our Body and Behaviour. Den Haag: Rathenau Instituut.

61.

Veling

McGinn

(2021) Qualitative research in HRI: A review and taxonomy. International Journal of Social Robotics 13: 1689–1709.

62.

Vinck

Blanco

(2003) Everyday Engineering: An Ethnography of Design and Innovation. Cambridge: MIT Press.

63.

Von Bose

Treusch

(2013) Von ‘helfenden Händen’ in Robotik und Krankenhaus. Zur bedeutung einzelner Handgriffe in aktuellen Aushandlungen um Pflege. Feministische Studien 13(2): 253–266.

64.

Voss

(2021) More Than Machines? The Attribution of (In)Animacy to Robot Technology. Bielefeld: Transcript.

65.

Woolgar

(1991) Configuring the user: The case of usability trials. In: Law

(ed.) A Sociology of Monsters. Essays on Power, Technology and Domination. London and New York: Routledge, 57–99.