Infrastructuring AI and Robotics for Health and Everyday Life: Visions and Practices in a Japanese National Moonshot Research and Development Project

ABSTRACT

This article examines how emerging infrastructural AI and robotics for health and everyday life are proposed and negotiated, drawing on ethnographic research within the Japanese Awareness AI Project. The project envisages distributed assistance relying on continuous measurement of bodily states, aiming to make technological systems recede into the background. The analysis reveals how the project envisions bodies and machines as tightly coupled, yet encounters profound sociotechnical frictions around data integration, consent and bodily safety. It also traces perceived side effects, including the commercial circulation of bodily data and the risk of reproducing social inequalities. The article argues that such attempts at infrastructuring, although recently curtailed by shifting policy priorities, remain sociologically significant. They highlight the unstable nature of infrastructure in the making and provide a critical reference point for understanding emerging reorganisations of dependence and responsibility and forms of assisted selfhood, around continuous bodily datafication.

1. Introduction

What kind of technology could be a lifelong companion to a human being?

This question, displayed on a board at the demonstration site of the case project discussed in this paper at Expo 2025 Osaka, signals a broader contemporary moment in which digital technologies are being deployed to address perceived societal challenges, particularly in the domains of health and everyday life. Across technologically oriented societies, digital technologies and more specifically modern autonomous and automated technologies such as Artificial Intelligence (AI) and robotics, are increasingly promoted as practical responses to demographic change, labour shortages in care and service work and rising pressures on healthcare and welfare systems (Domènech and Vallès‐Peris ²⁰²⁴; Elliott ²⁰²⁴; Wright and De Togni ²⁰²⁴).

These initiatives frequently frame digital technologies as tools for empowerment and participation: technologists promise to extend individuals' capacities to manage their own health and to stay connected to services and communities across the life course (Erikainen et al. 2019; Marent and Henwood ²⁰²³; Marent and Langstrup ²⁰²³; Petersen ²⁰¹⁹; Wright and De Togni ²⁰²⁴). At the same time, a substantial body of sociological work has shown that such promises are closely knitted with neoliberal forms of selfhood and surveillance, extending expectations that individuals continuously monitor and optimise their bodies and behaviours (Elliott 2019; Erikainen et al. ²⁰¹⁹; Lupton ²⁰¹⁶; Petersen ²⁰¹⁹). Discussions of self‐tracking and bio‐digital forms of citizenship point to a redistribution of responsibilities and risks and to a growing dependence on access to digital infrastructures as a condition for participation in contemporary societies (Lupton 2016; Marent and Henwood ²⁰²³; Marent and Langstrup ²⁰²³; Neff and Nafus ²⁰¹⁶; Petersen ²⁰¹⁹; Ruckenstein ²⁰²³).

Within this shifting landscape of digital health, the research presented here examines how AI and robotics come to be framed and pursued as infrastructure in their own right, rather than as mere tools embedded in pre‐existing healthcare practices or infrastructures. Drawing on sociological and related debates on digital health, algorithms and embodied technologies, alongside infrastructure, the paper foregrounds a case in which infrastructural arrangements are envisioned around continuous bodily sensing and adaptive support and in which AI and robotics are envisaged as body‐proximate infrastructural supports that must be materially coupled to bodies. By following these configurations at a formative stage from within its research and development process, the analysis extends empirical qualitative medical sociological work that has tended to focus on end‐users of already establishing technologies and it highlights how infrastructuring efforts themselves become a critical site for understanding emerging health‐related norms in the organisation of care.

In Japan, these digitally mediated health and citizenship dynamics intersect with pronounced population ageing and concerns about national competitiveness, making AI and robotics central to policy agendas and research programmes aimed at sustaining everyday living and health‐related support (Cabinet Office (Japan) ^{2024, 2025}; De Togni ²⁰²⁴; JST (Japan Science and Technology Agency) ²⁰²⁴; Wright ²⁰²³; Wright and De Togni ²⁰²⁴). Here, AI refers to computational systems capable of executing complex tasks traditionally associated with human cognition and increasingly permeating quotidian practices within and beyond institutional settings. Robotics is understood here as physical systems engineered to simulate, augment or substitute for various forms of animate and inanimate activities. Current policy and research explore such systems primarily for their potential to support healthcare and eldercare, even as they extend to domains such as science and production. Against this backdrop and as will be illustrated through the Japanese case examined in this paper, infrastructure, defined here as pervasive enabling resources in network form, has become a particularly useful lens for examining emerging imaginings of AI and robotics for health and everyday life (Bowker et al. 2010; Star ¹⁹⁹⁹). While existing scholarship has raised doubts about the broader feasibility and desirability of far‐reaching AI‐ and robotics‐based developments, including their likely resource‐intensive and costly character (De Togni ²⁰²⁴; Petersen ²⁰¹⁹; Wright ²⁰²³), this article sets aside these broader assessments to analyse emerging infrastructural visions and practices as they are currently being articulated in and around the case project.

1.1. Emerging Infrastructural Visions in Japan

Classic studies of infrastructure have highlighted that technical systems can become taken‐for‐granted conditions of everyday routines and work, often remaining unnoticed until breakdown or maintenance makes them visible again (Bowker et al. 2010; Star ¹⁹⁹⁹). In contrast, this paper examines infrastructure in the making, following attempts to envisage AI and robotics as everyday resources for health and everyday life whereas these systems remain unsettled (De Togni ²⁰²⁴; Hoeyer ²⁰²³; Lipp ²⁰²³; Petersen ²⁰¹⁹; Ruckenstein ²⁰²³).

The empirical focus is a Japanese research and development initiative within the extensive Moonshot Research and Development Programme (hereafter the programme), specifically Moonshot Goal 3, which treats AI and robotics as key technologies for envisioning a 2050 society wherein people and technology coexist in harmony (Cabinet Office (Japan) ^{2024, 2025}). Within this programme, the Awareness AI Project (hereafter the project) has pursued since its initiation in 2022 a vision in which AI‐powered systems continuously measure and analyse bodily states and behaviours through wearable devices and ambient sensors to provide adaptive support in health and everyday life (JST (Japan Science and Technology Agency) ²⁰²⁴). In its public‐facing materials, including those displayed in the Future Life Experience exhibition at Expo 2025, the project describes aims such as ‘creating a society where people can live proactively while robots watch over us’ and ‘teaching people the awareness they need to live constructively’. These formulations express technological aspiration but also hint at a paternalistic orientation, raising questions about how agency and ‘awareness’ are imagined—whether as autonomous capacities of individuals or as outcomes of infrastructural and algorithmic guidance embedded in everyday life.

Rather than emphasising spectacular humanoid robots, a form often privileged in Japanese robotics, the project has explored distributed, background forms of assistance intended to blend into daily routines and environments. Importantly, the project's institutional mandate has focused primarily on software and biosignal analytics rather than building robotic hardware from scratch. Reflecting this, its team consists predominantly of researchers in neuroscience, medicine, rehabilitation and AI, alongside robotics engineers. In doing so, it has articulated AI and robotics not as singular artefacts but as components of an emerging infrastructure for health and everyday life centred on the continuous datafication and modulation of the body.

This infrastructural approach is significant not only for what the project proposes to build but also for the shifting and varied visions and expectations of technological futures that they reveal (Bareis and Katzenbach 2022; Konrad and Böhle ²⁰¹⁹; Petersen ²⁰¹⁹; Wright and De Togni ²⁰²⁴). Following a 2025 review, Moonshot Goal 3 was restructured to focus on humanoid robotics (JST (Japan Science and Technology Agency) ²⁰²⁶). This policy shift responds to the rapid global acceleration of humanoid‐robot research and the Japanese government's desire to leverage perceived national strengths in this area. Although the Awareness AI Project received the highest possible evaluation during a review prior to its closure, it still ended as a Moonshot initiative due to the programme's restructuring. As a result, the broader idea of AI and robotics as infrastructure for health and everyday life is less prominent in Japan, even as the infrastructural visions traced here remain sociologically significant (e.g., Lupton ²⁰¹⁶; Petersen ²⁰¹⁹). At the same time, the Moonshot scheme explicitly expects funded projects to develop follow‐up activities, including dissemination, outreach and transitions into subsequent research or implementation initiatives. In this sense, the envisaged systems and the networks of people involved are still emerging and in flux; therefore, I use the present tense when describing project developments during the fieldwork, while the past tense or past participles are adopted where the temporal context clearly allows.

1.2. Aim and Research Questions

The paper aims to clarify how emerging systems of bodily monitoring and support are proposed as ways of reorganising health and everyday life before infrastructural visions become stabilised and to situate these proposals within sociological understandings of contemporary and future living. It does so by addressing three sets of questions: (1) what kinds of bodily data and technological arrangements are assembled as building blocks for such an infrastructure and how these are pursued in practice; (2) what practical issues are encountered as bottlenecks in this process; and (3) what side‐effects and hidden consequences are attached to these infrastructuring practices. These sets of questions also structure the empirical analysis, with each of the three findings sections corresponding to one of them.

2. Methods: An Ethnography of Infrastructure in the Making

This study adopts a qualitative, ethnographic approach, with the author embedded as a sociologist in the Awareness AI Project from May 2024 until the end of its funding period in November 2025. Fieldwork included observing project meetings, academic conferences and public events such as the Expo 2025 Future Life Experience exhibition and producing field notes in these settings and in one of the project's laboratories, where the author was based during the fieldwork. These observations were complemented by publicly available project documents and policy‐related materials that articulated the project's aims and its position within the broader Moonshot programme.

In addition, 25 interviews were conducted between November 2024 and July 2025 with individuals connected to the project in varying ways. The interviews combined semi‐structured prompts with open‐ended conversation and primarily involved researchers at Japanese universities and research institutes in robotics, healthcare and the social sciences, along with several external scholars and practitioners who offered more independent or explicitly critical perspectives. Typical topics included interviewees' disciplinary trajectories, experiences with AI and robotics, expectations concerning the sociotechnical implications of emerging technologies, views on health‐related norms and public engagement and perceived sociotechnical obstacles in research and implementation. Most interviews were conducted in Japanese, with a small number in English; all Japanese excerpts were translated into English by the author, using AI‐based translation tools to fine‐tune tone and register.

The author's position within the Awareness AI Project shaped both access and perspective. Employed as an ELSI (Ethical, Legal and Social Implications/Issues) researcher, the author was expected to contribute to ethical and social reflection and to practical tasks, including preparing exhibition materials, supporting demonstrations and presenting research at interdisciplinary meetings. This dual role reflects dynamics described in studies of ELSI and responsible research and innovation, where social scientists are often positioned between promoting innovation and maintaining critical distance (Balmer et al. 2015; Hughes and Stilgoe ²⁰²⁴; Mikami et al. ²⁰²¹). In this study, these tensions were addressed by including both supportive and critical voices among the interviewees and by noting moments of unease or disagreement in field notes as part of the analysis.

The study is inspired by work on infrastructural inversion, which analyses seemingly mundane infrastructures by shifting attention to the work and assumptions that usually remain in the background (Bowker et al. 2010; Star ¹⁹⁹⁹). In the present case, the focus is on infrastructural visions and prototypes in formation, tracing how AI‐ and robotics‐based infrastructures for health and everyday life are configured during their development and imagined to appear as natural parts of everyday routines.

The research was developed through an abductive and iterative engagement with the field, with the analytical focus emerging gradually from recurring patterns in observations, interviews and documentary materials (Atkinson 2020; Clarke ²⁰¹⁰). Early observations of attempts to make robotic assistance unobtrusive whilst increasing the visibility of bodily states, together with recurrent concerns about technological integration, consent and safety, shaped the subsequent analysis of physical data, infrastructural bottlenecks and broader implications for health and everyday life.

In presenting the analysis, interview excerpts, field observations and documents are not treated as neatly separated bodies of data that must be systematically compared; instead, they are used as distinct sources that are drawn on when they most clearly illuminate the key themes identified in the study.

Ethical approval for the study was granted by the Nagoya University Graduate School of Medicine's Ethics Committee (Approval No. 2024‐0224) and informed consent was obtained from all interviewees. In line with committee guidance, ethnographic observations in public or project settings did not require separate written consent. To preserve confidentiality, quotations are attributed using minimal contextual descriptors (e.g., ‘one collaborator’) to convey relevant positionalities whilst preventing re‐identification.

3. Results

3.1. Coupling Bodily Data and Infrastructural Arrangements

The infrastructural vision pursued in the Awareness AI Project rests on a tight coupling between the backgrounded technology and the monitored bodies. Rather than operating largely independent of individual bodies, the envisaged systems depend on ongoing streams of bodily data as their primary resource, making bodily processes increasingly visible to technical systems whilst receding into the background of health and everyday life. In doing so, the project pushes further a tendency noted in medical sociological and related work, in which bodily datafication becomes a key medium through which contemporary configurations of patienthood, participation and surveillance are organised (Armstrong 2019; Erikainen et al. ²⁰¹⁹; Hoeyer ²⁰²³; Piras and Miele ²⁰¹⁷; Ruckenstein ²⁰²³), but it does so by treating bodies not only as sources of data but as indispensable components of infrastructural support.

According to the project's logic, unobtrusive integration of AI and robotics into daily life and healthcare requires broad and continuous datafication of bodily states and movements (JST (Japan Science and Technology Agency) ²⁰²⁴). Development efforts focus on non‐invasive measurement in pilot or controlled settings, framed as precursors to healthcare and everyday‐life spaces. A central area of work is gait‐related monitoring, including walking speed, using ‘smart’ insoles equipped with pressure sensors and markerless motion capture. The project tested these systems in a supermarket, hospital corners and laboratory‐affiliated facilities. In parallel, electromyography (EMG) via wearable sensors is tested primarily in laboratory settings, with both lines of work framed as precursors to future deployment in everyday living spaces or urban contexts to render selected bodily processes visible in familiar, often non‐clinical settings. In this sense, the project aligns with but also extends medical sociological accounts of how digital health tools bring monitoring into ordinary environments (e.g., Kristensen et al. ²⁰²¹; Piras and Miele ²⁰¹⁷), by explicitly framing these bodily signals as building blocks of an emerging, body‐proximate infrastructure rather than only as inputs to self‐ or institutional tracking practices.

In describing these developments, some project members drew parallels with earlier shifts from hospital‐centred care to continuous oversight in everyday life, noting that, for example, ‘if you wear a device and just live your everyday life, that in itself can count as or be taken as, healthcare or rehabilitation …. The same goes for things like markerless motion capture, where you do not even have to wear anything on your body’. Their narratives emphasised making bodies and bodily signals continually observable through sensors and analytics and in this sense resonate with David Armstrong's discussion of ‘surveillance medicine’ (Armstrong 1995), in which expanding networks of observation extend medical attention into daily routines. At the same time, participants stressed that they intended to make the systems work in the background, which, in their view, distinguished their aims from more overtly intrusive or punitive forms of monitoring, with one researcher who explained that

the idea of Society 5.0, which we also subscribe to, is a society where technology coexists with humans. Coexistence, as I see it, doesn't necessarily mean people consciously using technology, but technology just being there as a matter of course and helping us almost unconsciously (muishiki ni). By becoming an ordinary part of life and coexisting with us in this way, it gives people more leeway for human‐like activities. That's really what we should be aiming for, I think.

Crucially, the project's institutional mandate focused primarily on software and biosignal analytics rather than building robotic hardware from scratch, a task largely allocated to a separate but linked project. This focus was reflected in its team composition, which consisted predominantly of researchers in neuroscience, medicine, rehabilitation and AI, alongside robotics engineers. Because of this division of labour, Awareness AI concentrated on making these systems responsive to human states. In collaboration with this related project, Awareness AI also experiments with physical intervention modules, such as flexible limb‐support devices that adjust to users' movements and individual characteristics. Earlier prototypes required users to switch on assistance manually. In contrast, current work seeks to replace this explicit control with sensing of bodily movements, so that the devices can change their form responsively without the need for conscious switching. Here, too, the aim is to provide support that feels embedded in everyday routines, with bodily signals triggering intervention whereas the robotic systems remain as unobtrusive as possible. Project discussions often describe this orientation as aiming for ‘unconscious (muishiki)’ support, meaning assistance should operate largely outside users' active awareness rather than through deliberate, effortful engagement. These aspirations echo but also complicate self‐tracking studies that foreground reflexive engagement with data and explicit self‐management, whether self‐initiated or organisationally pushed (Kristensen et al. 2021; Lupton ²⁰¹⁶; Piras and Miele ²⁰¹⁷), by imagining support that works precisely when users do not attend to it.

The invisible side of this arrangement can be understood as a form of background aid, experienced less through its prominent presence than through its reliable functioning and becomes noticeable mainly when it fails (Star 1999). As infrastructure studies remind us, such invisibility is always relative and situational and is usually most acutely experienced by end users rather than to those responsible for designing or overseeing these systems (Bowker et al. 2010; Karasti and Blomberg ²⁰¹⁸; Star ¹⁹⁹⁹). For project members, this implies an ongoing design task to ensure that sensing components and analytical or assistive systems can operate quietly in the background. At the same time, the flows of bodily data they generate would remain available for analysis and potential intervention when needed, underscoring that infrastructural work here is as much about making bodies legible to machines as it is about keeping machines unobtrusive for users.

Researchers involved in developing biosignal measurement technologies often describe bodily functions as central to ‘what makes a person human’, emphasising biological processes over more abstract conceptions of personhood. Many argued that clarifying these bodily functions is a prerequisite for AI or robots to exist ‘naturally’ or ‘without discomfort’ in personal spaces, as one EMG researcher explained:

What I'm basically trying to do now with biosignal measurement is to understand what kind of state a person is in and, behind that, what makes a person ‘human’ in the first place. That's why I'm measuring things like surface EMG and analysing them.

The researcher linked this to recent developments in generative AI, suggesting that systems such as ChatGPT show how ‘human‐ness (hito rashisa)’ can be learnt from large volumes of human data and folded into everyday routines. By analogy, the researcher frames biosignal analysis as a way for robots to understand the bodily expressions of human‐ness, enabling them to fit more naturally into people's movements and routines rather than remain clearly external devices. A senior researcher from the collaborating project that explicitly frames its work in infrastructural terms articulated a similar orientation, contrasting their approach with humanoid‐centred imaginaries: ‘our project is not about having a single humanoid robot that can support everything. Rather, the basic premise is that the shape of the robot and the way it provides support should differ each time, depending on the person and the purpose’. From this perspective, robots would not be replacements for existing workers or roles, but tools that people can use to do things they could not otherwise do or to do them differently.

These concerns resonate with work that traces how the use of bodily information shapes contemporary forms of identity and subjectivities, but here they are articulated directly in terms of what bodies must disclose in order for AI and robotics to operate as background technical presence in everyday personal spaces.

From these perspectives, making bodily data visible is treated as an essential requirement for the envisaged infrastructural integration. The infrastructure is therefore imagined not merely as an external resource, but as something that must be continuously fed by and adjusted to, bodily states and behaviours. This requirement fundamentally distinguishes the envisioned infrastructure from traditional infrastructures such as water supply or electricity networks, which, as one roboticist explained to me, ‘are out there’ operating independently of individual bodies and ‘can be accessed when needed’. In contrast, the model pursued in Awareness AI is inseparably tied to individuals' bodies and their data; without such data, the technical system cannot function as intended. The body itself thus becomes a central component of infrastructural work, as the site of continuous sensing and adjustment.

Within this vision, bodily visibility and machine backgrounding are mutually reinforcing. Technologies such as the smart insoles and markerless motion capture would unobtrusively collect data on posture and movement during everyday activities, enabling the early detection of subtle changes such as ‘pre‐frailty’, a term frequently encountered in the field, without drawing overt attention to the devices themselves. For one roboticist, the ideal is a situation in which the technical system ‘feels like an extension of one's own body’ or, conversely, is so seamlessly integrated that users ‘feel they are acting entirely on their own’. In this sense, the more effectively bodily processes become legible to AI‐based analysis, the stronger the impetus to make the technical apparatus itself fade into the sensory background. Here, the envisaged infrastructures do not simply extend self‐tracking, but begin to shape what might be described as assisted selfhood, in which everyday capacities to act and to care for oneself are both supported and configured through ongoing interaction with backgrounded AI‐ and robotics‐based systems.

Several collaborators explicitly contrasted this orientation with familiar images of spectacular humanoid robots, describing their aim instead as creating distributed systems that quietly assist everyday activities in the background. One interviewee characterised attributing personhood to robots as ‘creepy (kimochiwarui)’ and another stressed that they were not trying to build ‘Doraemon‐like robots wandering everywhere’ but rather robots tailored to particular people and purposes across different environments.

These practices and narratives articulate a specific version of infrastructural AI and robotics in which enhanced bodily visibility and engineered machine backgrounding work as interdependent processes. The seamless operation of technical systems would enable new forms of bodily datafication (Armstrong 2019; Hoeyer ²⁰²³; Ruckenstein ²⁰²³) and expanding datafication in turn serves to justify further integration of AI and robotics into the background of health and everyday life.

The focus on the interdependence between the bodily datafication and technological backgrounding in the project extends debates in digital health. Reorganisations of patienthood and participation (Erikainen et al. 2019; Marent and Henwood ²⁰²³), as well as shifts in patient–provider relations in ‘clinical self‐tracking’ (Piras and Miele 2017), are already well documented. The present article adds to these discussions by highlighting that the project envisages AI and robotics to blend into daily routines at an almost unconscious, continuous level, going beyond the predominant emphasis on deliberate self‐management and explicit forms of communal data sharing. While work on self‐tracking has underlined the ‘relational affordances’ of digital devices and associated forms of care (Kristensen et al. 2021), the present study shows that considerably broader realms of everyday life and health are imagined as sites of technological backgrounding and bodily datafication, with individual effort deliberately kept to a minimum. This points to more passive, infrastructural possibilities and to a reorganisation of dependence and of responsibility within the entanglements between human agencies and autonomous and automated technologies, which differ from the forms of individualised subjectivity foregrounded in self‐tracking research.

3.2. Practical Bottlenecks in Infrastructuring AI and Robotics as Encountered in Project Developments

Rather than mere technical flaws, the practical bottlenecks encountered during the project's developments represent critical sites of sociotechnical negotiation, where the envisaged infrastructure is constantly threatened by potential breakdowns and maintained through ongoing reparative efforts (Lehtiniemi and Tarkkala 2025; Ruckenstein ²⁰²⁵). As recent scholarship on algorithmic systems highlights, the process of infrastructuring involves a cycle of breakdowns and repair before any system can fade into the taken‐for‐granted background. In this context, efforts to build AI and robotics infrastructure for health and everyday life confront three interrelated obstacles: coordinating devices, data flows and analytics to enable unobtrusive support; defining consent for continuous, unprompted bodily monitoring; and ensuring bodily safety in systems that intervene directly in users' physical routines.

The first type of bottleneck arises from attempts to integrate sensing, analysis and feedback across multiple devices and sites. Project members often stress that genuinely unobtrusive, everyday forms of support require seamless coordination between hardware and software, including data analytics, so that measurements can be processed quickly and translated into timely, meaningful feedback. This challenge became particularly salient in field trials using the smart insoles in a shopping mall environment, where the envisioned flows of data and responses proved difficult to maintain in practice. Below is an excerpt from fieldnotes:

On a weekend in October 2024, the project conducted preliminary measurements with ‘smart’ insoles equipped with four pressure sensors per insole in a large shopping mall, as preparation for a public ‘dynamic health check’ event scheduled for the following day. A booth on the first floor, next to a bookstore, invited visitors to have their ‘walking type’ analysed with sensor‐equipped insoles and to try tailored exercises with a health‐walking instructor, with shopping vouchers offered to early participants. Around noon, during a brief preliminary session, several project members and collaborators asked colleagues, including me, to serve as test participants. I wore the instrumented shoes and was instructed to walk back and forth along a ten‐metre path twice. A core member later explained the output to me, noting that my gait had been plotted in a ‘general’ group on a health index using gait‐related measurements as indicators of age‐related physical decline; the analysis, performed via a cloud‐based system, took about one minute. According to the measurements, my right foot showed a slightly longer ground contact time, but my walking—‘landing on the heel and rolling through to the toes’—appeared acceptable. Later in the afternoon, I again joined the same member and collaborators for further preparatory trials for the following day's ‘dynamic health check while shopping’. This time, I wore the shoes, placed a smartphone in my back pocket, put on a shoulder‐strap device for location logging and wore a smartwatch on my left wrist while walking freely through the supermarket area for about 10 to 20 minutes. In this run, the measurement appeared to have stopped partway through, apparently due to a connection failure and a laboratory colleague told me they had just experienced the same issue in a similar test. A second attempt led to the same problem and I left before the system was fully stabilised.

Subsequent preparations for the official demonstration suggested that some of these data had in fact been recorded, even though they were not accessible at the time of the trials. These episodes illustrate how easily tightly coupled flows between sensors, mobile devices and cloud‐based analysis can falter in busy commercial spaces and how fragile the ‘seamless’ support imagined by project members can be in practice.

A second set of bottlenecks concerns how consent is conceptualised in relation to continuous bodily monitoring. For many participants, the central promise of using physical data lies in predicting and classifying health states in chronic, gradually developing conditions such as frailty or lifestyle‐related diseases, rather than in acute or transient illnesses. In their view, meaningful prediction requires data that capture subtle, long‐term changes rather than occasional snapshots.

As one participant put it:

When individuals take the initiative to measure their condition because they feel something might be wrong—for example, deciding to get an MRI because their knee hurts—it's already too late by the time they notice the problem themselves.

From this perspective, continuous, ambient monitoring, likened by several interviewees to regular mandatory check‐ups, is valued for enabling earlier detection than voluntary, episodic checks. In this framing, dependence on continuous monitoring is not simply a technical convenience. Instead, it appears as a moral and practical requirement for responsible care, which in turn places pressure on how consent should be defined, obtained and renewed over time (Flaumenhaft and Ben‐Assuli ²⁰¹⁸; Fylan et al. ²⁰¹⁸).

At the same time, researchers frequently present their work as being only marginally concerned with identifiable personal information. As one participant described:

We're not really interested in people's names or anything like that. What matters is the actual measurement data, like gait speed or specific biometric signals. Once the models are operational, it's advisable to remove any personal details as soon as possible so we process only the features, not the data source. Of course, when we're conducting trials and experiments, sometimes we need to know a little more to ensure things are working correctly, but the goal is to have a setup where people themselves decide if and when they want to share their data. It's a bit like personal health records: if you go to the doctor, you decide what you show them.

Here, consent is framed less as a one‐time authorisation for data capture and more as an ongoing capacity to decide when and how one's data should be shared, even as the technical infrastructures in question depend on continuous and often backgrounded flows of bodily information. While existing digital health scholarship has highlighted the complexities of consent and data sharing—often framed as a ‘Faustian bargain’ where personal information is surrendered in exchange for perceived empowerment (Petersen 2019)—these accounts generally presume a model of conscious engagement. In such models, individuals are positioned as active participants expected to actively interpret their data and take responsibility for their health (Erikainen et al. 2019; Lupton ²⁰¹⁶; Piras and Miele ²⁰¹⁷). However, the Awareness AI project faces a dilemma regarding consent for infrastructural support, as user awareness is deliberately minimised to ensure seamless integration. While project members acknowledge the ethical importance of explicit consent, they also maintain that demanding conventional, fully conscious engagement could compromise the systems' technological efficacy in detecting and addressing subtle health issues. This dilemma introduces a distinct sociological and governance challenge: how to meaningfully sustain the ethical requirement for ongoing consent when the infrastructural goal is precisely to render the technology itself unnoticeable.

A third perceived bottleneck centres on bodily safety in robotic systems that intervene directly in movement. In the project's limb‐support work, for example, researchers experimented with devices that sense users' sit‐to‐stand movements and then provide assistance by subtly adjusting their physical form. These tensions between individual control and background optimisation are reflected in the development of robotic assistance, where designers aim to support users whilst minimising demands on their attention and effort. Fieldnotes from a trial session with one such device, where concerns about falling and losing balance, as well as worries about giving control over the robot, arose as a matter of course, show that questions of safety are negotiated and at times explicitly questioned in practice (De Togni ²⁰²⁴):

On an October morning in 2024, I observed and participated in preparatory sessions with a prototype limb‐support device borrowed from a collaborating laboratory. The device was attached to a roughly two‐metre‐high rack by several hanging belts. It consisted of a relatively chunky upper section, similar to the armrests of a chair, where both arms could be placed and a pair of ‘legs’ extending downward that could extend and contract in a coiled, spring‐like fashion. Project members were working to modify it so it would respond to sit‐to‐stand movements rather than operate via a basic on/off switch. Around 10 AM., two junior researchers from the collaborating lab and a doctoral student from another partner group joined the local team, including the principal investigator, several lab members and me. As they partially disassembled the device and explained its internal structure, wiring and control logic, one lab member recorded the process on a smartphone. I stopped what I was doing at the coffee machine when I realised the scene was being filmed, acutely aware that the device was in an early, unfinished state and being documented. Over lunch, the discussion turned explicitly to safety: one senior member insisted that ‘this could never be used directly with patients’ and the visitors mentioned the risk of older adults ‘getting seriously hurt’, including fractures, even though formal ethical approval for trials had already been obtained, framing the prototype as suitable only for healthy volunteers under close supervision.

In the early afternoon, two younger members of the host lab took turns trying out the device. As they stood up, the coiled supports extended and retracted in a somewhat unpredictable way, at times stretching farther than expected or nearly catching them underfoot. However, the device did appear to bear part of their body weight; each attempt to stand up looked precarious. When I was invited to try it myself, I declined, explaining that I was still recovering from a recent minor bicycle accident and my refusal was also a spontaneous response to unease about entrusting my balance to a system that the engineers themselves described as not yet safe enough for patients.

Later that month, further internal trials involved all male lab members as test participants, combining sit‐to‐stand exercises with multi‐site EMG measurements on the right side of the body, handled mainly by collaborators visiting from a partner institution.

As scholars at the intersection of medical sociology and STS have recently emphasised, integrating robotics into healthcare or non‐institutional settings requires moving beyond technical specifications to examine the broader sociomaterial conditions of safety and care (Domènech and Vallès‐Peris ²⁰²⁴). In the unpredictable spaces of everyday life, robotic systems remain inherently fragile, presenting a constant threat of breakdown (De Togni ²⁰²⁴; Lehtiniemi and Tarkkala ²⁰²⁵). Consequently, ensuring bodily safety becomes less about achieving perfect machine autonomy and more about an ongoing process of sociotechnical repair and adjustment. This dynamic reveals a paradoxical reversal of what Neves et al. (2025) critique as the unidirectional delegation of care to AI. Rather than the technology autonomously caring for the user, maintaining safety within these infrastructural arrangements often requires humans to continually accommodate and ‘care for’ the machine's clumsy or unpredictable behaviours (Lipp 2023).

Across these examples, the perceived bottlenecks highlight that building AI and robotics into the infrastructures of health and everyday life is not simply a matter of technical problem‐solving. Rather, it is a process fraught with sociotechnical frictions and the constant threat of breakdown. Whether managing the ethical dilemmas of consent for passive data collection or compensating for the inherent physical fragility of robotic bodies, infrastructuring requires continuous reparative efforts (Lehtiniemi and Tarkkala 2025). As Neves et al. (2025) suggest, delegating care to AI and robotics is often discursively framed as a seamless repair for human inefficiencies; yet paradoxically, the ambition to create an unobtrusive background of automated support in this project demands intense and ongoing human intervention. Ultimately, this demonstrates that making machines invisible in everyday life requires rendering them highly visible objects of human care and negotiation.

3.3. Perceived Potential Side Effects and Hidden Consequences of Infrastructural AI and Robotics

Beyond the project's stated aim of supporting health and everyday life, participants voiced concerns about the potential side effects and unintended consequences of infrastructural AI and robotics. These concerns centre on three primary issues: the commercial and managerial circulation of bodily data; the normalisation of particular health metrics and activity patterns; and the possibility of more direct interventions into physical and emotional states.

Participants noted that systems designed for health support could yield very different effects once embedded in broader commercial and everyday environments (Elliott 2019; Petersen ²⁰¹⁹; Zuboff ²⁰¹⁹). One collaborator illustrated a possible future setting in which commercial and health‐related sites link tightly with each other through personal data flows:

Supposing one's personal data could seamlessly transfer, imagine going to a shopping centre and using a cart that can immediately download your information or tapping your smartphone so that your data flow to the cart and allow it to support you optimally. It might say, ‘You ate this yesterday, so how about trying this special offer today?’ or ‘Here's a recipe based on what you bought and the current sales’. The same applies to rehabilitation and sports facilities: by checking in, the system could recall your recent activity levels, saying, for example, ‘You ran this many kilometres recently, your stamina has improved to this level, so how about this kind of training today?’ In this way, various environments could provide personalised, timely support just by accessing your data.

In such scenarios, everyday objects and spaces become active nodes in a larger system of personalised support and health‐oriented infrastructures slide into broader architectures of recommendation and behaviour control (Armstrong 1995; Hoeyer ²⁰²³; Marent and Langstrup ²⁰²³; Petersen ²⁰¹⁹; Ruckenstein ²⁰²³).

From this view, systems initially framed as tools for personalised health support also appear as attractive instruments for commercial targeting or workplace oversight. Some interviewees mentioned potential secondary uses of bodily data in commercial settings and noted that repeated exposure to specific metrics can gradually redefine what is considered ‘normal’ behaviour. Gait‐tracking technologies, for example, were discussed not only as tools for detecting pre‐frailty but also as potentially applicable for retail analytics or workplace oversight, such as mapping shopper behaviour in supermarkets or monitoring employee routines. One collaborator noted that large retail chains already use detailed customer segmentation based on purchase histories and loyalty cards and ‘you could easily imagine them wanting to track how their “best customers” move through a store to redesign layouts so that those people buy even more’. The same interviewee added that similar logics could apply in office settings: ‘in a normal workplace, some managers would probably be happy to know who is “slacking off”, going out for cigarettes all the time or always getting coffee’. Another commented that deployment ‘cannot rely only on goodwill; if no business incentives are involved, such systems will never scale’.

These discussions highlighted concerns that infrastructures built around continuous bodily data may impose specific health norms, for instance, by treating certain gait patterns or activity levels as baselines for desirable behaviour. Participants linked this to existing inequalities in access to digital tools and in the skills needed to use them. These discussions also highlighted concerns that infrastructures built around continuous bodily data may reinforce existing inequalities (Bowker et al. 2010; Burrell and Fourcade ²⁰²¹; Petersen ²⁰¹⁹; Rose ²⁰⁰⁷). One interviewee remarked that, at present, there is often a split between those who ‘use’ AI and robots and those who risk ‘being used’ by them:

Right now, it often feels like a relationship of users and the used. Some people use AI and robots effectively, but those who cannot may end up on the receiving end. To avoid being used in that way, you have to be careful and, above all, you need to understand the other side: things like how far robots can adapt, what the limits of AI are and what kind of information literacy is required.

Another participant pointed to how existing digital devices already contribute to educational divides:

The convenient devices we have now are already creating something like an educational class divide. Children in households that can use these tools can study, but children in other households where such devices are not available cannot. That kind of situation is already a problem to some extent.

These accounts capture a core sociological concern: rather than universally empowering citizens, data‐driven infrastructures risk reproducing and deepening existing social inequalities (Hoeyer 2023; Petersen ²⁰¹⁹; Ruckenstein ²⁰²³). For these interviewees, the ability to benefit from AI‐based infrastructures depends not only on access to devices but also on the capacity to understand the limits and possibilities of algorithmic systems. As critical studies of digital health note, such technologies often demand specific forms of knowledge and active engagement that are unevenly distributed across populations (Erikainen et al. 2019; Lupton ²⁰¹⁶). Furthermore, these infrastructural arrangements are never seamlessly universal. They inevitably risk creating new forms of marginalisation, excluding those who fall outside their standardised parameters and for whom the infrastructure becomes a barrier rather than an enabler (Lupton 2016; Petersen ²⁰¹⁹; Star ¹⁹⁹⁹).

While the literature widely recognises the importance of structured forms of oversight and public deliberation (Balmer et al. 2015; Mikami et al. ²⁰²¹), interviewees in this study rarely articulated concrete visions for participatory or citizen‐based initiatives, a pattern also noted in previous research on robotics and digital health in Japan (De Togni ²⁰²⁴; Wright ²⁰²³). When I raised participatory possibilities, they tended to elicit agreement or cautious interest. Yet, discussions about public engagement have so far remained confined mainly to internal ELSI meetings and other professional forums, rather than taking the form of broader citizen participation. Expectations that more participatory activities might develop within the project—such as extending ELSI‐related work beyond professional settings—were occasionally voiced by collaborators and by the author, but largely remained unrealised once the project came to an end.

External commentary added a further layer to these concerns. One social scientist, participating as an independent expert in ELSI discussions, drew on a Japanese science‐fiction novel to evoke a technologically saturated future in which pervasive health‐monitoring systems continuously regulate bodily states and even emotional conditions:

If Awareness AI‐like technologies mature, they might preemptively modulate neurochemicals based on predictive analytics. This scenario might sound like science fiction right now, but if we don't think through these possibilities now, we could be unprepared when they arrive.

For this commentator, such scenarios indicated that systems of the kind explored in the project might one day shift from supporting health‐related behaviour to influencing moods and neurochemical states, with uncertain consequences for consent and responsibility, as well as for individual bodies. In this sense, speculative references to science‐fiction futures served to bring potential unintended consequences into view before technical capabilities catch up.

Overall, the accounts discussed in this section suggest a shifting set of perceived side‐effects and hidden consequences, including the secondary use of bodily data for commercial oversight, the normalisation of particular health metrics and interventions and the possibility of more direct intervention in physical and emotional states. These possibilities are not framed as external additions to technical work; they are voiced by people involved in research and development and are likely to influence whether and on what terms, such systems would be taken up in practice. Strikingly, these concerns centre less on robot‐specific properties than on the commodification and governance of bodily data that are continuous with wider critiques of digital health and monitoring (Erikainen et al. 2019; Lupton ²⁰¹⁶; Petersen ²⁰¹⁹; Zuboff ²⁰¹⁹), suggesting that the project's robotic elements may, at least for now, serve mainly to intensify and reconfigure already familiar dynamics rather than to introduce entirely new kinds of risk.

Ultimately, bringing these side effects and hidden consequences into view highlights the sociological value of examining infrastructural work in formation rather than only after stabilisation. Observing these arrangements before they become firmly established makes it possible to capture what Ruckenstein (2025) terms the ‘darker qualities of repair’, namely, the often invisible and frustrating burdens placed on individuals who must constantly adjust their routines to compensate for the system's demands or data‐extractive logics. In doing so, this analysis complements existing studies of digital health by specifying what it means to conceptualise AI and robotics as infrastructures for health and everyday life (Erikainen et al. 2019; Hoeyer ²⁰²³; Petersen ²⁰¹⁹). It provides one way of bringing human‐oriented, as contrasted with machine‐centred, infrastructural imaginings into view, specifying how people might be configured not simply as passive data sources, but as participants in shaping the infrastructural conditions under which these technologies become part of ordinary social worlds (Clarke 2010; Star ¹⁹⁹⁹).

4. Conclusion

This article has examined the imaginings and practices surrounding emerging infrastructural AI and robotics for health and everyday life, using the Awareness AI Project as a case study. The project envisioned a dynamic, interdependent relationship between human bodies and technological systems, whilst also highlighting the profound practical bottlenecks and sociotechnical frictions inherent in such infrastructuring efforts.

In sum, the article traces the visions the project advanced, the perceived immediate challenges and the future considerations. First, it shows a proposed tight coupling between technological backgrounding and continuous, non‐invasive bodily measurements. Second, it examines key bottlenecks encountered during development, notably difficulties in meaningfully integrating monitoring and support whilst ensuring safety, revealing how conspicuous systems resist seamless integration into the background. Third, it analyses articulated side effects, including the commercial circulation of bodily data and the implicit reinforcement of particular health norms.

As noted in the introduction, following a renewed policy emphasis on humanoid robotics, the specific project described here was brought to an early close. While this closure might be interpreted simply as a technological or organisational ‘failure’, it underscores precisely the unstable, highly contested nature of infrastructure in the making from a sociological perspective. The shifting of national funding priorities reveals how infrastructural visions are vulnerable to policy disruptions long before the technologies themselves can stabilise or ‘fail’ in everyday use.

Building on this, the article makes two main contributions to medical sociology. First, it foregrounds the interdependence of bodily datafication and technological support as a distinctive foundation for emerging health infrastructures. Existing sociological scholarship has demonstrated that care infrastructures rely heavily on the everyday, intimate work of human caregivers (e.g., van Pijkeren et al. ²⁰²¹). Concurrently, research on digital health and self‐tracking has highlighted the redistribution of health responsibilities and shifting care relations (Kristensen et al. 2021; Lupton ²⁰¹⁶; Petersen ²⁰¹⁹; Piras and Miele ²⁰¹⁷). The present study extends these insights by specifying the envisioned, distinctive infrastructural dynamics of AI and robotics. The technologies pursued in the Awareness AI Project are intended to dynamically capture bodily functions to provide automated, backgrounded support. The project envisages bodily signals, such as gait or muscular responses, not only as sources of information but also as triggers for adaptive assistance, positioning the body within a responsive network where sensory data and algorithmic intervention are tightly linked. The analyses here invite medical sociology to attend more closely to emerging forms of assisted selfhood, understood here as modes of selfhood and governance shaped by ongoing, backgrounded bodily engagements with technology.

Second, by focusing on infrastructure at a formative stage rather than after consolidation, the analysis complements sociological work that usually centres on already stabilised systems (Bowker et al. 2010; Star ¹⁹⁹⁹). The Awareness AI Project offers a view of infrastructural work in motion, making visible the envisioned arrangements through which continuous bodily monitoring and automated assistance appear as ordinary features of health and everyday life. Attention to this early phase makes it possible to trace which future scenarios and concerns are allowed to develop and which are muted or sidelined and it recasts datafication and algorithmic governance as ongoing tests and trials rather than accomplished facts (Hoeyer 2023; Ruckenstein ²⁰²³). Moreover, this approach is relevant for examining digital health practices beyond the research and development stage. Tracing the struggles to push highly visible components into the background during the making of an infrastructure provides a critical foundation for understanding the often hidden maintenance work that would be required to sustain it in the future. This perspective positions the very construction of background health infrastructures as a key empirical site for examining emerging forms of control and dependence in the organisation of care.

Ultimately, the infrastructural vision traced in this article calls for a sociological imagination that moves beyond individual user‐device relationships to consider how AI and robotics are mobilised to reorganise the background conditions of health and everyday life. Such an approach highlights infrastructure as a deeply political arrangement that shapes the boundaries of participation, dependence and responsibility in contemporary societies.

Author Contributions

Hiroto Shimizu: conceptualization, data curation, formal analysis, writing – original draft, methodology, investigation, writing – review and editing.

Ethics Statement

All research involving human participants received ethical approval from the Nagoya University Graduate School of Medicine's Ethics Committee (Approval No. 2024‐0224).

Consent

Written informed consent to participate was obtained from interview participants. For the ethnographic participant observation component, the requirement for informed consent was waived by the committee as it was deemed impracticable to obtain.

Conflicts of Interest

The author was employed as an ELSI (Ethical, Legal, and Social Issues/Implications) researcher at Nagoya University under the Awareness AI Project, which was funded by the JST Moonshot R&D Program. This article was written as part of the author's concurrent roles as a sociologist and ELSI researcher within the project. The author's primary disciplinary identity remains that of a sociologist and every effort has been made to maintain scholarly independence in conducting and reporting this research.

Shimizu, Hiroto . 2026. “Infrastructuring AI and Robotics for Health and Everyday Life: Visions and Practices in a Japanese National Moonshot Research and Development Project,” Sociology of Health & Illness: e70259. 10.1111/1467-9566.70259.

Data Availability Statement

Data summaries supporting this study's findings are available from the author upon reasonable request.