Understanding occupants’ experiences in quantified buildings: results from a series of exploratory studies.

Rendering technology invisible within the built environment is not a new concern; Weiser’s vision for the computer in the 21st Century [69] set a primary direction for the later Ubicomp movement and contemporary IoT environments. Contemporaneous examples of the application of such practices are often referred to as smart buildings: buildings that integrate data-collection and processing technologies (IoT, AI) with a view to manage their energy and operation. An example of such buildings is the Edge [28]; an IoT- enabled office building in Amsterdam currently owned by Deloitte. The Edge showcases the benefits of IoT infrastructure and Building Information Modelling (BIM data processing platform) for post-construction building data analysis and use [53, 54]. The many sensors in the building collect predominantly environmental and occupancy data; processed by BIM for automated energy performance visualisation, building usage monitoring and post-processing for energy analysis [28].

Framing what a ‘smart building is not that simple; there is great ambiguity to the term [14, 28, 42]. Following Buckman et. al [14], a smart building collects and uses data points and AI with a purpose of adapting to the complex needs of their occupants over time. Inherent to their term is the occupant-centric approach of smart buildings [14]; smart buildings should be occupant-based, creating active participants [13, 55] through integrated technologies that provide feedback both towards and from occupants, and methods for user control [14].

Human-Building-Interaction research takes this human-centred dimension of smart buildings further, focusing primarily on human experiences, values and needs of the occupants of smart environments [3, 26, 47, 64]. While the "Smart Agenda" for the built environment emphasizes on improving efficiency, cost, and sustainability; HBI’s focus is on addressing people’s complex interactions and collective experiences with and within smart buildings [3, 26, 47, 64]. Human Building Interaction (HBI) research focuses on how occupants interact with buildings - the study of the interface between the occupants and the building’s physical space and the objects within it [64] - to provide a foundation to inform and direct the development of human-centred (and not merely user-centered) smart buildings. Adding to that, Adaptive Architecture research [31, 62, 71] focuses on the power of the physical and material dimensions of architecture as an interface, following an embodied and enacted approach to interaction [31]. HBI research ultimately shifts the focus from ‘smart buildings’ and ‘users’, to a wider smart agenda towards achieving sustainability, meaningful usage, and wellbeing in the buildings together with their occupants [10, 26, 55, 58].

The phenomenon of lived-in quantification emerges as the result of the constantly increasing tendency to inhabit smart buildings; referring to the intensive data-collection in smart environments [28, 41, 63]. More and more office and domestic environments are built to be permanently monitoring using a rich networks of sensors, collecting and processing and massive amount of data sets [63]– i.e. environmental, post-occupancy, operational data, user-generated data from mobile devices [16, 22, 54, 63] (see Newcastle future homes ¹ for instance) – exhibiting highly quantified environments. While ‘smart’ focuses on the potentials of IoT and automation [14, 28] -i.e. what we do with the data-, the term ’quantified’ addresses the form and scale of data collection – i.e. the amount and types of data collected – prior to the complexities of processing and using it. Although drawn from the concept of the ‘quantified-self’ [22, 24, 35, 57], quantified buildings present a very different case. As data tracking is spatialized, contextualized, and (opposed to self-tracking) often involuntary; quantified environments compose a unique and highly complex emerging research agenda [63].

Quantified smart buildings can indeed provide the ground data infrastructure for further research in improving lived-in spaces and the experiences in them. Referred as living-labs or lived-in Places following Alavi et al.’s categorization [4], some smart buildings are intentionally built as a real place and a research infrastructure in the same time – e.g London PEARL Living Labs ² for instance [4]. The envisioned HBI research outcomes of such living labs are interventions that bring value through supporting and enhancing the collective experiences [15, 31, 58, 71] of their occupants for the long term [3, 10, 58]; empowering them through the use of accumulated data - for example, by visualizing sensed data to the building occupants [1, 4, 6, 31]; support the health and wellbeing of their occupants – e.g. data-driven restorative environments [43, 44, 45, 65, 72] and improve the built environment and the life in it [31, 63].

Still, there are plenty examples where the line between research infrastructure and lived in smart building is hard to be drawn. In some cases, what differentiates a living lab from a lived-in smart building is the data access and ownership, the access to the space and infrastructure, and the focus and scope of a potential research agenda for the place [19]. The Edge described above [28] could be seen as such an example, as well as the office building used in this work (see section 2.4). The availability of the open datasets and access to lived-in quantified workplaces creates great opportunities for HBI research [3, 26, 47, 58]. But the reality of lived-in smart office buildings has unresolved complexities with regards to their users’ experiences, such as the feeling of control on their environments [42], and perceived and actual privacy [25, 46, 50]; whereby further research is meaningful. Boarder present challenges of lived-in quantified spaces relate with data sharing -i.e. what and how data is shared between individuals, and between individuals and the organization - data visibility – i.e. what data is the user able to experience in the building and how - data use – i.e. applications for wellbeing and collective life - data control, and ultimately, the privacy of the building occupants [23].

We briefly sketch out below some relevant background research taking place in "buildings as living labs" [4] with a focus on office buildings, drawn predominantly from the field of Human Building Interaction [3, 26, 47] and some from Adaptive Architecture [31, 61, 62, 67, 71]. The works below address how data can be collected and used in the office buildings to engage and support their occupants’ wellbeing and aspects of the collective life [13, 65]. There are different approaches to data collection; some studies follow an ‘instrumented people’ approach [4]– i.e. using wearable tracking devices– whereas other use passive sensing or deploy interactive prototypes to facilitate data collection and engagement in the buildings. All works seek to deploy their interventions in lived-in offices for longer periods on time [4]; with a view on informing human-centred smart offices. The studies in real lived-in smart buildings are very limited [4]; highlighting the contribution of our work.

A core body of past work focuses on data and technologies for environmental wellbeing, enhancing the experience of microclimate in smart workplace buildings as well as the productivity of employees [6, 16, 18, 38]. Physical comfort is a key relevant term throughout literature, with studies addressing multiple aspects of the perceived indoor climate and developing technologies to support the environmental wellbeing of the building occupants [5, 6, 16, 18] through technologies that facilitate climate data collection and awareness [6, 16]. As an example, recent studies look at the impact of air quality notifications on wellbeing and the collective behaviour in office spaces [21, 74].

Other works have accomplished an understanding of how the design of the built environment impacts wellbeing and social interactions [20, 38, 39, 40, 56, 63]. Researchers have employed different technologies to unpack relationships between group behaviour and physical environments, through passive & active sensing [5, 13, 68]. Some studies have used wearable badges to track social interactions in a workplace [5, 13, 68], to illustrate how the physical design of workplaces supports or inhibits specific types of interactions, how different behavioural profiles thrive in specific spatial configurations [5], and how an organizations’ physical design combines with organizational structure [13].

Researchers have deployed a variety of technologies and artifacts to surface data in office buildings – e.g. physical displays, ambient displays [15, 31, 32] –often embedded in the building’s existing infrastructure – i.e. using interior decorative elements such as room dividers [17, 43, 44], walls and windows [8, 70]– or using objects commonly used in the building – lamps, clocks, lights etc [44]. A relevant example is Yvonne Rogers’ work [56] on data displays for employees’ sedentary behaviour; exploring different ambient and physical displays including lights, screens and pendants as physical displays to communicate physical activity data and the impact on employees’ behaviour [56].

Physical space and its impact on perceived and actual privacy remains a highly vague and complex matter in HCI and Ubicomp research [33, 34, 46, 49, 50, 73], where lies one of the contributions of this work. Reflecting relevant work, Nepal et al [48] have explored the benefits of use and privacy concerns on passive sensing in the workplace; unpacking that the lack of control over the sensor data collection was a significant privacy concern [48]. Other work includes research on changing perceptions on data in the workplace and privacy concerns [25, 52].

The case we describe in this paper is a smart building as a quantified ‘living lab’; a lived-in office, study and research building part of university, developed with the potential of being a research facility – albeit not explicitly stating so. Being highly monitored, the building provides the ground data infrastructure for HBI research and data use applications to take place; while operating as an office/university building. The sensory infrastructure has been developed with regards to building owners’ and building manager’s needs for building operation management, and with a view to be shared for research projects. The paradox is that the most frequent users of the building have little awareness on the data collected and its potential use taking place; and little involvement in any decision making regarding any of the above. This is often the case in smart offices; whereby the employees have no say and often feel ‘monitored’ by the excessive data collection taking place, even if there is little interest in their activities.

Perceived privacy in such buildings remains a highly vague and complex matter in HBI research [73]. Achieving awareness and acceptability of data collection practices from users in lived-in smart buildings remains another open challenge, where lies one of the contributions of this work. Similarly, research on design for wellbeing in quantified buildings remains a field rapidly evolving. Relevant research has informed contemporary building design practice– see WELL ³ and LEED ⁴ certificates; as well as the development of diverse IoT technology to communicate and collectively manage data for wellbeing in offices [REFs]. Still these two dimensions – i.e. the physical design and data integration - do not meet as much as much they could. Open challenges remain around physical design and materiality in smart buildings, wellbeing, and the development of building-integrated technology that empowers their occupants.

Aiming to address the above, our work attempts to study this quantified ‘living-lab’ and unpack the occupants’ experiences, values and needs while inhabiting a smart building. The contribution of our work to HBI research is twofold: a) we contribute to the limited number of studies in lived-in smart office buildings by focusing on their occupants’ experiences, b) based on data collected, we propose a design agenda for the user-centred development of similar smart buildings.

The building were the studies took place is a five-storey modern office building in a city in the UK, which has been advertised – and is widely known - as a smart building. It is a highly sensing and monitoring environment with an extensive array of embedded environmental and occupancy sensors. The building hosts office, teaching, event-space, and research facilities; while the ground floor is an atrium open to public and often serves as event space. Although not explicitly built as a research facility/living lab, the building works as such: it is built to both serve as a real workplace while providing opportunities to conduct research with the data. Data is logged publicly via an API, allowing open access to real-time and historical (timeseries) data of different spaces in the building.

Participants (Tables 1 and 2) were recruited from occupants of a shared workspace based at the office building above. Participants were employees, researchers and students working in the same space. Their familiarity with regards to data and software/hardware varied; none of them could be characterized as an expert, but many of them had relevant knowledge on broader technology and data disciplines – e.g. HCI research. Participants were all aware of what the building monitors, but had very limited previous exposure to, or interaction with, the data produced. Still, participants were not recruited based on their technical knowledge and data familiarity, but based on them being full time workers at the same open plan office space hosted in this building for a period greater than 6 months (minimum was 6 months, greatest was 3 years). In that sense, participants were treated as a community of interest because of their lived-in experience of a smart building. Finally, participants were not incentivised for taking part in the workshops.

The following workshops were conducted during January, February, and April 2020 (Table 1). Workshops lasted between 30 and 60 mins. The first workshop took place physically, while the rest were conducted online using Zoom Software due to COVID-19 lockdown. Each workshop was treated as a separate entity, starting with a presentation framing the ‘smart buildings’ agenda and problem space, and then continuing with a specific angle (as described in the above section). Participants were introduced to the building’s API at workshop 01 and were encouraged to interact with the data. Participants were also encouraged take part in most of the workshops – e.g. many participants in Workshop 2 also participated in Workshop 3.

The workshops were designed as a program of work that would organically evolve over time, responding inductively to participants’ output and expressed concerns and interests. Methodologically, this aligns with a pragmatist orientation to research, eschewing an a-priori determination of experimental variables to be tested. We deliberately chose to have a level of freedom to adapt to changing circumstances - such as the move to remote data collection during COVID-19 lockdowns and the varying interests and engagement levels of the participant pool. Accordingly, participants’ recruitment also evolved as we progressed with the studies. For instance, focus groups had the form of an open discussion driven via a presentation for anyone wishing to attend, without a strict requirement on the number of attendants. Other activities such as the design fiction in Workshops 3,4 had different requirements as the format is easier to manage with a small number of participants; therefore, we kept the overall participant number open but divided them in subgroups of 2-3 participants per group.

Using Zoom as the main communication channel had pros and cons: remote engagement enabled wider participation but limited in the ways people collaborated and interacted with each other. Tools had to be invented to provide the participants with a common spatial context (office) during story-telling activities – such as the development of the 360 toolkit (see Workshops 3, 4). Overall, participants seemed to perform well when working in smaller groups via zoom rooms. Participants also reported that being able to develop their stories at their own time using online material allowed space for in-depth thinking and self-reflection.

All workshops were audio recorded with the participants’ permission - those hosted in Zoom were also video recorded with the participants’ consent. Audio recordings were transcribed by the lead researcher; the answers from google forms were exported in excel format and used as text. Transcriptions of all workshops were coded and subjected to thematic analysis [11]. We chose thematic analysis as the method to analyse collected data, as it allows flexibility and reflexivity when analysing complex qualitative datasets. The data analysis procedure included selecting quotes from the data (codes), performing a group clustering session (first order themes) and then further clustering (second order themes); executed by the first author. A reliability check was performed by the rest of the researchers, and repeating of the clustering process to finalize themes. Data was ‘open coded’ which entailed that data was clustered without a predefined elaborate coding scheme.

Throughout discussions, confusions and concerns over the types of data collected were unveiled. Many participants expressed they are not aware what data is collected, what is the purpose of collection, if this data is personal or not and whether their consent should be given or not."Why do we need to capture all this? […] If you want to get a sense of how the building is then walk around." (P06 / W1). "This building is supposed to be highly monitoring but I’m not sure that I really understand exactly what and how is it monitoring and if we consented to." (P09 / W1). A few participants insisted that their informed consent should be given for all data types that might relate to human activity in the building, even if this data is not inherently personal."Have we actually consented to have all our toilets monitored?" (P02 / W1).

The above statements illustrate existing confusions over which types of data are personal and which are not, and they how they should be treated in relation to giving consent, giving room for more ‘personal’ interpretations of personal data. As the discussion evolved, it became clear that participants wanted a more nuanced agreement over what constituted personal data."I think it’s important to clarify what data (is personal)...for examples I think if you access the admin and say: Give me my personal data from this building, excluding cards etc, they will say: there isn’t any. I think we should be careful about what is what we call personal data" (P07 /W1).

The discussions further explored what forms of data collection are perceived as appropriate and acceptable in shared spaces such as workplace buildings. Data obtained by ambient sensors are mostly perceived as unproblematic by many participants."I think all things (referring to temperature, occupancy, air quality) except for the video recordings feel quite alright, they sound quite anonymous." (P04 / W2). This was contrasted however, with data that was more explicitly tied to individual activity."I am aware of examples where the smart card systems being used by line managers to track what time someone is coming in and out of the building. That’s definitely not acceptable." (P02 / W1).

That being said however, the term ‘passive data’ emerged in some of these discussions, referring to data on secondary effects of human activity. Passive data was mostly related to collecting data after an action or human activity has happened – e.g. using sensory devices or qualitative observation and excluding any human subjects involved. Such data was perceived to be inherently privacy-friendly, highlighting the importance of the temporal dimension of data collection practices. "You could infer the use of taps from the sound of water coming on and off on the pipes, which I thought was a neat way of collecting data without having to actually use any consents" (P02 / W1). "Monitoring secondary effects of human activity can preserve anonymity while providing data about peoples’ preferred ways of using the space." (P15 / W2). A few said passive data can be real time – e.g. data from ambient sensors as suggested above – as long as users are not actively monitored. As this didn’t exclude disclosures that might come from combining different data streams, this view remained questionable among some participants.

Participants’ interpretations of privacy were further influenced by ideas of scale, across a number of dimensions. These dimensions included physical size (from person to workspace to building); the pace and frequency of data collection (e.g. continuous versus sporadic data collection); and form of data processing – e.g. data aggregation, combining data streams etc. These were all ways in which scale of data could be modified and had attendant impacts on perceptions of privacy or intrusiveness.

Thinking first about physical scale, it is evident that the scale of a desk is perceived differently from the scale of a room in terms of privacy and acceptability of data collection practices - e.g. occupancy sensors at each employee’s desk in a shared office space can be privacy threatening, whereas the same data collection method in large meeting room poses less of a threat. "I think if the outcome is having my individual office being monitored that’s different, like, a meeting room. It’s useful to know which meetings are in use and which are not, but then if I’m being tracked in my office, then that’s a different response." (P05 / W1).

Data collection at small scale also requires a different set of interventions due to potential privacy loss from combining different data streams. Participants expressed concerns over compromising privacy as a result of data localization and contextualization – e.g. combining occupancy sensory data with organizational data. "Well it is not personal data here; roughly because there’s, for every given sensor area, there’s maybe five or 10 desks but go up to the sixth floor: almost everyone has one at their office. So, where does that boundary end? What happens if you have four people in an office and some of them only work part time, they become identifiable." (P02 / W1).

Continuous monitoring also raised concerns relevant with the perceived privacy of individuals. It created mixed feelings, with some highlighting the importance to collect as much data as necessary, at the points that are necessary, for the purpose necessary. "It’s more about specific applications of things at places needed - ensuring that you, for example, have a sense of ‘safe lifting’, that’s very different than heart rate monitoring all the times - although both can be used to measure the same thing." (P01 / W1).

Regarding the broader concerns over privacy loss, some participants argued that if the value of data use is distinct and desirable for the individual, it could compensate with potential privacy loss. Others stated that there should not be any trade-off between value and privacy: "I don’t mind giving my personal data away if I’m going to benefit from it." (P05 / W2). "It shouldn’t necessarily be that trade-off between privacy and usefulness. And privacy by design means that you can have high functionality without surrendering your data" (P03 / W2).

Data aggregation came across throughout discussions as a process that appeals to privacy concerns of many participants. Data aggregation in the built environment has different dimensions; participants mentioned spatial dimensions and the use of varied types of data. "In aggregate, we’re not talking just about is it like an individual office versus a large open space where there’s lots of people; That’s one form of aggregation, but what about aggregation and lots of different types of data." (P02 / W1).

Aggregation as a dataset conceptually appeals to privacy concerns of many participants. Aggregated data was seen as privacy - friendly enough to be used in public to raise awareness around collective behaviours and to direct organizational responses. "A lot of it for me is about whether data is aggregated or not. […] If it is aggregated data then, although it might be possible to identify me from that, it takes a significant more amount of effort to do that at the very least." (P01 / W1). "If you do it in a way that is aggregated, so for example, it only signals when heart rates become above a certain level. And they would only transmit this kind of data at the end of the week as an ‘overview statistics’ […] the organizational body could never really track individuals based on that." (P04 / W1).

Many participants expressed the view that although a lot of data is collected in the building, they do not see or feel how it is utilized in any (beneficial) way (data capture effectively takes from them, without giving anything back). For example, the building collects but does not act or respond based on the collected datasets in ways that may improve occupants’ wellbeing and their daily experiences in it. "A very different point but taking that it’s a smart building; I’m actually a bit disappointed, because I feel like they are only measuring but I don’t see any evidence of the data being used for or for maybe for environmental reasons, or well-being. […] It doesn’t sound like or feel like a smart building to me." (P04 / W1).

The points below illustrate the perceived underutilization of collected data, but also the lack of infrastructure that can create awareness on data collected, and meaningful experiences in the building. Mentioning air-quality levels in the building, participants referred to how their senses help them become aware of potential problems, rather than the sensory data. "For example, you can’t really see but you feel the oxygen level drop over the day in the lab, you really feel like you get more tired. If they could do something about that using the sensors, they just pumped some air in or whatever, do something about it, but they don’t." (P04 / W1). "When there was a gas leak, they found out because of its smell not because of the sensors." (P05 / W1).

Focusing on existing visualizations of the buildings’ data (through the associated API), participants mentioned that it is rather hard to read and make sense of. Although the data is openly available, it is not accessible and useable by the average user. As highlighted below, what is missing are the tools to enable users make sense of it and use it in a meaningful way. "It’s just an output more or less of like raw data. And there’s a certain level of interference that’s kind of missing that to make the data useful to somebody who’s not a visualization specialist or something." (P03 / W1).

Participants agreed that surfacing collected data to create awareness could transform their experiences in the building. There were some ideas on how this could be done through data visualizations and physicalizations. Key ideas were surfacing environmental and occupancy data, free movement in the building; and technologies that allow the building occupants to become aware, experience data, and use it as they feel it’s meaningful. "A thing that can be useful is providing ways to surface that information and just letting people use it how they would see it fits. So noise case, for example, if you could easily see, you know, what was currently the quietest space to work, then you could take yourself there […] almost like doing a Google search in the building for noise and then being like, I’ll just go to that space etc." (P01 / W1).

For some participants, simply having more interoperable data can generate value as it becomes easier to analyse and use; for others however, there was a feeling that the value of data capture would be raised by exploring how the building supports the occupants to be smarter, and take collective action. A central idea among proposed use-cases of data was the need to empower the building occupants through giving them more control for example, over using data, as part of a lightweight infrastructure within the buildings. "Our conception of the ‘smart’ building lies entirely in the inbuilt data and information networks that a building contains, with a view to managing its energy use. […] (this) is a limited vision, and that perhaps smartness should come from the occupants, rather than the environment." (P25 / W4).

Participants proposed that processing accumulated data – e.g. as environmental and usage data, movement data – could help users develop awareness about their own behaviours in the buildings and support smart decisions at an individual and collective level on how they use and manage the built environment. "The many measurements that are made in the office space are also synthesised into an app that employees can use to find their favourite places in the office […]one month after the new office space has come into use and the app has collected enough data to inform you about your preferred usage patterns of the building." (P04 / W3).

It was also postulated by participants that there is an inherent long-term value to building data. It was suggested that accumulated datasets could be used for the systematic study of usage patterns in buildings to inform user-centred building design and adaptation, and space management, by directly learning from other occupants’ past and present activities, and the application of collective analytics."[…] when I think of value, I first think how to optimize, but it might be valuable to know how other people have configured their rooms or desks, and where they’ve spent most of the time […] It would be valuable if you could sort of see that it in a different way, and optimize how your space is configured." (P14 / W2).

However, intriguingly it should be remembered, as one participant suggested: "I think, the sort of main argument around smart cities or against smart cities, is that a city is not all about optimization."(P03 / W2). And so thinking about how data can support value-added activity beyond this is valuable. The previous quote suggesting using data not necessarily to optimise space use but to reveal potential for reconfiguration and re-use. It was also clear that our participants also understood there to be a social contract one might enter in to around the collection of data, which was in essence for the greater good, and for which short-term sacrifice of privacy might lend itself to longer term utility. "(about smart offices and energy footprint) It means of you giving out some data because you enter some things in your phone, and probably your home collects some data about your behaviour, but I think if you scale it up, and if you consider this in the long-term perspective, it can have a very valuable impact to society." (P15 / W2).

In many fictional narratives, the perceived value of data was highly linked with supporting health and wellbeing – e.g. buildings as restorative spaces. The concept of environments that learn to adapt to their users over time based on accumulated & processed environmental, occupancy and health data came across in many of participants’ narratives. Other ideas included managing environmental comfort, microclimate and task distribution in the building based on data. "Responsive room […] that can adapt to emotional states based on physiological data and environmental data that is collected. […] It can also do spatial adjustments depending on the number of people in the room." (P24 / W4). "The work environments could measure how much specific desk area was used and whether it had light there […] Maybe there should be a change in how these desks are being used based on data […] to better adapt it based on kinds of tasks and light." (P04 / W3).

The fictional narratives were particularly rich in demonstrating different perceptions of agency and control in adaptive environments, as well as the long-term consequences of AI training on accumulated data. Control is often shifted from the building occupants as the building is now responsible in managing their wellbeing and activities in the building. Sometimes users are aware and are given options to control what the environment is doing, in other cases the changes happen subtly and without awareness. "Adaptive environments can learn over time what settings are most effective e.g. increase temperature or change oxygen concentrations. It can learn what lighting changes are most effective to calm people." (P04 / W3). "The room reads body temperature, heartbeat, pitch of voice in additional to environmental data such room temperature, noise levels and light. Based on this data, the room adapts by offering a range of lighting options (including control of window shades), soothing music when stressed and regulates the room temperature depending if it is too hot or too cold. (P24 / W4)

Conversely, concerns of being controlled or manipulated by the environment based on collected health data were reflected in both utopian and dystopian scenarios- e.g. the building taking over control by imposing specific changes without the occupants’ awareness and desire. "Maybe the room would start to influence us, if we were reading or sleeping too long it would disturb us to do something different." (P26 / W4). "Maybe the room could do more to adapt to our mood and/or activity and support us in what we were doing - partying, reading, sleeping, eating etc. If we were sitting quietly the lights could dim, the walls could become soft in some way. If we were moving around, the lights could be bright, windows and curtains open. The issue might be that the environment starts to control us, or we start acting in an unnatural way to provoke a change in the environment." (P27 / W4).

Finally, concerns on data reliability came across in both discussions and story-telling activities when addressing health in the built environment. "But is data reliable? Is the data valid in terms of measuring what we think it is measuring, and is it meaningful? I’m reluctant to rely too heavily on quantified data without a better understanding of the issues." (P02 / W1). "Does anyone track their sleep? Sometimes you wake up in the morning and you feel like you had a terrible night, you feel really tired and your data tells you slept perfectly well...My boss was coming to my desk and saying you are really stressed…and you are not! Or if you are really suffering and the computer says no." (P10 / W1). "There’s the flip side as well, you go to occupational health because you just feeling really stressed and anxiety, anxious. And the computer says, no you are fine." (P06 / W1).

These concerns also support the view that the use of data should support and not replace human agency, which returns us to the previous section and the argument for using data to support smarter occupants not smart buildings. "I think the danger is if that data is used as a replacement for human interactions which is about care and well-being... they should support and not replace" (P06 / W1).

Our analysis has shown that there is a need to support a more pro-active approach [55] to how the built environment responds to collected data, encouraging actuations to take place and engaging users in experiencing and interacting with that data. Surfacing data at physical scale creates different dynamics [56]; awareness technologies in the buildings [7, 56] can bring latent aspects in the foreground of human experience– e.g. an example is where participants refer to being able to smell, see or feel changes in air quality instead of just collecting data about it [7, 9, 12]. Examples of such data physicalizations for awareness include the recent work on atmospheric interfaces [12] crafting a space for the development of novel ambient interactions around air quality.

More broadly, there are potentials for the design space of physical interfaces such as e-textiles, shape changing, haptics etc. [12, 27, 43, 44], to contribute to a physical and experiential agenda for smart buildings [29], creating applications at variable scales [37] -e.g. furniture, interior spaces, façade systems. Scale can have a greater impact [37, 56] in raising awareness on environmental aspects – e.g. air quality, energy use - as well as aspects of the collective life in the buildings – e.g. use of space. Beyond the many examples of relevant past works [7, 27, 31, 37, 59], there is space for expanding the design agenda through exploring potentials of physical interfaces and interactive materials at scale in the buildings [37, 43, 44]. Apart from physical scale of actuations, temporal scale of data feedback in the built environment has not been adequately explored [10, 51] – e.g. actuations that happen slowly, in an organic way; material changes that display data over time etc.- providing opportunities for feedback that does not monopolize attention, but makes users think [9, 29, 55, 56]. Examples such Organic UIs [43, 44] and slow HCI [9, 51] illustrate some of the attempts towards that direction, reinforcing a smart building agenda where smartness comes from the occupants and facilitating the ways in which they might respond to data, and not from a building’s data management system [42].

Beyond awareness, our findings highlighted the importance of data accessibility and use; the need for interventions that leave space for users to control how the use the data. Participants suggested the development of lightweight tools for open use of collected data by the occupants; shifting the control on data use from the building to its users. Improving data accessibility creates opportunities for levelling control [42] of the use of data, reinforcing an agenda of smart buildings that have potentially ‘DIY’ and ‘open-source’ dimensions. There is therefore room to design for spatial accessibility of data, providing opportunities for building occupants to engage with and use data in the buildings as they think it is meaningful [42].

Summarizing, the value of data for our participants was linked to its ability to be experienced, to be made perceivable (raising awareness of data) and for it to be openly used by occupants as they think it’s valuable. Different forms of data physicalizations and foregrounding of information can encourage a more proactive engagement in the building’s use of data by "turning control to the users" [42, 55], transferring the responsibility and choice of action from the smart building to support smarter occupants [3, 36, 42]. Not only is the accessibility of data highlighted, but the ability of users to meaningfully interact with it [3, 31, 59]. Such engagement techniques can empower occupant’s control in the quantified buildings, while having a tangible and/or visible value and positive impact of the use of data on their daily experiences.

Our analysis unveiled some of the existing confusions regarding data collection, what personal data is in quantified buildings, and whether a consent should be given for collecting data. Participants’ views on personal data varied, illustrating the fact that such environments have many complexities around privacy-friendly data collection. Our analysis also illustrated the underlying discrepancy between the perceived levels of privacy and the actual privacy [34, 50]. Reflecting the work of Nissenbaum on privacy as contextual integrity [50], our work illustrates that physical scale and architectural factors impact perceived privacy [46] and determine what forms of data collection and data types are perceived as acceptable [25, 46]. Moreover, it advocates that physical scale can potentially have an impact in actual privacy, illustrating the mismatch between current policy and the realities of lived-in smart buildings [52].

Besides physical scale, proportionality and scale (e.g. mass) of data collection was addressed as a privacy-determining factor -e.g. requests for data minimization – which is also left vaguely interpretable. The idea of the continuous monitoring and the massive datasets that are produced were related with potential loss of privacy. Participants addressed the importance of only collecting as much data as necessary, at the points that are necessary, for the necessary purpose, highlighting the importance of application - specific collection of data [34].

Past work on privacy and data collection in smart environments illustrates some of these problems [50]. As a result of the mismatch between physical and digital boundaries in quantified buildings, privacy is perceived as fluid in quantified buildings, and in a dynamic relation to architectural and social factors [50]. Perceived privacy is a proposition to be negotiated and contested, evolving, and changing throughout spatiotemporal scales. Any user-centred design solutions that are developed in that space therefore must consider the effects of spatiotemporal relationships on perceived privacy. This urging for a more architectural and physical design for privacy that promotes embodied awareness [23, 49, 52] and tangible control mechanisms [2].

Of note, GDPR ⁵ regulations (of critical importance to those of us living and researching in Europe) do not address any of the above concerns; principles such as the meaningful consent, data aggregation, data minimization and anonymization as mentioned in GDPR are general guidelines, with their application in the built environment creating complexities. GDPR cannot take spatiotemporal relationships and physical boundaries into account. This also leaves space for concerns and misinterpretations. There is also a difficulty of nuance, with individuals having variable and contextualised understandings of acceptability over matters of privacy and the use of personal data, and this creates extensive difficulties in translating privacy into a global building GDPR policy or architecture. These difficulties are inherent to the sphere of privacy in quantified buildings and therefore encourage design-led and context-specific interventions.

Our work illustrates that users have concrete expectations regarding how buildings function that have been well shaped by the years of making and inhabiting them [59]; and diverse expectations on what and how data in the buildings should or should not be used in the context [46, 50]. Data in buildings can be perceived as yet another ‘shearing’ layer [59] that represents a complex space of activity within a building, but it is not commonly seen as an inherent infrastructural ‘utility’ layer such as electricity, ventilation or plumbing. Thinking about data and occupants’ experience of it as another layer of building fabric, and as a physical/material element in buildings (e.g. data materialities)[2, 7, 9] could potentially shift some of the problems around perceived data privacy and acceptability. In that sense, having a GDPR-first approach when addressing data in buildings can lead to a deadlock, whereas prioritising data materialization and usability through user engagement can potentially allow different perceptions to evolve.

Overall, there is a need to clarify and inform the regulatory framing of personal data and privacy within the quantified built environment, and the processes through these regulations are applied in each context, prioritising user engagement and data use. Within that space, there is potential for design research in quantified workplaces to accommodate a physical design that allows negotiating privacy in different levels; designing technologies that allow a more ‘dynamic, user-centred and tangible privacy’ [2]. Moving forward towards ‘privacy -by- design’ quantified environments [34], buildings should prioritize making occupants aware of data collection taking place and allow them to ‘level their privacy’ – e.g. filtering the sharing of data depending on where they are (physical scale and spatial attributes) and the tasks they are performing (e.g. uses of space). This will likely require the development of appropriate socio-digital-architectural design patterns – which can be easily deployed by both building designers and building managers. Such design interventions should enhance both the perceived and actual privacy of the building occupants. Examples of such artifacts could be interactive room dividers that help users control and customize both physical and digital boundaries at the workplace; that enhance awareness on data collection through physical feedback – e.g. material colour change, light notifications - and provide tangible ways for users to control the amount and pace of data collected in the building’s spaces they inhabit – e.g. touch to control pace of data collection, data accuracy, or appropriate data streams.

The value of accumulated data in the long-term was related to data on collective use patterns of the built environment, used to inform present and future building design and management. As participants mentioned, the study of intuitive spatial use patterns can be systematized through data collection and inform future designs of adaptive architectures [47]. Data accumulation can provide insights and design directions based on the collective, ‘informal’ and ‘unplanned’ uses of space and the ‘organic’, ‘intuitive’ and ‘slow’ adaptation processes that happen in the built environment by its occupants’ activities [30, 51, 59], providing novel solutions to contemporary problems about how buildings can be better designed and built for user-centred patterns of use.

Collective analytics – i.e. open datasets produced through collective engagement in the buildings that are openly available for collective use – can provide new insights on how ‘smartness’ can be driven by the building’s occupants and not the building’s infrastructure. Surfacing’ or ‘layering’ these datasets through lightweight infrastructure that allow collective control on data use could radically change how the building occupants perceive and use quantified buildings. The process could be described as a passive long-term adaptation of both people and environments [59] – a symbiotic process of co-adaptation - based on iterative interactions with data in the prospect of serving broader societal and environmental values, such as reducing the buildings’ energy footprint [16]. Critical to this however is determining who has the right to set the agenda for those societal and environmental goals to which the built environment is then shaping human behaviour.

At the moment there is very little consideration of how collective behaviour can actually shape the ‘agenda’ for a smart building – something we should perhaps explore much further and which might fruitfully bring the research areas of ‘Digital Civics’ and HBI together.

Occupants expressed conflicting insights regarding control and use of data for health and wellbeing from the built environment. Occupants discussed the undesirable effects in adaptive environments whereby their health and behavior is managed - and in some cases controlled – by AI, with or without their awareness, giving buildings themselves a new kind of agency. Although participants expressed that a level of automation in the buildings’ response for wellbeing purposes would be desirable, awareness and the levelling of control of AI actions are again key considerations. As a few suggested, data should support and not replace any decisions around health and wellbeing in quantified buildings; pointing towards a more data-for-awareness approach, and transparency in the choice-of-action.

The above conflicts between the perceived value of data use, agency and AI control in the built environment can inspire researchers, architects and developers towards re-thinking and designing for human-driven AI in quantified buildings with a focus on health and wellbeing. Reflecting on the work of Urquhart et al., Schnädelbach et al., and Jäger et al. [67] on adaptative architecture, and the idea of human-building-collaboration by Rosenberg & Tsamis [58], the future of intelligent environments lies in developing interactive applications that embrace a full-body interaction with a building – i.e. using different forms of embodied actuations - and establish a collaboration between AI in the buildings and their users through constant feedback iterations [58].

Quantified environments could develop intelligent behaviours based on long-term iterative interactions -e.g. feedback loops - with their users to support health and wellbeing, whilst allowing users to control data use and actions initiated by the environments. Extending this to health and wellbeing agenda, the design agenda of restorative environments [65] has potentials to obtain a more human-AI collaborative approach, whilst enhancing behaviour awareness to empower human agency.

By exploring with building occupants their experiences of, and expectations around, a ‘quantified’ smart building, we have brought to the fore a number of tensions and concerns, in particular around privacy and data awareness/use, and shown how they can be nuanced in relation to spatiotemporal issues. At various points in the discussion above we have articulated some potential areas for further inquiry, some challenges that are then raised by these experiences which might help us set the agenda for future research in this space. Of course, some work has already been done to highlight agenda for HBI [3, 31, 60, 62, 63, 67], but here with a specific focus on lived-in quantified buildings and design interventions, there are several observable areas of future work which could be fruitfully explored.

Through these various proposed strands of research agenda – a potential new research landscape emerges that helps to better define an occupant-centred approach to smart buildings. Taking such an approach even in this limited study has already started to unpack some of the complexities around how issues of privacy might need to be dealt with in future and the challenges around supporting occupants’ desired sense of agency in the face of automated environments. With a renewed call for exploring the human-experience of smart buildings there is much potential for developing new, engaging and ultimately occupant-centred data-interactions within the built environment, and hopefully this work helps to scaffold future work in this area.