Automated decision making (ADM) systems are in the center of smart home technologies. With the recent advancements within the fields of artificial intelligence (AI) and internet of things (IoT), ADM systems are becoming increasingly smart and arguably more comfortable and accurate in their automated decision making. However, there are still questions regarding the design of human-machine collaboration in the home. How exactly do users engage with smart technologies in their homes? How do they maintain control over their systems when something does not work as expected, and how can we design for co-performance between users and AI? Our research investigates such questions in the area of user engagement in automated decision making (ADM) systems in smart homes. In this paper, we outline first ifndings and perspectives from an ongoing literature study on this topic, which we intend to investigate further in our future work.
A 1Fraunhofer-Institute for Applied Information Technology FIT,Schloss Birlinghoven,Sankt Augustin,Germany,53757
“Looking through history, we saw the enlightenment of sages, who think that one day machines can do most of the things in replace of humans, free of your mind, free of your hands, no thinking, no action, all the best is prepared for you.”
Automated decision making (ADM) systems are improving thanks to the rapid
development of artificial intelligence (AI) and internet of things (IoT) technologies. Smart homes with
automated decision making systems promise to make homes smarter, more intelligent, and
energy-eficient at the same time. The definition of smart home comes from Intertek in
September 2003. More specifically according to their DTI Smart Homes Project: a smart home is “a
dwelling incorporating a communications network that connects the key electrical appliances
and services, and allows them to be remotely controlled, monitored or accessed” [
Working towards this topic, we are interested to study how users engage with ADMs in their smart homes. How do users set up and interact with their smart home technologies, how do they respond to potential unwanted behaviors or fine-tune the systems to their individual and situated needs? By better understanding these practices, we hope to inform the design of ADM systems that allow co-performance between users and AI. This is especially important for the question on how to design ADM systems that are able to deal with conflicts of interests, such as how to navigate the fields of tension between automation and being in control, as well as between the need to preserve energy and provide comfortable living environments.
In this position paper, we outline the first stage of our research idea as a brief literature review on algorithmic decision making in the smart home, and discuss in which areas we see potentials for designing for co-performance between ADM systems and their users in diferent scenarios.
There are at least three roles to consider in the design of automated decision-making systems:
the designers, the users, and the afected persons. Designers are the ones who devise the
automated decision-making systems [
Understanding diferent kinds of roles involved in designing automated decision-making
systems is very crucial. It solidifies the foundation of designing human-centered automated
decision-making systems. For instance, designers usually have a good understanding of the
technological basis, but need to learn about the needs of users in order to build systems that
are successful in practice, calling for user-centered design processes [
In this position paper, we focus on the users as both the persons who interact with the automated decision-making system or are exposed to their decisions. We try to understand their needs based on diferent smart home application areas, analysing what problems might exist in the interaction and how to design a well-fit automated decision-making system in smart home that allows co-performance between AI and users. By doing so, we intend to put the human in the loop of ADM systems.
In the first phase, we investigate the application areas of ADM systems in smart homes. Based on a literature review, we identified several areas where ADM systems play a role. In this position paper, we only focus on three areas - eHealthcare, energy management, and entertainment. While we can foresee that there will be more “playgrounds” for ADM systems in smart homes in the near future, these are currently the most prominent application areas.
With the increased cost of healthcare of aging societies and the development of IoT, eHealthcare
has become a major service that smart homes can provide. The concept of Ambient Assisted
Living (AAL) has become a major research topic in the last decades [
This data is then analysed by automated decision making systems to provide diagnostic information, call help when needed, or provide health monitoring data for physicians.
ADM systems in eHealthcare so far usually provide little means for user engagement [14]. They are often setup by experts and provide largely passive functions such as disease diagnosis [15], emergency notification [ 16] and health monitoring [17] for the user, but without much means for user control. However, health situations of elderly people can be very diverse, and there is a huge potential in giving users more control over their ADM systems in order to address sensitive issues (such as privacy related to health data or interaction with potentially distressing diagnosis decisions), as well as allowing for more fine-tuned and situational support in complex situations (i.e., disease recovery).
Due to the low level of user control, there exists potential room to think towards higher user engagement, providing more interactive interface and making decisions more transparent. In turn, these will increase users’ trust and allow for more complex interactions with automated decisions[18]. However, there are also challenges due to the safety-critical aspect of those systems, that call for special support approaches and careful considerations.
For energy management, ADM systems are used to minimize the cost of energy consumption and maximize the comfort of inhabitants. This can happen on two levels: the first level refers to automated decisions based on the smart grid (e.g. starting appliances automatically when there is a low energy demand on the overall grid). The second level is based on sensor information from the smart home energy management itself (i.e. lower the heating when no user is present at home).
Data are also exchanged between these two levels, to identify behaviour patterns for grid optimization. Due to the challenges of the climate crisis, energy saving is often in the focus of such systems, though they also provide comfort to the users in terms of automating things that have previously been needed to be done manually (such as turning of thermostats when leaving the house)[19].
In energy management, ADM systems usually ofer a rather high level of user engagement, allowing (or requiring) to set up complex rules and time schedules for the operation of the smart home. This is required because the energy demand is usually varied person to person, and living situations will change over time. Furthermore, a lot of the energy demand is strongly related to the user behavior. Therefore, approaches such as providing eco-feedback, gamification and nudging are commonly considered here. As smart grid based automation (such as starting the dryer at night when energy is cheap) might interfere with individual living situations and requires user engagement, those systems will usually provide diferent usage scenarios for users to select (i.e., optimizing for comfort, saving energy, or security)[20].
While Energy Management ofers a high level of user engagement, we see a high potential for designing for co-performance due to the inherent fields of tension between energy saving and user comfort. Finding the sweet spot between diferent aims is dificult, and also a system which requires a high level of user engagement and interaction can be more challenging to set up(if it provides high levels of control) or rids of the user of control (in case of full automation). This especially afect users that live with these systems and are afected by them, but don’t operated the systems themselves.
Another area of ADM systems in the smart homes is related to entertainment and digital consumption. Here, automated decisions mostly refer not only to the recommendation of digital media (such as music streaming to smart speakers or movies on smart TVs), but also to other comfort functions such as providing easy access to the Internet based on voice interaction with smart Voice Assistants such as Amazon’s Alexa, Google Assistant or Apple’s Siri [21]. These ADM systems perform based on analysis the user preferences about digital media, as well as patterns that are detected in the preferences of the overall user base of the service providers (such as ‘customers who liked this, also liked that’). Having good recommendation mechanisms has turned out as a major competitive advantage of the digital economy (next to having control over platforms themselves) [22].
In the area of entertainment, users are passively engaged in the system. The level of user engagement is low. Users can provide feedback to the systems by “disliking” media or marking automated suggestions as “irrelevant”, but there does not exist ways to control the system besides that.
It seems that users are engaged in the systems all the time, but ADM systems remain transparent in the background and users don’t interact with the ADMs directly. Decisions in this area are highly related to the concept of taste [22], which can be a highly complex phenomenon. Under the such situation, we see a potential opportunity for future designs in it, especially involving the voice assistant into it. In terms of interaction with Voice Assistants, there is also a lot of potential for designing for a better co-performance using Voice Assistant. We can make a reference to other ADMs application areas with voice assistant[23], which has already established in practice and provide a lot of functionality in terms of user interaction and feedback.
In our study we have found diferent levels of user engagement and automation at smart homes across the diferent domains. As we have outlined, there are a lot of open issues to be solved, and also a lot of research opportunities in the diferent areas, which are often also entangled in practice–smart homes intend to provide safety, comfort, entertainment, and also means to preserve energy at the same time. Due to the high variance of individual needs, life situations, and the high complexity of the problem areas, we agree with the concern that there will be no fully automated solution in the sense of a “one-size-fits-all” approach, but we will need to consider the human in the loop for automated decision making [24].
In our future work, we would like to study the diferent domains in more detail, understanding how users appropriate such systems in practice, how they interact with each other, and how we can improve those interactions in terms of giving users more control without compromising on their comfort of living. This will not only require to make systems more transparent in terms of their decision making [25], but also to provide the means to users to co-perform with their home ADM systems. Hence, this will provide them the power and means to adapt a system to their individual needs and living situations without giving up their aim to preserve energy and contribute to the combat against the ongoing climate crisis. By studying touch points, interfaces, practices and fields of tensions between competing interests, we hope to inform the design of more adaptable ADM systems, support appropriation and empowerment of users of such systems in various domains.
In doing so, we intend to experiment with diferent forms of interfaces, provide diferent entrances to the ADM system configuration that are also suitable for less engaged-users. This will provide them a feeling that they are living with ADM systems but (so far) have not actively engaged with them. We would like calling for “smart assistants” that provide a high level of control but also good usability and appropriation support for the users that engage with them [26]. Our assumption here is that as individual personal assistants (IPAs) already provide a very low level entrance to ADM systems, it can also be exploited to provide users with higher levels of control. Therefore, we assume that IPA can be used as bridge to connect the human and machine in terms of supporting ADM appropriation and control, which we want to study further in our future work.
This work has been funded by the EU Horizon 2020 Marie Skłodowska-Curie International Training Network GECKO, Grant number 955422 (https://gecko-project.eu/). home ehealth, in: Proceedings of the 11th European Conference on Software Architecture: Companion Proceedings, 2017, pp. 102–108. doi:10.1145/3129790.3129801. [13] N. Agoulmine, M. J. Deen, J.-S. Lee, M. Meyyappan, U-health smart home, IEEE Nanotechnology Magazine 5 (2011) 6–11. doi:10.1109/MNANO.2011.941951. [14] L. Rundo, R. Pirrone, S. Vitabile, E. Sala, O. Gambino, Recent advances of hci in decisionmaking tasks for optimized clinical workflows and precision medicine, Journal of biomedical informatics 108 (2020) 103479. URL: https://doi.org/10.1016/j.jbi.2020.103479. [15] P. Khan, M. F. Kader, S. R. Islam, A. B. Rahman, M. S. Kamal, M. U. Toha, K.-S. Kwak, Machine learning and deep learning approaches for brain disease diagnosis: Principles and recent advances, IEEE Access 9 (2021) 37622–37655. doi:10.1109/ACCESS.2021.3062484. [16] S. Bolourani, M. Brenner, P. Wang, T. McGinn, J. S. Hirsch, D. Barnaby, T. P. Zanos, N. C.-. R.
Consortium, et al., A machine learning prediction model of respiratory failure within 48 hours of patient admission for covid-19: model development and validation, Journal of medical Internet research 23 (2021) e24246. doi:10.2196/24246. [17] H. F. Nweke, Y. W. Teh, G. Mujtaba, M. A. Al-Garadi, Data fusion and multiple classifier systems for human activity detection and health monitoring: Review and open research directions, Information Fusion 46 (2019) 147–170. URL: https://doi.org/10.1016/j.inffus. 2018.06.002. [18] K. Shailaja, B. Seetharamulu, M. Jabbar, Machine learning in healthcare: A review, in: 2018 Second international conference on electronics, communication and aerospace technology (ICECA), IEEE, 2018, pp. 910–914. doi:10.1109/ICECA.2018.8474918. [19] N. Castelli, A. F. Pinatti de Carvalho, N. Vitt, S. Taugerbeck, D. Randall, P. Tolmie, G. Stevens, V. Wulf, On technology-assisted energy saving: challenges of digital plumbing in industrial settings, Human–Computer Interaction (2021) 1–29. URL: https://doi.org/10.1080/07370024. 2020.1855589. [20] M. C. Bozchalui, S. A. Hashmi, H. Hassen, C. A. Canizares, K. Bhattacharya, Optimal operation of residential energy hubs in smart grids, IEEE Transactions on Smart Grid 3 (2012) 1755–1766. doi:10.1109/TSG.2012.2212032. [21] A. de Barcelos Silva, M. M. Gomes, C. A. da Costa, R. da Rosa Righi, J. L. V. Barbosa, G. Pessin, G. De Doncker, G. Federizzi, Intelligent personal assistants: A systematic literature review, Expert Systems with Applications 147 (2020) 113193. URL: https://doi. org/10.1016/j.eswa.2020.113193. [22] J. W. Morris, Curation by code: Infomediaries and the data mining of taste, European journal of cultural studies 18 (2015) 446–463. URL: https://doi.org/10.1177/1367549415577387. [23] F. Alizadeh, G. Stevens, M. Esau, I don’t know, is ai also used in airbags?, i-com 20 (2021) 3–17. URL: https://doi.org/10.1515/icom-2021-0009. [24] B. Wagner, Liable, but not in control? ensuring meaningful human agency in automated decision-making systems, Policy & Internet 11 (2019) 104–122. URL: https://doi.org/10. 1002/poi3.198. [25] https://www2.deloitte.com/, Transparency and responsibility in artificial intelligence, 2019. URL: https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/innovatie/ deloitte-nl-innovation-bringing-transparency-and-ethics-into-ai.pdf. [26] D. Pins, T. Jakobi, A. Boden, F. Alizadeh, V. Wulf, Alexa, we need to talk: A data literacy approach on voice assistants, in: Designing Interactive Systems Conference 2021, 2021,