Who benefits? A scoping review and taxonomy of human-plant interactions in urban environments

1 Introduction

More than half of the world's population now lives in cities, a trend that is projected to intensify (United Nations, Department of Economic and Social Affairs, Population Division, 2025). As urban environments constitute the primary backdrop of daily life, direct contact between urban residents and nature continues to decline (Pyle, 1978; Soga and Gaston, 2016; Gaston and Soga, 2020). At the same time, the integration of sensing, wireless communication, and intelligent systems is transforming built environments from passive physical containers into interactive partners capable of supporting human needs and promoting health and wellbeing (Becerik-Gerber et al., 2022). Against this backdrop, human-building interaction (HBI) has emerged as an interdisciplinary field at the intersection of architecture, building science, ubiquitous computing, and interaction design, reframing buildings as immersive, interactive systems that shape and are shaped by the people who occupy them, rather than as static backdrops or passive enclosures (Alavi et al., 2019; Garcia-Hansen et al., 2026). Within this framework, interaction is characterised as a “spatiotemporally immersive” phenomenon that persists over time and unfolds across different adaptive timescales, rather than being confined to a single moment (Alavi et al., 2019). Relatedly, Kirsh (2019) observes that architects tend to view occupants as active agents who continually reshape the interface through use. Interaction in built environments is thus an ongoing process, co-shaped by multiple agents. This brings into focus how specific forms of non-human life, particularly plants, are being incorporated as active agents in ongoing interaction processes within built environments.

Among the various forms of life, plants are one of the few living organisms that have been introduced into built environments at scale. They slowly but continuously grow, respond, and change in their environment, rather than remaining in a fixed state. When technology intervenes to augment their natural capacities, plants can be transformed into interactive beings that reduce anxiety and stress (Park et al., 2016), support capacitive gesture interfaces (Poupyrev et al., 2012). In experimental systems, plants have further been augmented through in-planta conductive materials for sensing and actuation (Sareen and Maes, 2019), or coupled with robotic systems in bio-hybrid architectural prototypes (Hamann et al., 2015). With advances in sensing and actuating technologies, plants are increasingly regarded as collaborative beings rather than passive materials in interactive systems (Loh et al., 2024). The living characteristics of plants, including continuous growth and dynamic responsiveness, structurally resonate with the long-term, multi-agent, and cross-temporal interaction structures emphasised in HBI.

Rather than abstract biological entities, plants are often integrated into everyday built environments through specific technological interventions. In human–nature engagement,

“design choices impact knowledge, attitudes and behaviour” (Webber et al., 2023).

These design choices are embedded in sensors, actuators, and algorithms, which shape what is to be perceived and responded to, and who is served. Technology is not merely a tool for interaction; it also shapes the relational structure between humans and plants (Graham et al., 2024; Harding et al., 2026). Recent scholarship has raised ethical questions from a design perspective: are plants collaborators or materials in design, and how much agency should be preserved for them? (Fell et al., 2022). These ethical inquiries also open space for analysis at the operational level. When plants are incorporated into interactive systems within built environments—which are typically not their natural environments—further questions arise: what does each party gain from the interaction? How are benefits distributed? To what extent does the designer retain control over the direction of benefits, and to what extent is decision-making ceded to the physiological processes of the plant?

Existing HBI research offers only a partial response to these questions. Prior work has primarily focused on quantifiable human benefits, including improvements in air quality (Deng and Deng, 2018), psychological and physiological stress reduction (Lee et al., 2015), enhanced work productivity (Nieuwenhuis et al., 2014), and the effects of architectural form on physiological stress responses (Valentine, 2024). The prevailing tendency is to treat plants as instruments in the service of building performance. This orientation is not unique to HBI. (Chang et al. 2022, p. 943) note that design research involving plants has long been shaped by Western anthropocentric traditions and call on researchers to “transcend imitation or parasitism, and move towards respectful, longer-term partnerships with the untapped potential of nature.” Understanding the full range of roles that plants can assume requires looking beyond the mainstream concerns of HBI. When the design perspective expands from unilateral human convenience to multi-species coexistence, the design of built environments needs to consider not only how to serve human users but also how to welcome plant species. To achieve this, HBI researchers can draw on the work of researchers, designers, and artists from other disciplines. Compared with applied HBI research, artistic, creative and design practices are less constrained by functional application requirements and often address boundary conditions in exploratory ways, ranging from unidirectional human benefit and bidirectional human-plant coexistence to extending benefits towards broader ecosystems. This paper takes the transferable insights of such practices as its point of departure. By reviewing these practices, it provides a reference for HBI researchers and practitioners to reflect on the role of plants and how benefits are distributed in human-plant interactions within built environments. Importantly, this focus on benefits should not be read as an assumption that such interactions are inherently positive. Many interventions may entail neutral or adverse effects for plants, including stress, damage, or death. Our emphasis on benefits is therefore analytical rather than evaluative, offering a tractable lens to examine how value is framed, attributed, and distributed across human and more-than-human stakeholders.

The paper addresses a central question: who benefits from technology-mediated human-plant interaction, and how are these benefits distributed among humans, plants, and broader urban ecosystems? It makes a twofold contribution: (i) a scoping review of 50 works on technology-mediated human-plant interaction in built environments; and (ii) a Beneficiary Taxonomy that conceptualises patterns of benefit distribution. Drawing on the analysis of this corpus, the taxonomy identifies five recurring types of benefit flow: Restoring, Nurturing, Autonomising, Entangling, and Extending. Rather than evaluating which orientation is preferable, it serves as a heuristic to identify patterns across cases and foster reflection. In doing so, the paper contributes an analytical dimension focused on the direction and distribution of benefits, offering a starting point for cross-disciplinary dialogue in the field that may lay the groundwork for future work that more systematically accounts for harms, trade-offs, and ethical tensions (Chang et al., 2022; Fell et al., 2022). This analysis focuses on the benefit orientation each work is designed to prioritise—expressed through its design, aims, and interaction structure—rather than the cognitive or emotional reactions audiences may form, and it situates this orientation within the spatial settings in which each work operates. The paper is concerned not only with human-plant interaction in general, but with how benefit flows are shaped by the spatial settings—galleries, domestic spaces, workplaces, public spaces, and community or ecological infrastructures—in which each work is situated.

The remainder of the paper is structured as follows. Section 2 reviews related work on technology-mediated human–plant interaction and the theoretical foundations informing design orientations in this field. Section 3 outlines our approach to scoping case collection and taxonomy construction. Section 4 presents the Beneficiary Taxonomy and analyses its five categories: Restoring, Nurturing, Autonomising, Entangling, and Extending. Section 5 discusses the implications of the taxonomy for reciprocal human–plant relationships and the alignment between design intent and benefit flow. Section 6 concludes the paper, including limitations and directions for future research.

2 Related works

This section situates the study within existing scholarship on technology-mediated human–plant interaction, and identifies a gap in understanding how benefits are distributed among humans, plants, and broader urban ecosystems. As a Review article, our paper combines conceptual positioning with a structured scoping review; this section therefore outlines key strands of work that frame the research problem, rather than presenting the full review corpus. Section 2.1 surveys representative practices across interaction design, media art, and architecture, while Section 2.2 examines the theoretical foundations and design orientations that inform these works.

2.1 Technology-mediated human–plant interaction practices

Built environments are

“man-made or modified structures that provide people with living, working, and recreational spaces,” (US EPA, 2017)

and the infusion of technology is changing how people live, work, socialise, and learn within them. HBI spans a broad range of concerns, including health and wellbeing, education, climate-responsive design, and the development of future-oriented built environments (Garcia-Hansen et al., 2026). Examples include smart homes (Peters et al., 2011; De Russis et al., 2013; Strengers et al., 2022), indoor environmental quality monitoring (Moore et al., 2018), interactive building surfaces (Bader et al., 2019), and assistive technologies (Carabalona et al., 2010; de la Guía et al., 2015). While this body of work is diverse in scope, it largely frames interaction in relation to human users, with built environments positioned as responsive systems that support comfort, performance, and everyday activities. Within this predominantly human-centred framing, plants—despite being among the most widely present living organisms in built environments—have rarely been considered as independent interactive participants.

When plants find suitable conditions within a built environment, they not only sustain themselves but can also help humans engage more actively with the buildings they inhabit. More broadly, technology-mediated human-plant interaction has emerged as an interdisciplinary area spanning interaction design, architecture, media architecture, media art, and bioengineering, encompassing a wide range of experimental and applied practices. Adjacent work on citizen science and urban forageable plants further shows how digital platforms can mediate public awareness, documentation, and stewardship of urban plant life, even when such work falls outside the interactive installation and creative-practice focus of the present review (Soares et al., 2024a,b).

Within this emerging field, early systems such as Infotropism (Holstius et al., 2004), Bio-Fi (Easterly, 2004), and PlantDisplay (Kuribayashi and Wakita, 2006) shared a common starting point: validating the feasibility of digital technology as a cross-species mediating layer. These systems explored different directions of information flow between plants and humans. Some translated plant bioelectric signals into human-perceptible outputs (Kuribayashi et al., 2007; Asaf et al., 2021), while others used plants as ambient display media actuated by external digital information (Easterly, 2004; Kuribayashi and Wakita, 2006) or used biomimetic phototropic robots to guide pro-environmental human behaviour (Holstius et al., 2004). Under this paradigm, the value of plants was primarily understood in terms of serving human users as information carriers or emotional regulation media (Hwang et al., 2010; Kawakami et al., 2010; Cho et al., 2015; Park et al., 2016; Angelini et al., 2016; Botros et al., 2016; Steiner et al., 2017).

Subsequently, designers began to reshape the role of plants in interactive systems by moving from instrumental uses towards more integrated and participatory forms of interaction. This shift is also reflected in work within media architecture that explores multispecies frameworks aimed at amplifying plant agency (Sheikh et al., 2021). Capacitive sensing technology has transformed living plants into multitouch interfaces (Poupyrev et al., 2012), while multimodal augmentation and data visualisation have enabled their use as information displays (Hammerschmidt et al., 2015; Wen, 2025). In-planta cybernetic systems—constructed by embedding conductive polymers within plant tissue—allow plants to function simultaneously as sensors, displays, and actuators (Sareen and Maes, 2019). Systems that drive plant movement to provide ambient feedback for digital media (Degraen et al., 2020), alongside design frameworks for augmented human-flora interaction (Seow et al., 2022), further demonstrate the potential of plants as active media for interaction.

Beyond mediating human-plant encounters, digital technologies can also support broader human–nature relations through post-anthropocentric design strategies (Webber et al., 2023; Kneile et al., 2025). More embodied forms of interaction include designs that regulate user emotion through soft-robotic tactile surfaces rendering plant biosignals via expressive movement (Sabinson et al., 2021), as well as strategies that assign anthropomorphic traits to plants to enhance human empathy (Bhardwaj and Joseph, 2020). These examples illustrate the diverse values of plant participation in interaction.

Recent studies further expand methodological diversity. Hansen and Rolighed (2024) show how interaction design can cultivate hermeneutic, existential, and socio-cultural sensibilities towards plants. The use of supervised learning to analyse plant biosignals and translate them into dynamic audiovisual expressions has been framed as reading the environment “from the perspective of plants,” offering an alternative to anthropocentric modes of perception (Hu et al., 2023). Hu et al. (2024) propose a design framework for plant-driven robotic actuators by integrating the properties of plant growth, ageing, and decay. In the field of interactive art, researchers have begun to examine interaction structures between human and non-human organisms through a relational lens (Xu et al., 2025; Wu et al., 2025).

At the same time, scientific findings on plant perception and communication are providing new epistemological foundations for technology design. Plants are not passive entities but living organisms with capacities for perception, communication, and environmental response (Mancuso, 2018). The concept of intelligence is expanding from a narrowly human attribute to a quality recognised across a broader range of living organisms (Trewavas, 2003). Gagliano et al. (2017) re-examine the expressive capacities of plants from linguistic and philosophical perspectives, offering a cross-disciplinary lens for understanding plant communication. At the practical level, decentring design theory encourages designers to incorporate non-human participants into collaborative frameworks (Forlano, 2016; Wakkary, 2021; Heitlinger et al., 2024), while plant-robot hybrid architectural systems use distributed robots to guide directional plant growth, building living structures with self-repair and self-organisation capacities (Hamann et al., 2015).

In terms of systematic classification, Chang et al. (2022) survey human–plant interaction works and analyse how plant signals, sensing inputs, and outputs are coupled within interactive systems, offering an effective entry point for mapping the technical implementation pathways across different works. However, this classification primarily focuses on the dimension of technical implementation, that is, how systems are constructed, how plant signals are coupled, and how interfaces are manipulated. Yet, the majority of projects to date have—intentionally or otherwise—treated plants as utilitarian objects in the service of human needs, deploying them as proxies for nature, triggers for human experiences, and interfaces for other actions (Loh et al., 2024). Technology amplifies the relational potential of human-plant encounters, but amplification alone does not entail an equitable distribution of benefit (Fell et al., 2022; Loh et al., 2024). When plants in built environments are no longer treated as ancillary tools for building performance (Loh et al., 2020) but recognised as living participants with needs of their own, questions of benefit distribution become central to the design of human-plant interaction. Every design implicitly takes a position on “for whom” such interactions are configured. This shift calls for a more explicit conceptual framing of human–plant relations, which is explored in the following section.

2.2 Theoretical foundations and design orientations of human–plant interaction

Plants continue to attract research attention in interaction design, in part because of the possibilities they carry. On the one hand, plants as natural sensing and responsive systems provide a material basis for building more sustainable technologies. On the other hand, design practices involving plants inherently hold the potential for rebuilding the relationships between humans and nature. However, as the previous section indicates, this potential remains largely unidirectional. Technology can amplify the relational potential of human-plant encounters, but amplification alone does not automatically redirect the flow of benefits, which requires consideration of the cognitive frameworks inherited by design practice. By examining these frameworks, we seek to understand how the distribution of benefits might be reconsidered in human–plant interactions.

In design practice, plants have long been treated as functional materials in the service of human goals, with their own needs and preferences rarely receiving independent attention. If plants are to be regarded as equal participants in interaction, researchers first need to understand plants as lifeworlds with their own interests and agency (Aspling et al., 2016; Borthwick et al., 2022; Kohn, 2013; Jensen et al., 2025). Design practices that prioritise human comfort and convenience, in a context where non-human entities continue to be depleted and driven to extinction, have in effect exacerbated threats such as climate change and species loss (Wakkary, 2021; Sheikh et al., 2023c). Reconsidering the position of non-human entities in design has become an increasingly active topic in design research and urban studies (Fieuw et al., 2022; Yoo et al., 2023; Eriksson et al., 2024; Fredericks et al., 2026).

Entanglement theory provides an ontological basis for this shift. It holds that humans and non-humans are inseparable from the outset, and that agents can only constitute themselves through interaction with other entities (Frauenberger, 2019). This perspective elevates non-humans from passive background to active participants in relational processes and offers grounds for rethinking the distribution of benefits in design. Foregrounding plants in artistic creation can itself evoke attention to their subjectivity, even though creative work inevitably involves a degree of appropriation (Ryan, 2015). Posthumanist design has been further articulated as a practice of “designing-with,” characterised by humility, relationality, care, and standing alongside non-humans rather than above them (Forlano, 2017; Choi et al., 2023). The concept of sympoiesis in interspecies mutualism (Haraway, 2016) offers a theoretical path towards this direction. In actual human-plant interactions, this relational restructuring is realised through the cultivation of an ecological self, whereby humans expand their ecological awareness and emotional connection through interaction with plants (DelSesto, 2020). This shift manifests in the field of human-plant interaction as an evolution in the role of plants, moving from utilitarian objects serving as proxies for nature, triggers for experience, or interfaces for action, towards “living co-inhabitants” interdependent with humans (Loh et al., 2024). In design practice, this shift is reflected in treating plants as collaborators rather than objects, and incorporating them into participatory, relational spaces (Crosby and Vanni, 2023; Janicki et al., 2024a).

Yet, has the theoretical turn translated into practice? Digital technologies increasingly mediate human contact with nature, which provides diverse forms of direct and indirect experiences (Webber et al., 2023). However, such work remains largely oriented towards enriching human experience, with comparatively limited attention to the needs or wellbeing of the natural entities themselves. An ethical review of interaction design papers shows that a considerable proportion of studies pose insufficiently considered risks to plants, leading to the proposal of ethical design principles grounded in a “Respect for Nature” framework (Fell et al., 2022). An analysis of 114 post-anthropocentric design artefacts further shows that even in works claiming to decentre the human, artefacts still tend to support human agency over that of natural entities. Making nature “legible” to humans is far more common than ceding control of nature (Kneile et al., 2025). This tendency aligns with what Wang (2024) describes as authoritarian nature, in which technologically mediated environments impose human-defined logics onto non-human life, rendering it observable, interpretable, and governable on human terms. Drawing on examples from Singapore, including the Gardens by the Bay Supertrees as large-scale media architecture installations, Wang illustrates how human-plant interactions are staged as spectacular, highly controlled sensor-driven experiences for human entertainment and consumption while masking underlying ecological asymmetries. Rather than redistributing agency, such designs risk reinforcing unequal power relations, “copy-paste urbanism,” and “greenwashing” (de Freitas Netto et al., 2020; Foth et al., 2026).

Alongside proposing human-plant interaction design frameworks, scholars have also called for research to move beyond mimicry or parasitic exploitation of nature towards respectful and long-term partnerships with nature (Chang et al., 2022). Design explorations of cross-species relationships corroborate this discussion. Research on human-fungi relationships suggests that collaborative survival strategies require designers to redefine who benefits from interaction (Liu et al., 2018). These studies point to a shared observation: advocating for the decentring of design at a theoretical level does not automatically alter the actual flow of benefit. Designers may embrace posthumanist frameworks in intent yet still replicate structures in which humans remain the primary beneficiaries in implementation.

Existing classification tools offer different entry points for understanding this field. Kim et al. (2023) developed a taxonomy of engagement levels for surfacing “livingness” in works that integrate living microorganisms. In built environment research, a structured taxonomy of interactive properties has been used to help researchers establish shared terminology and identify research gaps (Pulsipher and Giannakos, 2023). In the broader context of cross-species relationships, cross-scale integration of artefacts and organisms has also been proposed as an analytical framework for understanding how design can support interspecies mutualism (Tian and Herr, 2025). In the context of designing for participatory urbanism, more-than-human media architecture and middle-out design approaches have explored how digital technologies can mediate relationships across human and non-human actors, fostering collaborative processes that extend beyond human-centred participation to include concerns of multispecies justice (Caldwell and Foth, 2014; Foth and Caldwell, 2018; Sheikh et al., 2023b; Fredericks et al., 2026). Each of these works responds to specific questions, such as how to classify, how to display livingness, and how to foster participation.

Our review demonstrates that when human-plant interaction is deployed in urban environments, a fundamental question remains: who ultimately benefits? Do such systems enhance human experience, improve plant living conditions, or contribute to broader ecological outcomes? Existing classification frameworks describe interaction modes and technical composition but offer limited means to interrogate this distribution of benefits. This question is closely related to, but distinct from, two recent contributions: Fell et al. (2022) examine plant-human relations through a biocentric lens focused on research ethics and risks to plants, while Loh et al. (2024) argue for reconceptualising plants from utilitarian objects into living co-inhabitants. Neither asks how benefits are actually distributed across humans, plants, and built environments. To address this gap, this research examines practice cases from interaction design, media art, and media architecture through the lens of benefit orientation. We propose a Beneficiary Taxonomy that shifts analytical focus from how systems work to for whom they work. By studying the flow of benefits, this framework enables more explicit consideration of trade-offs and alignments between human and plant wellbeing. The following section outlines the methodological approach used to construct the corpus and develop the taxonomy.

3 Methodology

This section outlines the approach we used to construct the corpus and develop the Beneficiary Taxonomy. We first introduce the scoping review approach and case collection strategy. Subsequent sections (3.1 to 3.4) detail the four phases of corpus construction (identification, screening, eligibility assessment, and supplementation) across academic and creative practice sources. Section 3.5 then describes the inductive process of coding and taxonomy development based on the assembled corpus.

This study adopts a scoping review methodology to assemble a heterogeneous corpus spanning academic literature, specialist media art archives, and artist documentation as the empirical basis for developing the taxonomy. A scoping review is appropriate where evidence is distributed across diverse sources and formats and where the aim is to map a field rather than evaluate the effectiveness of specific interventions (Arksey and O'Malley, 2005; Pham et al., 2014). In contrast to systematic reviews, which typically address narrowly defined questions through structured synthesis and formalised inclusion, appraisal, and comparison criteria, scoping reviews support more exploratory forms of analysis; scoping reviews allow concepts, categories, and relationships to emerge inductively from diverse sources, tracing how a field has been studied, and identifying patterns and gaps (Munn et al., 2018). This approach enables us to first establish a broad view of technology-mediated human-plant interaction practices, then identify recurring patterns of benefit orientation, and subsequently develop these into a taxonomy. This study follows the framework proposed by Arksey and O'Malley (2005), incorporating refinements by Levac et al. (2010) on iterative searching and team consultation. PRISMA-ScR guidelines (Tricco et al., 2018) are used as a reporting reference and are applied adaptively to accommodate creative practice sources. The purpose of the case collection is to provide an empirical basis sufficient for developing the taxonomy, rather than to construct an exhaustive dataset.

The case collection focuses on artefacts involving living plants within built and urban environments. In many cases, these plants are displaced from their native ecological contexts and incorporated into designed systems that shape how they grow, respond, and interact. While some works rely on tightly controlled conditions in which plants depend on human or technological support, others operate in more open environments where plants retain partial ecological autonomy. Across these variations, plant life is nonetheless conditioned by design decisions that define how it is sustained, perceived, and engaged. Each interaction therefore reflects a prior human framing of care, use, and responsibility. The case collection seeks to understand how artists and designers consider the distribution of benefit in their works. Specifically, it examines the frameworks underlying these works that determine how benefits are distributed among human participants, living plants, and broader environments. While the analysis focuses on articulated or implied benefits, this does not assume that such outcomes are realised in practice; potential harms or asymmetries may remain implicit or unaddressed within the works themselves.

Following case collection, the corpus was analysed to identify recurring patterns of benefit orientation, which were then categorised and developed into the Beneficiary Taxonomy. We proceeded through four phases—identification, screening, eligibility assessment, and supplementation (summarised in Figure 1)—each of which is detailed in the following subsections.

3.1 Phase 1: Identification

The identification phase initially proceeded along two parallel paths: an academic database search and a search of a specialist media art archive. A third path, a curated personal archive, was later introduced in Phase 4 (Supplementation) to address coverage gaps that emerged during the first two searches (see Section 3.4 for details).

3.1.1 Academic database search

Three academic databases were selected to cover research in HCI, interaction design, and media art: (i) ACM Digital Library and (ii) IEEE Xplore cover the primary publication venues in HCI, including CHI, DIS, and TEI, among others. (iii) Web of Science was included because media art publications such as Leonardo are included in its Arts & Humanities Citation Index (A&HCI). Other databases such as Scopus were not included separately due to substantial overlap in coverage with Web of Science.

Search terms were structured around two concept clusters connected by the Boolean operator AND: plant and organism terms, and interaction and installation terms. The search period was set from 2005 to 2025 to capture the emergence and maturation of human–plant interaction as a distinct area of practice, coinciding with the broader development of interactive systems, biosensing technologies, and media art practices that integrate living organisms. Earlier precedents exist, but prior to the mid-2000s such works were sparse and less systematically documented across both academic and creative practice contexts. Wildcards (^*) were applied to core terms to capture morphological variants. Biology-related terms were specified explicitly rather than truncated, to prevent “bio” from retrieving large volumes of life science literature unrelated to this study. “Power plant” and “plant management” were added as exclusion terms to reduce false positives from energy engineering and agricultural literature. If a query returned fewer than 500 results, all records were retained. If results exceeded 500, the top 500 ranked by relevance were retained. Table 1 lists the search strings, filters, and results for each database. After deduplication across the three databases, a total of 1,419 records were identified (see Table 1).

Table 1. Search strings, filters, and results for database searches (ACM, Web of Science, IEEE Xplore).

3.1.2 Art archive search

Significant works in media architecture and media art are often documented in exhibition catalogues, festival archives, and artist statements rather than in peer-reviewed publications. Relying solely on academic databases would inevitably miss part of the practice in this field. To address this gap, three complementary, specialist media art archives were searched in parallel (Table 2): (i) The Ars Electronica Archive has documented international media art practice since 1987. It covers decades of competition entries and exhibited works and serves as a comprehensive repository of global media art. (ii) The STARTS Prize database complements the breadth of Ars Electronica with selectivity. Although the two sources overlap to some extent, the STARTS Prize is a globally open competition funded by the European Commission. Its awarded and nominated works represent forward-looking practice at the intersection of art, science, and technology, making it a valuable source for identifying landmark cases. (iii) V2_ Institute for the Unstable Media was included to capture additional, experimental, and interdisciplinary works, particularly those preserved through residency and festival programs. ZKM Center for Art and Media was considered but excluded due to substantial overlap with the Ars Electronica collection.

Table 2. Search terms and results for supplementary art-repository searches.

Each archive was searched manually using the term “plant” within its available interface. In the Ars Electronica Archive, the search was limited to the Prix category. The Pictures and Videos categories were excluded because their content provided insufficient information to support screening judgments. The Publications category was excluded because the works it contains are already represented in the Prix category. In V2_, the advanced search function was used to retrieve Articles, Publications, and Works containing the term “plant” within the period from 1 January 2005 to 31 December 2025. The STARTS Prize database was searched across its full collection with no additional filters. A total of 222 records were identified across the three archives (see Table 2). Across both paths, 1,641 records were identified in total (1,419 from academic databases and 222 from art archives) and advanced to the next phase.

3.2 Phase 2: Screening

The identification phase prioritised broad coverage and therefore included many records outside the study's scope. The screening phase aimed to exclude clearly irrelevant records through rapid assessment of titles, abstracts, and available metadata prior to full-text review. Screening was guided by two criteria designed for consistent, binary judgement based on minimal information: whether the work involved a living plant and whether it had been physically realised. These could typically be determined from abstracts, images, or metadata (e.g., whether the work was reported as a prototype, installation, or exhibition). More complex considerations—such as the presence of technological mediation or human participation—were reserved for the eligibility assessment in Phase 3.

All records were screened by title and abstract, with brief consultation of full texts and images where necessary. Records from the art archives were assessed using both textual and visual documentation. Ambiguous cases were resolved through discussion by the author team. Screening was conducted according to the following criteria (SC):

• SC-1: Living plants as a physical component. The work must include one or more living plants as part of its physical composition. Works addressing plants only as a theme (e.g., digital representations or virtual plants in AR/VR) were excluded.
• SC-2: Physically realised artefact. The work must describe an installation, prototype, or interactive artwork that has been physically realised. Conceptual proposals without functional implementation were excluded as this review is primarily concerned with how a work's intended benefit is configured within a realised system—its technological mediation, mode of interaction, and spatial deployment in urban and built environments—dimensions that can be more readily examined in physically realised works.

Following screening, 1,338 academic records were excluded, leaving 81 papers (ACM = 70, IEEE = 3, WoS = 8), and 191 artworks were excluded, leaving 31 non-duplicate records. In total, 112 records advanced to the eligibility assessment phase.

3.3 Phase 3: Eligibility assessment

All 112 records that passed the screening phase entered full-text review to assess whether they met the conditions required to analyse the distribution of benefits. Three eligibility criteria (EC) were defined in advance, each corresponding to an analytical precondition of the study (see Table 3). Each record was evaluated against these criteria, with inclusion determined on a binary basis. Records failing one or more criteria were excluded, and ambiguous cases were resolved through team discussion.

Table 3. Screening and eligibility criteria applied across phases 2 and 3.

• EC-1: Technological mediation in human-plant interaction. The work must include sensors, electronic components, or computational systems that mediate interactions between humans and plants, and produce perceptible outputs (e.g., visual, auditory, tactile, or olfactory). Technologies used solely for monitoring without designed perceptible output, as well as purely physical plant installations without electronic or computational mediation, were excluded.
• EC-2: Human participation as design intent. The work must incorporate human participation, whether through direct interaction, indirect engagement, or observational witnessing. This criterion ensures that a distribution of benefits between humans and plants can be analysed. Purely technical sensing studies, agricultural monitoring applications, and commercial products with incidental human involvement were excluded.
• EC-3: Built environment as spatial context. The work must be deployed in an urban context (e.g., galleries, exhibition spaces, workplaces, parks, or domestic settings), where the spatial context forms part of the experience. Works indifferent to location—such as laboratory prototypes, temporary demonstrations, wearable systems, or works in non-built natural environments—were excluded. Works connecting urban publics to plant systems at agricultural or larger ecological scales, e.g., GROW (Studio Roosegaarde, 2021a); The Plant Intelligence Plan (Zhang, 2022), are relevant but fall outside the urban/built-environment context required by EC-3, and were therefore not included in the coded corpus.

Following this phase, 81 records were excluded. A total of 31 works were retained (ACM = 18, IEEE = 1, WoS = 3, art archives = 9).

3.4 Phase 4: Supplementation

The academic database and art archive searches did not fully capture the range of plant-mediated interactive works. Many relevant projects are not published in academic venues or indexed in major art archives but have been publicly exhibited and documented through curatorial processes. Following a strategy similar to Kim et al. (2023), this study supplemented the corpus using a personal curatorial database maintained by the first author. The database has been developed since 2020 through sustained collection across multiple sources, including artist websites, snowball sampling of related works, keyword-based searches (e.g., “plant interactive installation,” “plant bioart”), and works encountered through exhibitions, festivals, and conferences. All entries were documented with supporting data (e.g., web links, images, descriptions).

Supplementary candidates were deduplicated against prior phases and assessed using the same screening (SC-1, SC-2) and eligibility criteria (EC-1–EC-3). In addition, they were required to meet two conditions: (i) evidence of peer review or curatorial selection (e.g., exhibition at an international venue, inclusion in a reviewed publication, or recognised award), and (ii) sufficient public documentation to support independent interpretation. While the database reflects the first author's research trajectory, potential selection bias was mitigated by applying identical eligibility criteria across all sources and requiring external validation through curatorial or peer-reviewed selection. The database initially contained 68 candidate works. After deduplication, 30 remained; following screening and eligibility assessment, 19 were included. Combined with the 31 works retained from previous phases, the final corpus comprised 50 works for coding and taxonomy development.

3.5 Taxonomy construction

Following case collection, the next step was to develop an analytical framework capturing both the flow of benefits and the configurations that enabled them across the 50 works. This was guided by three questions: (i) what benefits are produced; (ii) who benefits; and (iii) how roles are configured among humans, plants, technology, and environmental elements. These questions were kept deliberately open: no predefined definition of benefit was imposed, and beneficiaries were not limited to human or plant actors. The taxonomy was derived inductively, with recurring patterns identified, named, and organised into a classificatory framework.

A clarification is warranted regarding how “benefit” was interpreted. While no a priori restriction was placed on what could count as a benefit or who could be a beneficiary, benefit was consistently understood as the intended or implied positive outcome of a work, as articulated through its design, documentation, artist statements, or reported aims. The taxonomy therefore analyses how benefits are conceived, designed, and distributed, rather than whether they are empirically realised: we did not independently measure outcomes or verify plant wellbeing through physiological data, which is unavailable for most works in the corpus. Coding was accordingly based on the primary benefit orientation embedded in each work's design structure, interaction logic, and stated aims—not on every possible audience reaction, emotional response, or downstream effect, since coding such diffuse effects (through which almost any work could be said to benefit humans via learning, reflection, or aesthetic experience) would cause the categories to collapse into one another.

To ensure consistency, the three questions were operationalised into a codebook comprising six codes, each supported by guiding questions (see Table 4). Benefit Type distinguishes functional, experiential, and ecological outcomes. Plant Role and Human Role capture the roles assigned to plants (from passive medium to active participant) and humans (e.g., user, caregiver, observer). Interaction Pattern identifies the directionality of interaction. Technological Mediation specifies the role of technology (e.g., sensing, translating, actuating), while Environment Role considers the built environment as a container, stage, habitat, or participant. These codes provide structured perspectives for making value assumptions in each work visible and comparable.

Table 4. Codebook of six coding dimensions used for taxonomy construction.

The taxonomy was developed over approximately 2 months by a team of four researchers with backgrounds in interaction design, media art, human-building interaction, architecture, and urban studies. All 50 works were initially coded by the first author using the codebook, then reviewed iteratively by the full team, with attention to category clarity, distinction, and consistency; disagreements were resolved through discussion until consensus was reached. Coding was grounded in each work's own evidence, such as artists' or designers' statements, project documentation, and reported aims, together with the interaction structure of the work itself, rather than in the authors' subjective interpretation. We acknowledge that a gap may exist between how creators conceptualise their work and how its intended benefit is expressed in the realised system; we return to this gap between intended and actual outcomes in the Discussion (Section 5.2).

Cross-referencing the six codes revealed recurring patterns that consolidated into two higher-level analytical dimensions: the direction of benefit flow (human-oriented, mutual, or plant-oriented) and the scope of benefit (individual, dyadic, community, or ecosystem). We describe the resulting framework as a relational taxonomy. Unlike the strictly nested hierarchy of classical biological systematics (Linnaeus, 1753), the term “taxonomy” in applied and interdisciplinary research refers to a structured classification that organises concepts to reveal relationships between them rather than to impose a fixed genealogical descent. This relational usage is well established across fields: Münscher et al. (2016) employ a taxonomy in behavioural decision-making to map choice-architecture techniques along multiple dimensions, while Laubheimer (2022) describes taxonomies in UX practice as organised sets of concepts that provide context and reveal connections. Our framework operates in the same spirit. Its purpose is not to slot works into rigid categories but to identify patterns across the corpus and to understand the relationships between those patterns. Together, these two dimensions yield the five categories of the Beneficiary Taxonomy. Each category occupies a position along the two axes; where cells in the resulting grid remain empty, this reflects the fact that no works in the current corpus matched those particular combinations rather than a gap in the framework itself. Figure 2 provides an overview of how the five categories are positioned along the two dimensions.

4 Results

Our analysis of the 50 works revealed that the distribution of benefits in technology-mediated human-plant interaction follows recognisable and recurring patterns. These patterns are organised along two dimensions: the direction of benefits (who gains) and the scope of benefits (whether benefits are one-sided, mutual, or extend beyond the human–plant relationship into broader relational networks).

Based on these dimensions, we identify five benefit orientations—Restoring, Nurturing, Autonomising, Entangling, and Extending—organised into a Beneficiary Taxonomy (see Table 5). They describe a progression from unidirectional benefit, through shifting and reciprocal relations, to expanded ecological and community-oriented configurations. Figure 3 illustrates how these orientations differ in terms of relational configuration and benefit flow. The following sub-sections examine each orientation in turn. While individual works may exhibit multiple orientations, each was classified according to its dominant pattern of benefit distribution for the purposes of analysis.

Table 5. The Beneficiary Taxonomy: category definitions, design stance, and corpus counts (n = 50).

4.1 Category 1: Restoring

Restoring is defined as a design orientation in which benefits flow primarily to humans. Designers position plants as passive media that serve human restorative experiences. Technology captures the physical properties or physiological signals of plants and translates them into sensory outputs perceivable by humans, supporting emotional regulation, cognitive engagement, or behavioural motivation. In everyday life, a potted plant on an office desk may incidentally relieve stress, but such benefits are sporadic and uncontrollable. Works in the Restoring category systematise this incidental quality: the presence of plants is orchestrated into repeatable sensory experiences, while the question of whether the plants themselves are healthy or in suitable growing conditions falls outside the design agenda. In these works, the bidirectionality exhibited by the interactive system operates between humans and technology, rather than constituting biological reciprocity between humans and plants. This category comprises 18 works (36% of the corpus) and is the largest among the five categories. These works are deployed across the widest range of built-environment settings among the five categories, spanning galleries, domestic spaces, workplaces, outdoor public spaces, and care facilities.

Our analysis identified four design strategies within this category: touch-triggered, signal translation, behavioural mapping, and display instrumentalisation. Touch-triggered is a technological extension of the oldest form of human-plant contact. Sensors detect physical contact between humans and plants and convert it into predetermined sensory outputs. In Sonnengarten, urban visitors touching plant displays triggered LED lighting in nearby structures (Fastnacht et al., 2017) while in Flora Luma, bioelectrical signals from plants controlled coloured animation of self-illuminating sculptural objects (Frankjær and Kitel, 2016). In Please Touch the Plant on Your Way Up the Stairs, the built environment can become a constituent element of an interaction strategy: the designers embed plants along a daily commuting route on office staircases, so that pedestrian flow through the space itself becomes the condition that triggers interaction (Loh and Santo, 2010). Liquid Landscape applies the touch-triggered logic at an architectural scale with a 50-square-metre interactive landscape of soil and grass constructed over a deformable mechanical structure (Studio Roosegaarde, 2021b). Human touch can be used to elicit a plant stress response, for an aesthetic output, sustainable outcome, or enable formation of emotional attachments as in Touchology (Seo et al., 2015) which provided a plant-touch interface for children with autism and older adults in senior housing.

Signal translation converts physiological signals or physical properties of plants into visual, auditory, or olfactory outputs perceivable by humans. Synplant applies supervised learning algorithms to analyse the bioelectrical signals of plants responding to wind and rain. The results are translated into cymatics patterns, offering audiences a visual and sonic experience of environmental change from the plant's perspective (Hu et al., 2023). Beyond the immediate aesthetic experience, such visualisations of how plants respond to and are affected by environmental change can foster a less direct but meaningful benefit: by making these processes perceptible, they may gradually cultivate ecological literacy and a stronger sense of human-plant connection. In Movement, the sound field is generated from the Mimosa sp. plant's stress responses to human body movements (Hu and Chou, 2021b). However, this response constitutes stress rather than benefit. Empathic Growth simultaneously detects human galvanic skin response and plant biosignals, but the two datasets merge only at the visualisation layer rather than interacting at the material level (Wu et al., 2025). One Tree ID extracts volatile organic compounds from a tree and synthesises them into a perfume (Meyer-Brandis, 2019), and MB>CO2 physically releases CO₂ calculated from real-time internet activity into a sealed plant terrarium, causing the enclosed plants to wilt as the carbon footprint accumulates (Biersteker, 2022). Many of these works emphasise “non-human perspectives” or “cross-species empathy” at the discursive level, yet their interactive relationships consistently exhibit unidirectionality. That said, reading these works as simply unidirectional overlooks one meaningful dimension: by making plant responses and signals perceptible, and by showing how human actions affect plant welfare, they help audiences become aware of their own impact on plants. This does not make the plant a beneficiary, so the works remain human-oriented, but it is a potentially significant aspect of how they are intended to operate.

Behavioural mapping links personal data from humans with plant growing conditions, rendering plants as living indicators of human behavioural performance. Go and Grow maps step counts and social interaction frequency onto plant growing conditions, so that the flourishing or wilting of the plant directly reflects user performance (Botros et al., 2016). MossWater applies similar logic by linking water intake to moss irrigation (Zhou et al., 2021). In both works, plants do receive better care conditions as a consequence of improved human behaviour. However, system success is measured by whether human behaviour improves, not by whether the plant is healthy. This constitutes a difference between Restoring (C1) and Nurturing (C2): in Restoring, plant care is a means to encourage human behavioural change; in Nurturing, encouraging human behavioural change is a means to support plant care.

Display Instrumentalisation directly appropriates the physical morphology or physiological responses of plants to convey information. Plantxel uses the thigmonastic response of Mimosa spegazzinii to construct a controllable dynamic display. Pneumatic stimulation triggers leaf closing, with reopening rendering pixelated images (Gentile et al., 2018). Similar strategies appear in Plant Shadow Morphing (Colley et al., 2019), Overgrown (Degraen et al., 2019), and InfoPlant (Hammerschmidt et al., 2015), which transform plants into environmental information displays through shadow manipulation, whole-plant actuation via a servo-driven endoskeleton, and multimodal augmentation, respectively. Anastatica Sensibile represents the highest degree of plant vitality participation within this category: irrigation of 45 Selaginella lepidophylla plants is triggered by visitor density, and the plants self-regenerate upon watering, unfurling within a day and re-closing after 4 days as moisture depletes (Di Maro et al., 2014). The built environment, in this case a gallery, functions not as a habitat for plants but as a stage for their survival performance.

Restoring comprises 18 works that span a range of built environments from urban plazas to care facilities and from gallery installations to office desktops. They deploy technical means ranging widely from simple tactile sensing to machine learning algorithms, but one characteristic of the outcomes is consistent: benefits flow unidirectionally towards humans. At the same time, we observe that although several works belong to Restoring at the system level, the signal detection and translation technologies they employ, such as bioelectrical signal analysis, volatile organic compound extraction, and machine-learning-driven biosignal classification, can provide reusable technical foundations for constructing bidirectional human-plant interactions (C4 Entangling).

4.2 Category 2: Nurturing

Nurturing is defined as a design orientation in which benefits flow exclusively to plants. Designers use technology to mediate or augment human care practices, enabling measurable improvements in plant growing conditions such as light, water, and nutrient availability. In Restoring, plant care serves as a means to achieve human behavioural change. In Nurturing, this relationship is inverted: human behavioural change serves as a means to achieve plant care. This category comprises two works (4% of the corpus) and is the smallest among the five categories. Both works in this category are situated in gallery and domestic settings.

PlantPal employs a web interface where multiple remote users tend individually assigned 1 m² beds through a single robot, transforming a communal outdoor garden into a technologically mediated site of distributed care (Zeqiri et al., 2025). The interaction is bidirectional: plant growth status is relayed to humans via sensors, they adjust plant care strategies accordingly, and the adjusted actions are carried out on the plants through robotic actuators. This forms a closed loop oriented towards plant wellbeing. Unlike Go and Grow and MossWater in Restoring (C1), the success of PlantPal is measured by whether plants receive suitable growing conditions. StellaVerde advances the automation of care further still. Plants, fish, and a battery-less “spiderbot” form an autonomous aquaponic ecosystem sustained through biological and mechanical exchange, with the human “exempted from the decision-making process” (Krpič et al., 2023). Across PlantPal and StellaVerde, the extent to which technology assumes care responsibilities deepens progressively. Human involvement shifts from active caregiver to remote operator to observer. Regardless of how human participation varies, the measure of success consistently points to the objective condition of plants, namely photosynthetic efficiency, water supply, and growth performance, rather than human subjective experience. This plant-centred design orientation nonetheless remains a product of human agency: it is the designers who determine what constitutes the needs of plants, and who deploy technology to fulfil them.

4.3 Category 3: Autonomising

Autonomising is defined as a design orientation in which the designer transfers behavioural initiative from humans to plants. Technology externalises plant physiological processes into observable autonomous behaviours, enabling plants to drive system operation without human instruction: the designer builds the technological infrastructure and leaves plant physiological activity as the sole driving force of system behaviour. The human role shifts from user or caregiver to observer, and in some cases is absent entirely. This category comprises 10 works (20% of the corpus). This category is concentrated almost entirely in gallery settings, with a smaller number of works deployed in outdoor public space.

Plants cannot move on their own, and this biological constraint is overcome by technology in several works within this category. In Action Plant, the action potentials of Mimosa pudica are detected by a signal amplifier and mapped onto the movement of a robotic wheelchair structure, allowing the plant to move away when touched (Henriques, 2011). Nomadic Plants extends a similar logic to the autonomous migration of plant-microbe symbionts (Esparza, 2010). Hortum machina, B is a hemispherical mobile architectural structure deployed and tested in London that houses multiple plants (Sheikh et al., 2021). Plants sense their environment through both electrophysiological signals and environmental sensors, and autonomously determine the direction in which the entire structure moves through urban space. The specific conditions provided at a predefined location within the built environment no longer determine the fate of plants. Instead, plants have some ability to choose their environment (Camilleri and Sampaio, 2016).

In other works, plant autonomy is expressed not through spatial displacement but through control over mechanical systems. In Bamboo Bot, varying electrical-resistance signals read from a live lucky-bamboo stalk are mapped in real time to the joint movements of an industrial robot holding the plant (Bowen, 2024). In Plant Machete, bioelectric resistance signals from a philodendron's leaves drive a robotic arm wielding a machete (Bowen, 2022a). In Pneuma, each leaf functions as an independent light-sensing unit where photosynthetic electronic signals enable the installation to identify and autonomously follow the light source direction (Hu and Chou, 2020). Plantae Agrestis deploys a distributed cyborg plant system in a botanical conservatory where collective coordination patterns emerge from individually collected plant biosignals (Sareen and Kakehi, 2023). PL'AI introduces AI as an intermediary layer, with the robot observing cucumber growth and tendril movement in real time and responding accordingly (Petrič, 2020). In Plant Bot, plants train an AI system through image recognition to learn their own watering needs, after which the system autonomously carries out care (Bowen, 2022b). Anachronic Landscapes deploys an industrial robot to autonomously carry out daily irrigation and fluorescent fertilisation, with humans entirely removed from the care loop (Batliner and Newsum, 2017).

Plant autonomy in these works still operates within a technological framework preset by humans. Plant physiological signals do drive system behaviour, but how those signals are detected and how they are mapped onto specific movements or responses are determined by the human designer in advance. In Action Plant, Mimosa pudica triggers wheelchair movement, but the speed and directional mapping are preset by the designer. In Plant Machete, the plant controls a machete, but the choice of this object, the range of motion, and the force parameters of the robotic arm are all products of human design decisions. In other words, plants determine when and whether to act, but the form of action has already been established by the designer at the deployment stage. Plants do not act freely in an open environment. Rather, they express their physiologically driven behavioural preferences within a technological framework constructed by humans. This category is defined not by who benefits but by a structural feature, in which the plant's own signals drive system behaviour. It is also where benefit flow is least clear: the plant does not need the apparatus, and what is most visibly staged is a human-preset framework. Where a benefit can be discerned, it tends to lean towards humans (e.g., visual surprise, or a heightened sense of plant agency), but we do not speculate beyond what the artists themselves state. Plant Machete is thus included for its plant-driven control structure, not for any benefit to the plant.

The distinction between Autonomising and the preceding two categories lies in who initiates action. In Restoring and Nurturing, humans initiate interaction or design care systems. In Autonomising, plant physiological processes themselves drive system behaviour. Even when humans are part of the causal chain—as in Action Plant, where a human touches a Mimosa pudica whose leaf-folding then drives a mechanical system—the technical system reads only the plant's physiological signal, not the human input. The system's behaviour is therefore initiated unilaterally by the plant. A clear boundary also exists between Autonomising and the subsequent Entangling (C4): in Autonomising, humans function primarily as observers rather than direct beneficiaries, and do not form a reciprocal loop with plants through the system. In Entangling, by contrast, humans and plants constitute a bidirectional negotiation at the biological level.

4.4 Category 4: Entangling

Entangling is defined as a design orientation in which humans and plants form a bidirectional benefit relationship mediated by technology. Human physiological activity or behaviour produces biological effects on plants through the system, while plant physiological activity generates perceivable feedback to humans through the system. Together, these form a sustained loop of negotiation. In the preceding three categories, benefits are unidirectional: flowing to humans (Restoring), to plants (Nurturing), or driven unilaterally by plants (Autonomising). Entangling is the first category in the taxonomy that requires benefits to flow bidirectionally between humans and plants. This category comprises 9 works (18% of the corpus). We acknowledge that not all works classified as Entangling demonstrate this loop in its full, symmetrical form; some constitute experimental probes that explore the conditions under which such loops might emerge, and we discuss these limits in detail in Section 5. Bidirectional refers to the structure of the relationship—each party materially acts on and responds to the other—rather than to a guarantee of symmetrical physiological benefit. The benefit returned to the plant spans a spectrum, from direct physiological gain (e.g., the photosynthetic gas-exchange works discussed below) to a form of relational participation in which the plant's physiological state co-determines the interaction (in more exploratory, sensorimotor works). In these latter cases, the bidirectionality is best understood as perceived and intended by the artists, and our coding follows their stated framing.

Respiratory exchange is the most direct bidirectional mechanism in this category. In Dialogue, CO2 exhaled by humans triggers a photosynthetic response in plants, and the photosynthetic process is translated into audible sound feedback. Breathing and photosynthesis form an immediate cross-species sensory exchange. As Haapoja (2008) describes, “lights, sound, CO2, digital and analogue technology form a circuit of information which enables the viewer to perceive their interaction with the non-human environment not only as a physical process but also as a form of communication.” Ex.A.R.U. materialises this logic into a symbiotic breathing apparatus. CO2 exhaled by humans is delivered directly to plants, and the oxygen produced through plant photosynthesis returns to humans, forming a closed gas exchange loop. The work thereby prompts reflection on the “co-dependency” between humans and plants (Dolinšek, 2023). PlantConnect extends this mechanism in scale. Participants blow into CO2 sensors to trigger grow light arrays that promote rice photosynthesis, while voltage signals from plant-microbial fuel cells and CO2 level changes are analysed by machine learning algorithms to generate a spatialized sound environment (Castellanos, 2023). In these three works, CO2 from human breath participates directly in plant metabolic processes as a raw material for photosynthesis, and plants receive measurable physiological benefits.

In other works, bidirectional loops are realised through different material media. Project Florence converts language sentiment analysis into light control signals that affect the plant-growing environment, while translating plant physiological responses into reply text (Steiner et al., 2017). The loop is two-way: the participant's text drives the light the plant responds to, while the plant's sensed state co-determines a reply printed back to the participant, who reads it and continues the exchange. The human is thus also affected, so the interaction is perceived as a two-way exchange rather than one-directional control—a bidirectionality that reflects the artists' stated intention, even as the plant functions as a medium within the loop. Slowing Plants, Slowing Home selects microclimatic humidity as a shared interaction medium between humans and plants. Plant transpiration, smart curtain movement, and human breathing all contribute to humidity variation. The designers propose “a relationship-centred strategy” that uses humidity as an “inclusive feature of a microclimate” rather than the plant itself as the interaction medium, pointing towards a “plant-decentred perspective” (Wen and Wu, 2022). Sonifying Aquatic Plant Photosynthesis converts the bubble sounds produced by aquatic plant photosynthesis into ambient music through machine learning. The design intent is to motivate users' everyday care practices through music (Dai et al., 2025). Encountering Human-Plant Relations deploys plant-electronic devices in participants' living spaces and, through weeks of cohabitation, finds that interaction design can cultivate multiple forms of human sensibility towards plants (Hansen and Rolighed, 2024).

Syn-thesasia and Breathing establish direct coupling between human and plant bodies at the physical level. In Syn-thesasia, when human touch causes Mimosa sp. leaves to fold, EMS technology synchronously contracts the human arm muscles, making the human body directly experience the plant's stress response (Hu and Chou, 2018). In Breathing, a living plant's bioelectric signal, captured via a galvanic-skin-response circuit, drives a hybrid robotic apparatus of servos, fibre optics, and LEDs that responds in dialogue with the viewer's breath (Nóbrega, 2009). In both works, the bidirectional loop is primarily perceptual and sensorimotor rather than nutritional or developmental: humans gain an embodied awareness of plant signals, while plants do not derive direct physiological benefit. We therefore read these works as exploratory experiments in entanglement—probing how human and plant bodies might be coupled—rather than as fully realised reciprocal benefit systems. These works are divided between gallery and domestic settings, reflecting their dyadic, close-proximity character.

The works in the Entangling category span a range of bidirectional mechanisms, from respiratory gas exchange to microclimate negotiation, from bodily synchronisation to music-driven care. These works are deployed across galleries, domestic settings, and urban public spaces. Across these contexts, the bidirectional loop turns the built environment containing the installation into a site of negotiation, where humans and plants encounter, adapt to, and learn from one another, rather than a passive backdrop in which one party simply acts upon the other. In Slowing Plants, Slowing Home, the domestic microclimate becomes a site of negotiation. It is important to note that the works gathered under Entangling do not always provide a strong, symmetrical benefit loop for both parties; many function as experimental probes that ask how human and plant lives might be entangled, and what role the built environment could play in supporting such entanglements. Compared to the preceding three categories, technology in Entangling no longer serves the needs of one party unilaterally, but functions as infrastructure that sustains ongoing negotiation between humans and plants.

4.5 Category 5: Extending

Extending is defined as a design orientation in which benefits expand beyond the human-plant dyad to reach communities, the public, or broader ecosystems. Plants are no longer merely participants in individual interactions but are repositioned as members of larger ecological or social networks. The goal of technological mediation shifts from facilitating the human-plant relationship to awakening collective awareness, driving community participation, or restoring ecological connections. This category comprises 11 works (22% of the corpus). This is the most spatially dispersed category, with works spanning galleries, outdoor public space, domestic settings, community gardens, and flexible indoor–outdoor installations—consistent with its expansion of benefit beyond the human–plant dyad.

One group of works extends human-plant relationships to the community level through shared care. Telegarden (Goldberg et al., 1995) allows online users to remotely plant, water, and monitor seedling growth in a living garden via an industrial robotic arm. It attracted over 9,000 members in its first year and operated continuously for 9 years. Plant care here becomes a medium for global collective collaboration, and the garden itself serves as a social-ecological metaphor for virtual community (Goldberg et al., 1995; Kahn et al., 2005). Common Roots deploys a networked irrigation system among community neighbours. The study finds that shared plant care behaviour effectively promotes social connections between residents (McDonald, 2018). Ecoshroom explores communication between plants and mycorrhizal fungi. AI analyses bioelectrical signals and environmental data in real time, adjusting temperature, lighting, CO₂ levels, and humidity. Visitors participate in multi-species care through composting (Henriques, 2022). Flora Robotica constructs a tightly coupled plant-robot architectural system in which humans, robots, and plants continuously negotiate growth zones. The three parties co-evolve a living architectural structure (Hamann et al., 2015).

Another group of works translates plant-life processes into publicly perceivable ecological narratives through data visualisation. Voice of Nature connects 12 sensors to a living tree in Chengdu, collecting 1,600 environmental data points and generating digital tree rings in real time, one per second rather than one per year. Biersteker (2019) poses the central question of the work: “What if we could harness the natural climate monitoring abilities of trees to communicate the urgency of climate change to the general public?” Symbiosia, created in collaboration with plant neurobiologist Mancuso, measures photosynthesis, air quality, and chemical compound fluctuations of two trees in the garden of Fondation Cartier in Paris, further exploring root communication and symbiotic relationships between trees (Biersteker and Mancuso, 2019). Xylemia turns the focus towards the internal life flow of trees. Sensors reveal real-time sap flow and visualise the transport of water and nutrients from roots through xylem vessels. Its core message is

“in whom we can reflect, we protect.” (Biersteker, 2024)

Sensing Bodies extends the ecological narrative from climate issues to postcolonial history. Three plantation plants from the American South engage participants through biosensors and data displays in embodied interactions, fostering critical reflection on colonial violence and plantation capitalism (Janicki et al., 2024b).

The remaining works extend benefits in distinct ways. Flowish collects sap flow and soil moisture data from a distant forest and uses drip devices to cultivate algae colonies on cement sculpture surfaces in an urban park, connecting two geographically remote ecosystems (Chou and Hu, 2024). Core Sample comprises pneumatic floral elements, living plants, fungi, and soil forming a micro-ecosystem. The longer it remains in an environment, the more deeply its ecosystem merges with surrounding local vegetation (Wu, 2020). Moss is a kinetic sound sculpture driven by World Air Quality Index data. Living moss constitutes the sculptural body. Real-time air quality data from Berlin to Mumbai, Moscow, and Beijing is transformed into breathing patterns and soundscapes. The work aims to

“engages the citizens to take part in the discussion about air quality and earth democracy.” (Barotti, 2021)

The 11 works in the Extending category show that when designers shift their attention from the human-plant dyad to broader social and ecological networks, the role of the built environment transforms accordingly. Urban plazas become sites for collective climate perception. Community gardens become a medium for neighbourhood connection. Gallery spaces become arenas for postcolonial historical reflection (Sheikh et al., 2023a). Across the corpus as a whole, this spatial differentiation is most pronounced in Extending, which occupies the widest range of built-environment settings among the five categories, while galleries remain the dominant setting overall (see Table 6, Figure 4). This gallery-dominant pattern should be read with some caution, however, as it may partly reflect the composition of our search sources and the institutional nature of the works retrieved, rather than a definitive feature of how human–plant interaction is distributed across built environments; our aim here is descriptive, to document this distribution as it currently exists in the corpus, rather than to generalise beyond it.

Table 6. The 50 works classified by beneficiary category and built-environment context.

5 Discussion

In this paper, we have responded to the increasing diversification of technology-mediated human-plant interaction practices in built environments across HBI, HCI, and media art, alongside the lack of analytical frameworks for examining these practices from the perspective of benefit distribution. The Beneficiary Taxonomy introduced here provides researchers and practitioners with a set of analytical lenses centred on benefit flow with a view to enable a more systematic account of how benefits are distributed among humans, plants, and broader ecosystems. As such, benefit distribution emerges as a fundamental design consideration—one that shapes not only the roles assigned to plants within built environments, but also the terms on which humans coexist with non-human lifeworlds, now and in the future.

The following discussion develops these contributions in two parts. First, we synthesise the diverse configurations of benefit distribution identified in the taxonomy to articulate a trajectory towards more reciprocal human–plant relationships (Section 5.1). Second, we examine how design intent translates into, or diverges from, realised benefit flows in practice, which suggests key alignments and tensions for interaction design and built environment research (Section 5.2).

5.1 Towards reciprocal human-plant relationships: reconfiguring benefit distribution

By offering a set of analytical categories centred on benefit flow, the taxonomy aims to help designers make more reflective judgments about the benefit distribution when conceiving human-plant interactions. Over one-third of the works in the corpus direct benefits primarily towards humans. This proportion does not imply that these works are of lesser value. A plant-based touch interface designed for children with autism (Seo et al., 2015) and a behavioural guidance system linking plant growth to personal health behaviours (Botros et al., 2016) demonstrate contributions of designs oriented towards human restorative experience. However, the taxonomy draws attention to the fact that many existing technical pathways are available when designers wish to make different choices about benefit distribution. Supervised learning algorithms for analysing plant bioelectrical signals (Hu et al., 2023), for instance, can serve unidirectional sensory translation or be integrated into bidirectional negotiation loops. The critical distinction lies not in the technology itself, but in whether designers have made room to consider both the plant and human benefits of the system they develop.

Works using Mimosa sp. as an interaction medium illustrate concretely how benefit distribution shapes human-plant relationships. Both works draw on the stress response of the same plant. When the response is translated into a sound field and fed back to humans (Hu and Chou, 2021b), the plant serves merely as a signal source. When EMS technology synchronises human muscle contraction with the folding movement of its leaves (Hu and Chou, 2020), the human body is driven by the plant's physiological state, and the interaction shifts from unidirectional signal consumption to bidirectional bodily entanglement. The two works use the same organism and address the same stress response, yet the designers' different choices for the benefit distribution lead to a fundamental difference in the quality of the human-plant relationship. A similar observation applies to Anastatica Sensibile. The resurrection plant undergoes genuine self-regeneration upon receiving water (Di Maro et al., 2014), yet the irrigation rhythm is determined by visitor foot traffic, which can lead to plant death if there is insufficient human participation. Even where plant vitality is highly engaged, the direction of benefit remains determined by the designer's care logic.

During the classification process, we also found that boundaries between categories are often blurred in practice. Whether a work belongs to translating plant signals into a consumable human experience or to constructing a bidirectional negotiation space between humans and plants sometimes depends on extremely subtle design decisions, such as whether the system allows plant responses to alter human behaviour. In most cases, there is an insufficiently examined distance between the rhetoric of reciprocity and the reality of unidirectional benefit. Artefacts tend to prioritise human agency over the agency of natural entities (Kneile et al., 2025). The present taxonomy introduces control and benefit distribution as an ordering dimension, shifting the analytical focus to who has the authority to decide how benefits are distributed. The taxonomy does not offer value judgments on each category. Rather, it serves as a mirror enabling designers to recognise their actual position regarding control and benefit distribution.

At the same time, we find that the role of the built environment shifts across different benefit modes. This shift is visible across the corpus: human-oriented and plant-driven categories (Restoring, Autonomising) cluster predominantly in gallery settings, where controlled conditions support close observation; mutual and plant-oriented categories (Entangling, Nurturing) more often extend into domestic space, where sustained proximity between human and plant is possible; and the community/ecosystem-oriented category (Extending) is the most spatially dispersed, reaching outdoor public and community settings. Designers conceptualising human–plant interaction work should therefore treat the built environment not as a neutral backdrop but as a factor that shapes which kinds of benefit relationships are feasible. When benefits flow unidirectionally, space typically serves as a passive container for interaction. When benefits flow bidirectionally between humans and plants, the space that the built environment offers becomes more crucial for supporting human-plant engagement. Within domestic microclimates provided by the built environment, humidity, temperature, and light are no longer merely environmental parameters but critical materials that constitute the environment through which humans and plants engage in ongoing negotiation (Wen and Wu, 2022). When benefits extend further to communities and ecosystems, space itself becomes a site for collective participation and ecological action (McDonald, 2018; Biersteker, 2019). This finding is directly relevant to HBI researchers: different modes of benefit distribution place different design demands on the built environment.

The utility of the Beneficiary Taxonomy for design practice lies in providing an operable set of coordinates that help designers integrate and reconcile the needs of different participants. When plant biosignals are translated solely into human-perceivable outputs, the interaction serves to enrich human sensory experience. When signal translation is accompanied by the material participation of human physiological activity in plant metabolic processes (Haapoja, 2008; Dolinšek, 2023), the interaction moves towards reciprocity at the physiological level. Similarly, when the scope of benefit expands from individual plant welfare to the restoration of ecological networks (Zhang, 2022), the impact of design extends from the human-plant relationship to a broader ecological agenda. Moving towards reciprocal human-plant relationships does not require every work to achieve perfectly symmetrical bidirectional benefits. Rather, it invites designers to become aware of the diverse possibilities of benefit distribution and to translate that awareness into conscious design decisions.

5.2 Connecting design intent and benefit flow

A recurring phenomenon encountered during the analysis is a gap between designers' discursive descriptions of their works and the actual benefit flow at the system level. A considerable number of works in the corpus claim to address plant agency, cross-species empathy, or non-anthropocentric design principles. Yet when benefit flow is introduced as an analytical dimension, the system structures of these works still convert plant sensory capacities into sources that serve human benefits. One contribution of the taxonomy lies in providing a concrete and operable means of describing this gap.

This gap does not stem from designer oversight but reflects a difficult challenge. Ryan (2015) argues that foregrounding plants in artistic creation can draw attention to plant subjectivity. However, foregrounding is itself an act of framing. Designers shape human perception by choosing which plant characteristics to amplify, while selectively ignoring others, such as the slow temporality of plants, their dependence on soil microbial communities, and their vulnerability when displaced from native habitats. At the system level, transforming plant shadows into ambient display interfaces (Colley et al., 2019) or arranging plants as controllable pixel matrices (Gentile et al., 2018) is visually impressive. Yet whether participants will care more about a plant's survival after leaving the exhibition is a question that cannot be identified in the corpus itself. We hope that future research can track this question through longitudinal user studies, examining whether participants develop sustained behavioural changes after the interactive experience, such as actively caring for plants at home or developing new concern for the living conditions of plants in their everyday environments.

Works in the Autonomising category bring this tension into sharper focus. The 10 works in this category grant plants the power to drive system behaviour, responding most fully at the discursive level to claims of plant agency. However, while plants determine when and whether to act, the form of action is already established by the designer prior to the deployment stage (Henriques, 2011; Bowen, 2022a). Autonomising thus becomes the category where the gap between design intent and benefit flow is most pronounced: designers convey conceptual messages about plant agency, but whether plants themselves derive identifiable benefit remains an open question. This open question extends beyond the plant to the audience. Because most of these works are exhibited to prompt reflection on plant agency, a further tension arises: the operations that make a plant legible to us, that is, translating its signals into words, images, or movement, can also fold it back into a human frame of reference, so that a work may assert plant agency in its message while re-enacting a human-centred way of seeing in its form. We do not treat this as a flaw, but as a tension designers should acknowledge rather than assume their intended meaning is received.

Even in works featuring bidirectional interaction, the relationship between design intent and actual benefit requires careful examination. The immediate exchange between breathing and photosynthesis appears to construct a cross-species dialogue (Haapoja, 2008), but the grammar of this dialogue is entirely defined by the designer: which plant signals are selectively amplified, which are filtered, and what kinds of human behaviour count as valid responses. When we measure plant needs with sensors and construct interactive feedback accordingly, we confront what Wakkary (2021) calls the not-knowing things predicament: are we perceiving the true state of the plant, or only the version that our perceptual framework permits us to perceive? Between structural bidirectionality and ontological reciprocity, there lies a gap that has not yet been adequately described. One possible path towards narrowing this gap is to introduce richer indicators of plant state. Most current works rely on a single physiological parameter, such as bioelectrical signals or CO₂ concentration, to characterise plant responses. Future designs could explore integrating multidimensional plant-state data, including soil moisture, root activity, volatile organic compound emissions, and chlorophyll fluorescence. This will allow the system's perception of plant state to more closely represent physiological reality and provide means to respond more accurately to survival needs, rather than capturing only limited dimensions for more controllable or desirable outcomes, which are often needed for artistic statement, but may be harmful for the plant's wellbeing.

More fundamentally, classifying a work by its benefit orientation does not imply that its outcomes are benign: benefit and harm can coexist within a single intervention, and asking who benefits therefore entails asking who suffers. Even works coded as plant-oriented or reciprocal may still place plants under adverse conditions. For example, displacing them into gallery settings with insufficient light, or repeatedly eliciting touch- or stress-responses for the sake of an interactive or aesthetic output. Acknowledging this asymmetry does not require a separate taxonomy of harms, but it reinforces the need to read benefit claims critically rather than at face value.

For works that target plant welfare, judgments of benefit likewise depend on human knowledge frameworks. Light spectrum formula calibrations based on photobiology research (Studio Roosegaarde, 2021a) represent optimal growth conditions according to scientific consensus. Still, whether scientifically defined optimality equates to plant welfare remains ontologically open. Only two works in the corpus are classified as Nurturing. This scarcity may reflect the difficulty designers face in practice: in the absence of ontological consensus regarding the subjective experience of plants, what it means to design for plants remains a question without a ready answer. We hope that future research can establish closer interdisciplinary collaboration between plant science and interaction design, translating the latest findings from plant physiology, plant behavioural ecology, and ecology into operable design parameters, so that plant-centred design orientations no longer rely solely on well-intentioned human speculation but are grounded in a more robust understanding of plant needs.

The majority of works in the corpus are deployed in galleries and exhibition spaces, with interaction durations measured in minutes or hours. Yet actual human-plant coexistence in built environments unfolds over months and years. When human-plant interaction moves from short-term exhibition to long-term embedding in everyday urban life, the questions facing designers scale accordingly. At the community scale, Telegarden and Networked Communal Plant Watering System have already demonstrated how plant care can serve as a medium for community bonding (Goldberg et al., 1995; McDonald, 2018; Loh et al., 2023), but these practices have not yet been systematically connected with urban planning and architectural design. We hope that future research can explore how human-plant interaction might move from standalone installations to building or urban-scale infrastructure. Can networked care systems in community shared gardens be integrated into building environmental control systems so that plant growth requirements operate in coordination with building energy management and microclimate regulation? Can real-time plant data visualisations in urban public spaces, such as those explored in Voice of Nature and Xylemia, evolve from temporary art installations into permanent components of urban environmental monitoring networks? When HVAC, irrigation, and lighting systems begin to incorporate plant physiological states into their control logic, the built environment will no longer be just a passive container of human-plant interaction, but will become an active infrastructure that supports multi-species coexistence. This transformation requires new interdisciplinary dialogue among a wide range of practitioners, including HBI and HCI researchers, architects, landscape architects, urban planners, engineers, and facility managers. Because the benefits analysed here are those intended or claimed by each work rather than independently verified, the relationship between intended benefits and actual outcomes, including plant wellbeing, remains an important direction for future research.

The taxonomy judges the direction of benefit based on the designer's stated intent and observable system behaviour, rather than making claims about the actual effects of works. This is a deliberate methodological choice. The gap between intent and effect may occur at multiple levels: participants may experience a work that achieves bidirectional interaction at the system level merely as a novel sensory entertainment; participants in a work whose benefits are directed entirely towards humans may unexpectedly develop care for plants through tactile contact. What the taxonomy captures is the benefit orientation embedded in the design structure, not the cognitive and emotional responses and changes that are experienced independently during interactive events. Combining the taxonomy with empirical user experience research, particularly long-term deployment studies in domestic environments and workplaces, is a necessary direction for validating these findings and advancing the taxonomy from an analytical tool towards design guidance.

6 Conclusion

This paper has examined the diversification of human–plant interactions across HBI, HCI, and media art through the development of a Beneficiary Taxonomy. Comprising five orientations—Restoring, Nurturing, Autonomising, Entangling, and Extending—the taxonomy maps how benefits are distributed between humans and plants, tracing a shift from unidirectional extraction towards more relational configurations. Rather than prescribing optimal solutions, it offers a critical lens for interrogating how plants are positioned within interactive systems and for supporting more reflective design practices.

The notion of “benefit” is understood here as an interpretive and contingent construct. While reciprocal framings are increasingly articulated, they are only partially realised in practice. The taxonomy therefore not only maps possibilities for redistribution but also exposes power asymmetries and persistent anthropocentric biases limiting reciprocity in current design approaches. Plants frequently remain instrumentalised within design logics that continue to prioritise human experience, even when framed as care or engagement. These tensions are compounded by epistemic limitations. This study relies on designers' intentions and observable system behaviours rather than longitudinal evidence of plant wellbeing, and the question of what constitutes a “benefit” for plants remains unresolved. While plant science is advancing understandings of plant responsiveness and communication (e.g., Trewavas, 2003; Gagliano et al., 2017), translating these insights into design-relevant notions of plant welfare is still emergent. The taxonomy does not resolve this ambiguity but instead surfaces the assumptions through which benefit is currently defined, opening space for more ethically attentive practice.

Importantly, the taxonomy also reveals a broader constraint: design alone cannot fully achieve reciprocity. While designers can shape systems within their sphere of influence, the conditions structuring human–plant relations are embedded in wider cultural and governance frameworks. Advancing multispecies justice requires challenging the anthropocentric foundations of contemporary governance towards a shift from ownership to stewardship and from legislation to obligation (Sheikh et al., 2023b). Such a shift reframes nonhuman life not as passive recipients of human care but as participants in shared ecological relations. This point towards a deeper reorientation of and beyond design practice. Relationist approaches to urban governance emphasise interdependence, collective responsibility, and long-term care, moving from individualised benefits towards an ethics of obligation (Graham et al., 2024). These perspectives resonate strongly with the enduring knowledge systems of many First Nations peoples, which position humans as embedded within more-than-human worlds and bound by ongoing responsibilities to care for land and life. In this context, the challenge is not only how to design for better distributions of benefit, but how to reposition design within a broader cultural shift towards custodianship. Human–plant interaction design must therefore be understood as part of a larger project of reconfiguring relationships with the living world—acknowledging limits, respecting nonhuman agency, and fostering practices of care that extend beyond immediate human concerns.

The Beneficiary Taxonomy provides a starting point for this work. By making visible how benefits are imagined and distributed, it points to both opportunities and persistent blind spots. Ultimately, its contribution lies in prompting a more fundamental question: what would it mean to design not only for multispecies futures, but as part of a commitment to being “good ancestors” (Graham et al., 2024), that is, sustaining the conditions for life, human and nonhuman alike, across generations?

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript. Generative AI (ChatGPT, OpenAI, GPT-5.3; Claude, Anthropic, Opus 4.6 and 4.7) was used in a limited capacity to assist with language editing and refinement of the manuscript. All AI-assisted copyediting was reviewed, verified, and revised by the author(s), who take full responsibility for the accuracy, integrity, and originality of the work. No generative AI tools were used to generate data, analysis, or figures.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcomp.2026.1870559/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Karolina M. Zielinska-Dabkowska, Gdansk University of Technology, Poland

Reviewed by: Amardeep M. Dugar, Independent Researcher, Chennai, India

Nirit Binyamini Ben-Meir, Queen Mary University of London, United Kingdom

Filipi Miranda Soares, Mathematiques Informatique et Statistique pour l'Environnement et l'Agronomie, France