Data poets: Co-interpreting urban ecologies with generative AI prototypes in psychogeographic walks

Abstract

The ‘Data Poets’ project proposes a place-based design research methodology that explores the tension between the experiential nature of human sensorial engagement and the potential of generative AI to shape – or distort – socio-ecological knowing in urban spaces. This approach looks to the human as a practitioner and embodied agent, rather than a data source, in exploring urban environments. The project aims to challenge objectivist data narratives of urban ecology (the study of relationships between living beings, here focusing on humans, and the environment), emphasising the subjective, phenomenological aspects of urban life. The Data Poets are AI-powered devices designed for use in psychogeographic walks (produced in 2020, using GPT-2 and Google Vision), as well as an open interactive website inspired by the devices (produced in 2024, using GPT-4V). The devices serve as data collectors and dialogic ‘others’ that interpret, echo or contrast the sensorial experiences of participants. This interaction creates a critical and dialogic space where the AI’s output becomes a reflective text, intriguing for its capacity to both align with and diverge from human perceptions. The article discusses what can be perceived and recalled through human senses and cognition: the smells, sounds, textures and lived experiences of the city, and how AI can provoke dialogues to produce an interpretation of these experiences. Given recent debates about the deployment of AI technologies in urban studies, I aim to critique their epistemological and ontological implications in urban contexts.

Introduction

Recent advancements in artificial intelligence (AI) have reopened debates about its role in urban ‘observing, sensing, imaging, and mapping’ (Weng et al., 2024). With claims of abilities ‘to make sense, to facilitate, or to revolutionise, our understanding of the world’ (p. 2), AI is currently employed in tasks such as object detection, image classification and 3D reconstruction. Models like convolutional neural networks (CNNs), recurrent neural networks (RNNs) and transformers aim to enable new ways to interpret and engage with urban environments (Weng et al., 2024). Beyond the paradigm of the Smart City, ‘Urban Artificial Intelligences’ are emerging with the potential ‘to develop their own narratives and to act autonomously in the pursuit of emergent futures’ (Cugurullo et al., 2023: 14).

Offering a critical, practice-driven response to these developments, this article presents the Data Poets project, a prototype exploration of the integration of artificial intelligence (AI) in psychogeographic practices. When referring to AI from now on, I mean specifically Generative AI like Large Language Models (specifically, GPT-2 and GPT-4 were used) and Image Recognition/Machine Vision models (Google Vision and GPT-4V). This research employs AI-powered devices (the titular Data Poets ¹ ) during psychogeographic walks to collect and interpret sensory data. Their function is to facilitate the gathering of soft data: intangible yet meaningful information like feelings, opinions and perceptions from urban residents. These devices, activated by participants, serve as data collectors and dialogic devices, echoing or contrasting the participants’ experiences through the generation of poems (see Figure 1). The Data Poets project’s objective is not to quantify this data, but to question how AI models shape – or potentially distort – the production of meaning about place in these participant-led explorations.

Figure 1.

Visual abstract.

The devices help question and discuss how these forms of ‘non-biological intelligence’ (Cugurullo et al., 2023) might ‘experience’ and cohabit urban spaces: if ‘the control of the city . . . is also influenced by AIs whose logics and actions sometimes diverge significantly from ours’ (Cugurullo et al., 2023: 1), then investigating these divergences may be a necessary contribution to urban studies. Current debates in Science and Technology Studies (STS) further call for qualitative exploration of AI models. One such approach is ‘synthetic ethnography’, as described by De Seta et al. (2024). They define it as ‘a methodological approach for the qualitative study of generative models grounded on the experimentation with field devices’: a form of practice-based experimentation with Generative AI models.

The Data Poets project, while preceding this formalisation, can be understood as an early exploration that aligns with principles of synthetic ethnography. The project was started in 2019 as a self-directed research activity, part of a contract with the Glasgow School of Art, and continued later independently from 2020 as part of my artistic practice. I aim to critically recontextualise this practice project, evaluating it through the lens of 2024 technology and a broader philosophical and theoretical framework. This evaluation calls for a critical examination of certain aspects and assumptions underlying the original project; consequently, it does not advocate unconditionally for the use of AI in urban studies. Rather, it explores its potential applications and implications through two pilot psychogeographic interventions in Glasgow (in 2021 and 2024) and an interactive participatory website (launched in 2024). This article offers a reflection on the design process and its theoretical implications; however, it currently lacks sufficient empirical testing beyond the two pilot walks and the ongoing open website. Consequently, it should be viewed primarily as a positional proposal and reflective critique: it presents a pilot project.

I introduce a methodological approach that combines psychogeography with generative AI, which offers a critical theoretical reflection on the ways AI mediates perception and meaning. The Data Poets project aims to demonstrate how AI-generated poetic outputs, derived from visual data, can provoke reflection on the embodied, subjective and sensory experiences of place. Through a case study, I explore how AI-generated poems can mediate human interactions with place, not as a detached analytical tool, but as part of an assemblage of a sensory and material practice. In doing so, objectivist applications of AI are challenged, which invites a rethinking of how technology might contribute to participatory and situated ways of knowing.

In practice, the Data Poets project involves participants collecting visual inputs, which are then processed to generate textual data in the form of poems. Using devices described as ‘Data Poets’, or alternatively, participants’ phones, the input data is captured and fed into an image recognition model. This model’s output is processed by a Large Language Model to create poems. In the 2020 version of the project, these poems were printed, whereas the 2024 version displays them on a website. Participants engage with the project by taking the devices on walks or using their phones to record the necessary data. The project is an assemblage in which the Data Poets play a central but not exclusive role (see Figure 2), comprising (1) recruitment and explanation of the project’s aims, (2) facilitation of participant walks, (3) interpretation of generated poems and (4) collective reflection.

Figure 2.

Study procedure.

The two pilot studies discussed – conducted in 2021 and 2024 – provide initial findings that demonstrate the potential of this approach. The interactive project website (built in 2024) is open to the public, with contributions made from participants worldwide, which brings a new dynamic to the project and is also reflected upon in this article. However, given the experimental nature of these pilots, the data should be read as illustrative examples of the methodology’s possibilities rather than definitive insights into the specific places explored.

Theoretically, the article positions AI as neither neutral nor omniscient but as a contextual and partial ‘observer’, whose outputs reveal both the capabilities and limitations of computational systems in engaging with human experience. AI is examined as an active agent in generating socio-ecological knowledge. Drawing on debates in phenomenology, post-phenomenology and critical design, I critique the tendency to view AI as disembodied or objective and instead foreground the material, interpretative and socio-political dimensions of its use. This framing situates the article at the intersection of urban studies, critical AI research and participatory design, and offers a critical perspective on how AI can enhance or disrupt the experience of urban spaces.

The article is structured into six key sections to guide the reader from the practical implementation of the Data Poets project to the theoretical frameworks that inform and are challenged by it:

Part 1 (The Data Poets) focuses on the Data Poets project, its execution, participant input and reflections on its methodology. It establishes the project’s rationale, introduces the prototypes and shares insights from pilot walks and the website. Building on this empirical basis, Part 2 (Design methods and practice) situates the Data Poets project alongside other design research methodologies to clarify the broader context and examine relevant precedents for engagement and data production/collection in urban studies. Transitioning to a theoretical, human-centred lens, Part 3 (Experience) draws on Phenomenology and Post-phenomenology to discuss the embodied and situated aspects of walking and sensing in urban spaces, which serve to understand ways technology interfaces with and reframes these sensory engagements. Decentering the human, Part 4 (Being in place) then explores the ontological differences between human and AI existence. Contrasting AI’s atemporal and disembodied character with human lived experiences in urban settings enables us to question the authenticity and relevance of AI’s ‘perception’ of place. Then, having established the divergences between these two ‘interpreters’, the article progresses in Part 5 (On interpretation) to examine the act of interpretation itself. I discuss the interpretative models that shape how we read urban spaces to frame how meaning is actively constructed, reimagined, or contested by both human participants and AI-generated representations. Finally, the conclusion critically recontextualises the practice and learnings in light of the preceding theoretical discussion, and looks to future opportunities.

The Data Poets

The Data Poets project was first conceived as a research tool aimed at provoking playful engagement between people and their urban ecological milieu. I initiated the project in 2019 within a Glasgow School of Art research contract, and have carried the work forward independently since 2020. To achieve embodied engagement with the environment, the project’s physical form became a primary consideration. What physical shape should this object take to best facilitate a sensory connection to place? This endeavour prompted reflections on the forms AI might assume if it existed not just as disembodied voices or screen-based entities, but as tangible elements integrated into the physical world (see Figure 3). The organic aesthetics of ceramics and the cultural significance of pots and vessels inspired a physicality that provides a provocative intersection between AI and human senses. This concept deviates from popular AI imaginaries and representations, disembodied voices like Apple’s Siri or HAL 9000. ²

Figure 3.

Initial sketches for the design of the Data Poets.

In this context, the Data Poets function as interpreters of the city-text, utilising AI to idiosyncratically read and interpret urban environments. Initially, engaging with Bernard Stiegler’s (1998) notion of technics as a ‘prosthesis’, AI was envisioned as an extension of the human body. This process led to the creation of a collection of AI entities, each with specific sensory roles: the Aesthete observed visual stimuli, the Bard listened to auditory events, and the Fountain Printer translated these sensory inputs into tangible, printed poems. The Data Poets project was conceived as a set of physical prototypes designed to capture various sensory data inputs – visual, auditory – and generate textual outputs (see Figure 4 for a booklet of printed poems from a test walk), thus encouraging user interaction and engagement with urban environments. The physical forms of these devices were designed to reflect and challenge user interactions with technology.

Figure 4.

Printed Poems from a 2020 test walk, with additional illustrations by Charline Roussel.

The design of the Data Poets (see Figures 5–7) emphasised physical interaction with AI, moving away from disembodied screen-based or voice-based forms towards designs that engage the whole body. The prototypes, fabricated through selective laser sintering (SLS) and nylon powder, achieved a unique tactile quality (a slightly grainy texture akin to unglazed ceramic). This was meant to encourage participants to explore these objects in a more embodied manner. For instance, the Aesthete (Figure 5) is a neon yellow, symmetrical device equipped with a camera, intended to be held at waist height with both hands, reminiscent of a dowsing rod. This design metaphorically suggests its use in navigating and discovering the aesthetic aspects of the cityscape. Two crescent-shaped buttons are integrated to trigger the camera. The ripples on the main interactive part suggest fluidity, as if the device is constantly absorbing and processing visual data.

Figure 5.

The Aesthete.

Figure 6.

The Bard.

Figure 7.

The Fountain printer.

Similarly, the Bard (Figure 6) features an asymmetrical form that combines elements of a canister and a microphone. It includes a mesh of holes representing the microphone component. The form factor is designed to be held in one hand. The texturing and shape encourage the perception of gathering auditory data, which is conceptually stored in the jug-like shape, and later transformed into poetic outputs. ³

The final prototype, the Fountain Printer (Figure 7), is a vivid red, oblong shape designed to be worn like a pendant, housing a receipt printer. Its design includes plus-shaped buttons and a curved inward lid that symbolically acts as a receptacle for received information. The receipt paper hanging out resembles a tongue inside the red lips of the object lid, suggesting that the device ‘speaks’ or ‘outputs’ the poetic interpretations it generates. The inward curvature and the visible paper output emphasise its role as a receiving device.

The 2020 and 2024 iterations of the Data Poets project used three models: respectively Google Vision and GPT-2, and GPT-4/GPT-4V. Each played the following roles:

2020 version (see Figure 8):

• Google Vision (Cloud Vision API): This model uses Convolutional Neural Networks (CNNs) for object detection and image labelling. Its outputs often resembled a ‘shopping list’ of elements within the frame (e.g. ‘Jacket - 97%; Outerwear - 96%; Standing - 96%; Night - 92%; Pedestrian - 85%’).

• GPT-2: This LLM generated poetic outputs based on the labels and contextual information provided by Google Vision. The resulting poems were often nonsensical but revealed interesting juxtapositions between the image, the structured data and the generated text.

2024 version (see Figure 8):

• GPT-4V: The multimodal version of GPT-4 added scene description capabilities, which were used as a basis (hidden from the user) for generating poems. Unlike its predecessor, GPT-4V’s outputs were coherent and clearly contextually relevant to the provided images.

Figure 8.

Comparison of the 2020 and 2024 iterations of the Data Poets, focusing on the evolution of the data flow.

The initial integration inside physical prototypes of the Large Language Model GPT-2 for text generation and Google Machine Vision for image recognition enabled a conceptualisation of AI as something participants could feel, touch and interact with in physical spaces. The prototyping process revealed challenges of representing AI in non-traditional forms and offered insights into how people perceive and interact with AI. Reflecting on the initial project text, there is a tendency to anthropomorphise the technology:

If cameras had a poetic soul, would we show them things differently? What would be our motivation for taking them for a gentle walk along our favourite path? If microphones had a musical ear, how would they sing? Would we make them listen to the roar of the highway or the happy chatter of the birds in the park?

Text for the Data poets (Welisch, 2020)

While this (unintentionally misleading) anthropomorphism was intended to make the objects seem friendly and approachable, it might have been more effective to embrace the uncanny and provoke critical reflection. Pushing the boundaries of user comfort might have probed a more reflective engagement with the implications of AI.

Three pilot walks were conducted with the Data Poets. The first was held in 2021 (following COVID-19 safety guidelines at the time) around the Glasgow Canal (see Figure 9), in the context of the Glasgow Architecture Fringe. The five participants were mainly tourists and architecture enthusiasts, self-selected as they decided to come to the event after seeing it advertised on the Architecture Fringe website. The general area for the dérive was selected and advertised before the walk. The physical models of the Data Poets were brought to the walks, and sample poems were shown, but the prototypes were not fully functional on the go, so the poems were emailed to participants after their participation. This distance between participation and receipt of the poems introduces a temporal and experiential gap, which might have influenced the participants’ connection to the sensory data they collected and diminished the reflective aspects of the project.

Figure 9.

The 2021 Glasgow Canal Walk.

The second and third pilot walks were held in 2024 in Govan (see Figure 10). The context for these walks was an existing academic research project looking at the role of a historical building in Govan. The walk thus started from this building. Participants were the academics and stakeholders already involved in this project (six total participants). Instead of the physical prototypes, a website version of the programme was crafted, which participants could access through their smartphone browsers. The poems were generated during a debrief session. The following quotes come from these sessions.

Figure 10.

One of the 2024 Govan Walks.

Reflections from participants provide practical insights into this methodological framework. For instance, one participant noted,

I thought I’d take a recording as well. And what it was picking up was the thing that I noted was the children playing, screaming, birds singing, but it’s also this constant drone of machinery that I think everybody else here will have tuned out . . . and actually all sounded quite harmonious unusually. (Participant 1, 16 April 2024)

being attentive to sounds can reveal a new layer to soundscapes, with a renewed aesthetic appreciation.

Another participant commented, ‘I can see where this is kind of valuable for acquiring a kind of set of, of kind of fixing perceptions, you know, or at least fixed views of things’ (Participant 2, 16 April 2024). Capturing sensory data can both fix and challenge perceptions. ‘This staircase reminded me of New York tenements, and it made me think Glasgow tenements don’t usually have those. So I just felt compelled to photograph it’ (Participant 3, 16 April 2024). Personal memories and associations can be triggered by urban elements, and this can enrich the interpretive process by connecting individual experiences with broader narratives.

Finally, the website version of the Data Poets was made publicly available online so that (self-selected) participants from around the world could engage with the project and generate poems. The site was promoted on social media and relevant blogs to attract participants (153 total participants as of 26 January 2025).

Reflecting on these pilot walks, it becomes clear that the AI devices are only one element in a broader socio-technological assemblage that includes websites, emails, other digital tools, as well as human and broader socio-ecological factors (like the COVID-19 pandemic during the first pilot walk). This assemblage plays a significant role in shaping human-technology interactions. The design and implementation of the apparatus – encompassing the physical devices but also the supporting infrastructure – influence the participants’ experiences and the overall success of the project.

As the designer of the Data Poets project, my role extends beyond creating the objects to include recruiting participants, explaining the concepts and facilitating the interaction process. This role inevitably affects the outcomes, as the success of the objects is closely tied to their deployment and the context provided by the facilitator. For example, the distance introduced by the email delivery of poems in the first pilot walk, while practical, may have diluted the immediacy and engagement intended by the project.

The transition from physical prototypes to a web-based platform (see Figure 11 for the interface of the platform) utilising AI models GPT-4V and OpenAI’s machine vision systems in 2024 aimed to overcome the logistical limitations of the initial prototypes. While this shift enabled real-time poem generation and facilitated data collection, it transformed the experience from a tactile interaction to a screen-based one. This digital mediation potentially detracts from the immersive, psychogeographic intent of the project.

Figure 11.

The Web-based interface of the 2024 Data Poets, retrieved from https://datapoets.gaston.pro/.

Transitioning from physical prototypes to digital interfaces presented technical and conceptual challenges. The initial prototypes, designed using 3D printing technologies and incorporating Arduino and Processing, emphasised hands-on interaction and the physicality of space, key components in psychogeographic exploration. The shift to GPT-4o and OpenAI’s GPT-vision required maintaining the poetic and unpredictable nature of AI outputs, a quality that was somewhat diluted with more sophisticated AI models. The initial outputs from GPT-2, though often nonsensical, lent themselves to creative interpretation (see Appendix 1 for poem comparison). Conversely, GPT-Vision’s detailed descriptions of scenes often resulted in clichéd poetic forms from GPT-4o (overuse of poetic tropes and imagery, including metaphorical use of ‘tapestry’ and ‘whispers’), necessitating the development of poetry rules, inspired by the Oulipo movement (Baetens, 2012; Seaman, 2001), used in the prompt to reintroduce randomness and stylistic variation. There is a problem of frictionless fluency, where the newer models’ eloquence leave little space (or creative friction) for working out interpreted meaning.

Moving to the empirical material gathered through the website version of the project, the following quotes are selected from some of the submissions of the 153 users (visitors having submitted at least one image) of the site. They illustrate some of the engagement that people have had with the places they chose to upload to the platform. The prompt was ‘Add some detailed context for the data poets. Why was the moment significant?’

1. ts my private backyard. My secret garden

2. A peaceful hike in my new home State of Oregon.

3. This is The city that as born and grew Up (João Pessoa) Despite this city’s flaws (Traffic Is Rubbish) it is a good place to live.

4. Saw this man fixing roofing tiles on the way home, thought he looked like Santa Claus on the roof.

5. tall sheer cliff, arduous climb. I lost my phone sliding down.

6. I got married and attended my sister’s funeral on the same day.

These quotes exemplify people’s experiences of place being produced by personal memories, emotions and social contexts. Unlike the LLM used, which can only generate direct associations from visual data, humans imbue places with personal significance and symbolism derived from life events and interactions. This is evident in the humour found in everyday observations, the ambivalent emotional experiences linked to significant life events, and empathetic moments spent with nature and non-human elements (e.g. birds, walking with dogs). Finally, the digital platform offers logistical advantages: having the platform online has opened up greater participation to people all over the world (see Figure 12). The physical prototypes, however, still have a unique potential for immersive, embodied interaction.

Figure 12.

Map of the 237 poems on https://datapoets.gaston.pro/ as of 26 January 2025 – of note is the current prevalence of poems from the Global North, mostly the United States and Europe.

Design methods and practice

The Data Poets project situates itself within a broader tradition of critical and speculative design practices and theories. This section outlines key precedents, exploring how they could inform this study’s approach to collecting and interpreting data to create engaging, provocative representations of place. I will critically examine these methods, highlighting their contributions, and demonstrate how they inform and contrast with the approach taken in the Data Poets project.

The project inherits key conceptual and methodological approaches from Alison Powell’s (2018) data walking framework, while extending its inquiry to include the role of artificial intelligence. At its core, Powell’s work on data walks challenges the dominance of top-down data paradigms by introducing a bottom-up methodology for engaging with urban data. As Powell articulates, the ‘data walkshop’ is a process of ‘exploring and defining data, big data, and data politics from the perspectives of groups of citizens’ who engage with their urban environments through walking, observing and reflecting (Powell, 2018: 213). Similarly, the Data Poets project adopts walking as a method for sensory and critical engagement, but adds a material presence for AI as a mediating and interpretive ‘participant’ in the exploration of place.

For Powell, data walking ‘creates a phenomenology of data’ by linking movement, context and knowledge production (Powell, 2018: 213). While Powell focuses on the politics of datafication and the subjective experience of participants, the Data Poets project includes ‘non-biological intelligences’ (Cugurullo et al., 2023) to explore how AI systems ‘perceive’ urban spaces and contribute to the assemblage of sensory, material and social practices.

Powell’s reflections on the subjective and contested nature of data are particularly relevant to the epistemological questions posed by the Data Poets. She notes that ‘the workshops . . . highlight the inevitable consequences of deciding that one thing, rather than another, might become data’ (Powell, 2018: 217). The Data Poets project aims to expand on this idea by questioning how AI, with its own operational logics, decides what is worth interpreting or transforming into poetic outputs. This raises new questions about how AI, as an agent, may increasingly contribute to urban imaginaries and data narratives. The project probes the extent to which AI systems might offer alternative ways of knowing, while simultaneously reflecting the biases, limitations and power dynamics inherent in their design and operation. The Data Poets project shares the ambition of ‘surfacing the everyday experiences and reflections’ of participants (Powell, 2018: 214), but extends this to include the perspectives of AI systems as ‘participants’ in the production of urban knowledge. This transition reflects the broader shift from the Smart City to what Cugurullo et al. (2023) describe as ‘Urban Artificial Intelligence’, where human and non-human intelligences co-produce urban futures.

The Data Poets are also indebted to the design methodologies of Anthony Dunne and Fiona Raby, Laurens Boer and Jared Donovan, and the Interaction Research Studio. These researchers have demonstrated how product design artefacts can be used as speculative, critical, or provocative tools. Anthony Dunne and Fiona Raby’s Critical Design explores the ideological underpinnings of consumer products, encouraging users to reflect on how these objects shape their experiences and perceptions. Dunne (1999) critiques mainstream industrial design for being ‘comfortable [propagandising] needs designed by others, thereby maintaining a society of passive consumers’, while advocating that ‘design research in the aesthetic and cultural realm should draw attention to how products limit our experiences’. By creating a poetic distance between people and electronic objects, Dunne suggests that ‘sensitive scepticism might be encouraged, rather than unthinking assimilation of the values and conceptual models embedded in electronic objects’. He argues that design is always ideological. ‘User-friendliness helps conceal this fact. The values and ideas about life embodied in designed objects are not natural, objective or fixed, but man-made, artificial, and mutable’.

In ‘Designing for Homo Ludens’, Bill Gaver (2002) emphasises the importance of play and intrigue in design. Gaver argues that design should seek to engage users at multiple levels, ‘from the aesthetics of form and interaction, to functionality, to conceptual implications at psychological, social and cultural levels’, and open new forms of interaction. The designer’s role in this is not prescriptive or reactive: it isn’t ‘pre-scribing cures for people’s ills’ or ‘developing technologies that people know they want’, instead Gaver suggests that designers should act as ‘provocateurs, seeking out new possibilities for play and crafting technologies that entice people to explore them’.

An aimless walk in the city centre, a moment of awe, a short-lived obsession, a joke – all are defining and valuable facets of our humanity, as worthy of respect as planning, logic or study. Play is [. . .] an essential way of engaging with and learning about our world and ourselves.

[. . .] Not only should technologies reinforce pleasures that people know, but they should suggest new ones.

Continuing on the theme of Design suggesting alternative ways of thinking, Laurens Boer and Jared Donovan’s Provotyping methodology places emphasis on embodying conflicting perceptions in tangible forms to provoke critical reflection and dialogue. Their work in participatory innovation illustrates how traditional design methods often overlook conflicting perspectives, whereas provotypes (provocative prototypes) reveal underlying issues and stimulate creativity. In their study on indoor climate systems, Boer and Donovan (2012) developed two provotypes which embodied tensions related to air quality and comfort. These objects provoked family members to reflect on their environment, discussing the visibility and impact of indoor climate information (Boer and Donovan, 2012). This principle is taken forward in the Data Poets (themselves a form of provotype), where the AI devices aim to create a dialogic space that contrasts human and AI perceptions, and embodies tensions between human sensorial experiences and AI interpretations.

The Interaction Research Studio’s probe tools capture and reflect personal experiences to understand the subjective dimensions of human interaction with environments. The Domestic Probes study (Boehner et al., 2012) aimed to provide insights into the home context and explore new roles for technology within it. Participants received probe packets containing items like a ‘Dream Recorder’ and a disposable camera with specific photo requests. Figure 13 shows an example kit of the probes used (left) alongside some responses (right). Over a month, participants interacted with these items, returning responses that offered designers evocative glimpses into their home lives, which then inspired speculative design proposals.

Figure 13.

The Interaction Research Studio’s Domestic Probes (Gaver, 2007).

Similarly, TaskCams (Figure 14) are open-source digital cameras designed for Cultural Probes studies, offering a straightforward and customisable tool for engaging participants in user studies. TaskCams display questions or prompts and tag each photo with the displayed text, allowing researchers to capture contextual data about participants’ lives and environments. These tools address the limitations of traditional disposable cameras by providing a reusable and adaptable option that meets the methodological needs of Cultural Probes studies. The Everyday Design Studio at Simon Fraser University used TaskCams in a study exploring alternative living situations, customising them to align with the values of different participant groups, such as zero-waste practitioners (Boucher et al., 2018).

Figure 14.

TaskCams (Interaction Research Studio, n.d.).

While probe tools effectively capture personal experiences, they can lack the capability to process and interpret this data with the participant. The Data Poets’ use of AI aims to address this gap with the potential to provide reflective dialogue near-real-time.

Another key influence, the Datacatcher project (Gaver et al., 2016) involved deploying custom-built, location-aware devices (shown in Figure 15) to democratise big data and generate engagement with socio-political issues. These handheld devices displayed local data and prompted user interaction through playful questions. The project’s methodology included batch deployment and documentary filmmaking to capture participant responses. While some participants found the devices uninteresting or questioned the data’s accuracy, others appreciated the design and used them to start social interactions and discussions about local conditions.

Figure 15.

The Datacatcher (Interaction Research Studio, 2015).

Participants engaged with the devices not only as conveyors of data but also questioned their accuracy and relevance; a response researchers viewed as a valuable critique of big data’s growing societal and political influence. Gaver et al. (2016) emphasise that ‘research devices are not simply deployed’, but are embedded in ‘a complex ‘world’ created through strategies and tactics that shape their audience, identities and meanings’. This led participants to consider the prototypes ‘both as products and as embodied research’. This perspective emphasises the importance of considering the deployment process as part of a broader socio-ecological system, an insight particularly relevant to the Data Poets project, which has yet to be tested at scale.

Experience

Phenomenology and post-phenomenology

Phenomenology, as it has evolved within geography, provides a theoretical framework for understanding the interactions with the Data Poets and urban ecologies. Traditionally, phenomenology focused on the subjective experience of individuals as they perceive their environment. Hepach and Kinkaid (2024) argue that phenomenology has significantly influenced geographic debates on key concepts like landscape and place, serving as a counterpoint to structuralist and positivist approaches. Engagements with post-phenomenology move beyond the subject-centred perspective, considering the relational and emergent qualities of human–environment interactions. Post-phenomenology acknowledges the autonomous existence of objects and their capacity to influence human experience. Post-phenomenology rethinks intentionality ‘as an emergent relation with the world, rather than an a priori condition of experience’ (Ash and Simpson, 2016: 48). This approach shifts the focus from purely human-centred experiences to include the roles that objects and environments play in shaping those experiences. Therefore, throughout this project, phenomenology will be used not to single out the experiences of the subject as they appear to them but as a way to contrast perceptions with mediated interpretations and representations.

In the context of sensory perceptions, the Data Poets project considers sight, sound and body positioning in relation to the designed artefacts as key components of the interactions facilitated. Phillips (2015) emphasises the pedagogical value of a multi-sensory approach to geographical fieldwork – incorporating hearing, touching, looking differently, smelling and getting lost – to create a curiosity-driven, enthusiastic exploration of urban environments. The value of these approaches, for Phillips, is demonstrated ‘when Iain recognized a stranger in Manhattan, or when Rosa noticed a play of colours on the ground, when Henry lost himself in the touch of a rock or when Ben found a kind of music in the city’ (p. 628). This aligns with my methodology, which includes both visual and sound recordings and the embodied interactions through the physical forms of the AI devices. The sense of smell, difficult to capture through commercially available sensors, is one of the senses participants were encouraged to think about, mainly through facilitation and discussion, to augment the devices’ readings.

Factors such as time of day, weather, temperature and other perceptual data were considered to help provide a more context-rich understanding. Sensory geographies also emphasise the significance of haptic knowledge (understanding through touch and bodily engagement), which helps build a comprehensive sensory experience of urban spaces (Paterson, 2009). Paterson notes the difficulty in recording and expressing these experiences through language, and suggests using poetry may be better suited to transcribe ‘one set of sensations into another language’, as ‘evoking and describing sensuous dispositions and haptic knowledges benefits from the styles and methods involved in experimental or creative writing’ (p. 785). The use of poems aims to bridge this expressive gap.

By situating these sensory expressions within the broader theoretical framework of urban ecology, we can better understand the relationships between living organisms and their environments. Urban ecology encompasses the study of these relationships (Timon McPhearson et al., 2016). My project adopts a socio-ecological perspective, focusing on the interactions between humans and their urban surroundings. This perspective integrates the relationships among social, ecological, economic and built infrastructure systems. While the current scope of the Data Poets primarily involves human participants and AI, future work should consider incorporating non-human actors such as plants and animals to fully address ecological dynamics. This inclusion would align with the broader ecological framework, which considers organisms at various levels, from individual to biosphere.

Perspective

In this part, I contextualise this project, which aims to understand lived relations to urban ecologies, within a broader theoretical discourse that specifically addresses the perspectives from which one experiences urban spaces. Michel de Certeau notably offers a framework for situating the perspective of the observer of urban space through his distinction between the detached voyeur and the engaged walker. De Certeau (1984) contrasts the elevated, detached view from the top of the World Trade Centre with the intimate, ground-level experience of walking through the city. From above, the city appears as a coherent, totalising text that invites a voyeuristic gaze detached from the everyday practices of its inhabitants (de Certeau, 1984: 92). This elevated view, which de Certeau describes as ‘seeing the whole’ and ‘looking down on’, provides a seductive yet artificial sense of mastery and control over the urban landscape. This flattening perspective can be likened to the digital map, the satellite view of Google Maps, or the voyeuristic surveillance of delivery apps like Deliveroo, Uber and Amazon, allowing you to track the labour of your assigned driver. The understanding of the lived, embodied experiences of urban spaces is thus limited. In contrast, the ‘ordinary practitioners of the city’, the walkers, engage with the urban environment through a tactile, sensory and embodied practice that resists the voyeuristic reduction of the city to a spectacle (de Certeau, 1984: 93). These walkers create a story through their movements, inscribing the city with ‘unrecognised poems’ and paths illegible from a detached vantage point.

The Data Poets project juxtaposes these perspectives, with AI devices perhaps being the detached voyeur and human participants embodying the immersive walker. Donna Haraway’s concept of situated knowledge further critiques the ‘view from nowhere’ (or ‘God Trick’): the disembodied, seemingly objective perspective often associated with traditional scientific inquiry and data-driven approaches (Haraway, 1988). Haraway argues for a more situated, embodied approach to knowledge production that acknowledges the partial, situated nature of all observations, like that of de Certeau’s description of the embodied practices of walkers. The situated knowledge sought by the project is that of the participant: What can these walks reveal of the lived practices and tacit connections people have to places visited?

Daniel McQuillan’s (2017) ‘Counter Mapping the Smart City’ provides a complementary discussion of the Perspective of the Map, in which he critiques the prevailing use of Smart City technologies for mapping and surveillance, reminiscent of the historically colonial tendencies of mapping. These often serve capitalist and control-oriented interests. He suggests counter-mapping as an anti-hegemonic tool that can repurpose these technologies to regenerate social capital by mapping communal assets and mobilising engagement. Counter-mapping (Peluso, 1995), coined by Nancy Lee Peluso, is a way to use the masters’ tools against them, of appropriating ‘the state’s techniques and manner of representation to bolster the legitimacy of “customary” claims to resources’ (Peluso, 1995: 384). Counter-mapping is the practice of creating maps to challenge and redefine established narratives imposed by dominant power structures. This practice is particularly meant for marginalised or indigenous communities to assert their territorial claims and articulate their perspectives. By shifting the focus of mapping from top-down surveillance to on-the-ground participation, counter-mapping challenges the deterministic narratives imposed by Smart City infrastructures and opens up new avenues for understanding and experiencing the city (McQuillan, 2017). The Data Poets project attempts to enact a form of this practice, perhaps not by producing a more ‘accurate’ map, but by generating counter-data: poetic, subjective and often absurd outputs that challenge the logic of quantifiable, objective urban representation.

The phenomenological stakes of this practice can be expressed through Baudrillard’s concept of hyperreality. If the map has indeed become the territory, then we owe it to ourselves to better shape the map:

Simulation is no longer that of a territory, a referential being or a substance. It is the generation by models of a real without origin or reality: a hyperreal. The territory no longer precedes the map, nor survives it. Henceforth, it is the map that precedes the territory – precession of simulacra – it is the map that engenders the territory. (Baudrillard, 1994: 1)

Referencing Borges’ ‘On Exactitude in Science’, where a detailed map commissioned by an overzealous cartophile emperor eventually engulfs the entire empire, Baudrillard (1994) prompts us to question whether we experience the empire itself or its map, if the representation even has an original reference at all. As we navigate using Google Maps, does the representation engender the represented? If so, shouldn’t we assert control over the map, reclaiming its production to ensure that our experiences align with our intended reality? Counter-mapping perhaps offers a means to regain this agency, appropriating the tools of the hyperreal to redefine and represent our spaces according to our collective needs and perspectives.

The Data Poets project aims to take on McQuillan’s advocacy for counter-mapping by enacting an idiosyncratically anti-functionalist data-mapping, foregrounding the subjective, embodied experiences over hegemonic, top-down narratives. The Data Poets present and produce alternative representations rather than capture and mine the city as a computable resource. By providing AI-powered devices to participants, the project can hopefully provide an alternative to surveillance-oriented Smart City mappings.

Being in place

The concept of the dérive, defined by Guy Debord’s 1956 ‘Theory of the Dérive’, offers a method for engaging with urban environments through purposefully aimless wandering, guided by the flows and barriers of the infrastructure. This practice emphasises playfulness and the awareness of psychogeographical effects: the emotional and behavioural impact of geographical settings. Debord’s description of the dérive as a movement driven by the currents of urban landscapes allows for an exploration of cities beyond traditional, functional navigations (Debord, 2007). The dérive is the fundamental walk-based methodology used in the deployment and context of the Data Poets, which are technical objects of Socio-Ecological knowing.

However, any use of technology to mediate or interface with the world (here, more specifically, urban ecologies) inevitably brings to mind Heidegger’s critique of the instrumental definition of technology. In ‘The Question Concerning Technology’, Heidegger warns against the instrumental view of technology: viewing technology as a means to an end. He introduces the concept of ‘standing reserve’, where, through technology, we treat the world and ourselves as a resource to be ‘revealed’ or exploited. This is caused by ‘enframing’, the technological perspective through which the world is shaped and projected. A straightforward example is provided by Abraham Kaplan (1964) with his ‘law of the instrument’: ‘Give a small boy a hammer, and he will find that everything he encounters needs pounding’ (p. 28). The example given by Heidegger is the use of the Rhine for hydroelectric power: the Rhine is no longer perceived as a poetic and cultural symbol but as a potential source of electricity through the hydroelectric dam. This instrumental view reduces natural elements to utilities, awaiting exploitation, and obscures other forms of revealing, such as artistic creation (Heidegger, 1977). Heidegger contrasts this with ‘poiesis’, the bringing-forth that occurs in artistic creation, which reveals the world in a non-instrumental way.

If traditional mapping reveals by constraining and delimiting (what Heidegger would call enframing) resources, borders, property and nations, then counter-mapping reveals in a way that challenges and creates new understandings and possibilities. The maps critiqued by counter-mapping impose fixed boundaries and static definitions; they reinforce existing power structures and socio-political realities. In contrast, counter-mapping seeks to deconstruct these limitations, offering new ways of seeing and understanding space that contest hegemonic narratives.

The Data Poets project embodies a tension between technological enframing and poetic revealing. The AI devices used in psychogeographic walks collect data and engage participants in a dialogic process that echoes or contrasts their sensory experiences, thereby creating a reflective space to engage with the enframing of AI’s computational perspective. Importantly, one must also reckon with the apparent spatial disembodiment of the AI models themselves. Seemingly housed within the devices or in an incorporeal, obscuring and ambiguous ‘cloud’ (Bridle, 2018: 8), the labour and environmental costs of the models are simply displaced, made possible by an invisibilised but very material infrastructure of data centres, human data-workers and fine tuners (Rowe, 2023) mostly in the global south, rare-earth mineral mining (Crawford, 2021) and more. Crawford argues that the imaginaries of artificial intelligence are filled with ‘algorithms, data, and cloud architectures, but none of that can function without the minerals and resources that build computing’s core components’ (p. 30). These costs, ‘from harvesting the data made from our daily activities and expressions, to depleting natural resources, and to exploiting labor around the globe’ (p. 31), should not be overlooked in the design of these devices and their deployment. There is an opportunity to deconstruct and engage with the technologies that make the devices possible.

Don Ihde’s (1990) concept of embodiment, hermeneutic and alterity relations (see Figure 16) in human-technology interactions helps us further negotiate the role and relationship between the participants and Data Poets. Embodiment relations occur when technology becomes an extension of our bodies, seamlessly integrated, for example, when wearing glasses. Hermeneutic (usually meaning the interpretation of religious texts, here meaning textually interpreted more broadly) relations involve technology providing data or information that requires interpretation, like a thermometer, which does not convey the sensation of heat but instead offers numerical data that users interpret.

Figure 16.

The Data Poets call for a mix of Embodiment, Hermeneutic and Alterity relations.

In addition to these, Ihde describes alterity relations, which are particularly relevant to AI. In alterity relations, we interact with technology as if it were another entity or being. This is evident in interactions with voice assistants like Siri or Alexa, where users converse with the technology as if it possesses its own agency, despite knowing it is a machine. The world is not directly the focus of this relation. The AI devices in the Data Poets project encapsulate these three modes of interaction. They embody sensory extension and require interpretative engagement, while also introducing an element of alterity, as participants interact with the AI as an external ‘other’.

The integration of AI in this context raises questions about the relevance and authenticity of AI in capturing and interpreting the human-environment dynamic. AI, particularly Large Language Models (LLMs), represents a form of atemporal existence, lacking the continuity of past and future experiences that characterise human temporality.

While GPT-4V’s (the model currently used in the Data Poets project) architecture remains undisclosed, current MLLMs (Multimodal Large Language Models) are trained on large-scale image-text pair datasets (Yin et al., 2024). Training typically involves: (1) Pre-training, aligning modalities such as vision and language; (2) Instruction tuning, teaching models to follow natural language instructions; and (3) Alignment tuning, refining responses to align with human preferences via feedback mechanisms like reinforcement learning. These stages allow MLLMs to demonstrate ‘visual reasoning’ and adapt to unseen contexts.

Floridi (2023) outlines the philosophical implications of such training approaches: models operate statistically, processing patterns in data to generate outputs. He argues that these models ‘can do statistically – that is, working on the formal structure, and not on the meaning of the texts they process – what we do semantically’ (p. 1), lacking semantic understanding or the capacity to engage with the world in a lived, meaningful way.

The datasets underpinning such models reveal their significant challenges. While we cannot definitively confirm the use of LAION-5B in GPT-4V’s training, its prevalence in the field and its composition exemplify the commercial and cultural biases inherent in such datasets. Buschek and Thorp (2024) explain that: ‘Training a model on LAION-5B is meant to give it a comprehensive representation of the world, to build a kind of vocabulary of things and concepts’. Yet, this scale means that ‘human curation of the dataset borders on the impossible’. Consequently, the dataset reflects ‘how search engines see the world’ rather than human perspectives, driven by ‘commercial logics’.

LAION-5B exhibits clear biases shaped by its construction. Approximately 45% of websites in its foundational Common Crawl dataset are in English, privileging English-speaking culture over the other 107 languages combined (Buschek and Thorp, 2024). This training data structure shapes these models’ interpretation and generation capabilities in ways that do not reflect broader human experiences.

LLMs engage in hermeneutical interpretations devoid of lived experience, operating in an amotivated and reactive manner. Conversely, Heidegger’s notion of ‘being-in-the-world’ describes the essence of Dasein (literally being-there, the human mode of existence) as a being fundamentally embedded within its environment, characterised by temporality, phenomenological experience, motivation, social contextuality, embodiment and an active engagement with the world. This concept presents human existence as present in both physical and social matrices, influencing and being influenced by them (Heidegger, 1962). The following table seeks to parallel both modes of existence, Human and AI, to delineate key differences in their ontology.

Contrasting with this embedded human experience is the operational framework of AI (Figure 17), specifically Large Language Models (LLMs) like the ones used in the Data Poets project. This disjunction between human and AI experiences raises critical questions about the efficacy and authenticity of AI in capturing and interpreting the human experience of the environment.

Figure 17.

A (over)simplified comparison of AI and human modes of existence/ontology.

Susan Sontag’s critique of photography provides a relevant parallel discussion between capture and experience (Sontag, 1977): ‘Photography has become one of the principal devices for experiencing something, for giving an appearance of participation’. She observes that tourists ‘feel compelled to put the camera between themselves and whatever is remarkable that they encounter’, giving shape to the experience in that they ‘stop, take a photograph, and move on’ (pp. 6–7). Photography flattens time and turns the world into a ‘series of unrelated, free-standing particles; and history, past and present, a set of anecdotes and faits divers’ (p. 17). Sontag argues that the act of taking photographs can fragment reality into discrete, unrelated moments, compartmentalising the interconnectedness and continuity of experience. The camera, by isolating moments, confers on each the character of a mystery, making reality atomic and manageable but also opaque (Sontag, 1977). The risk is reducing interconnected experiences to separate data points and images. LLMs present similar risk: each prompt and response is a discrete atemporal point, artificially connected through the ‘memory’ of a context window. The Data Poets, by presenting their interpretations as poetic outputs, acknowledge their inherent fragmentation and bias; they offer one partial perspective rather than making objective claims to reality.

Phenomenology, the study of things as they appear to us (as discussed in Part 3.1: Phenomenology and Post-phenomenology) provides a potentially flawed yet valuable framework for understanding these interactions. Husserl’s (1931) original conception of phenomenological reduction advocated for ‘bracketing’ experiences (pp. 107–108), disregarding or setting aside one’s doubts and unique point of view to reach an ideal, albeit unreachable, phenomenological understanding. John D. Caputo’s (1987) radical hermeneutics might serve as a supplementary approach, embracing the multiplicity of readings and the biases each participant brings. This is particularly important in urban settings where multiple layers of historical, social and personal narratives overlap. By incorporating Caputo’s interpretive framework, the Data Poets project aims to foreground these multiple perspectives, enhancing our understanding of urban environments as complex, layered and subjective texts. In practice, this theoretical integration informs the methodology of the Data Poets project. The AI devices employed in these walks extend human sensory capabilities yet require interpretative interaction from participants (in that they have to make sense of the generated poems and choose what to input).

The critique of technology’s potential for deskilling and reducing embodied perception, notably articulated by Don Ihde, further problematises the role of AI in this context. Ihde provides the example of South Pacific islanders who replaced their complex art of phenomenologically reading waves with the simpler hermeneutic practice of using a compass for navigation. In this process, the islanders likely gained productivity, but lost a practice which granted them a unique aesthetic appreciation of the milieu they are embedded in. The introduction of AI technologies, which aim to mimic human cognition, may lead to similar consequences. While tools like calculators enhance abstract capabilities without fundamentally altering our relationship with mathematics, AI technologies that replace complex human activities may erode the skills and sensibilities of engagement with environments (Ihde, 1990).

The Data Poets project must navigate these tensions, deploying AI to ‘enhance’ human sensory engagement while bringing attention to the reductive tendencies of technological enframing. By creating AI devices that extend and interpret human experiences in a poetic form, the project aims to strike an ambivalent balance between technological dialogue and poesis with a preservation of the participants’ sensorial and emotional qualities.

On interpretation

Poetics

The choice of the name ‘Data Poets’ warrants a deeper analysis of poetics. Aristotle, the author of one of the founding texts about poetry, defines a poet as one who ‘by conscious art or mere habit, imitates and represents various objects through the medium of colour and form, or again by the voice; so in the arts above mentioned, taken as a whole, the imitation is produced by rhythm, language, or “harmony,” either singly or combined’ (Aristotle, 2008). This definition sees poetry as a form of imitation, representing the real. However, poetry extends beyond imitation; Poetry in the context of the Data Poets is not just about representation but also about providing a way of understanding or interpreting the real. Poetics’ attention to the form of the representation, which Aristotle emphasises heavily, suggests a play on structure and rules within which to express oneself.

Throughout this article and the Data Poets project, there is an unresolved tension between content – what data is captured and what it reveals of the city – and form – how this data is captured and what it reveals of the difference between human and machine perceptions. Should one uncritically embrace the data poets as a ‘tool’ to map the city, at the risk of overlooking the biases brought with its particular position? Or should one focus on the non-human intelligence, at the risk of overlooking the unique qualities of the environment one is roaming? If there is a medium that can engage with both form and content, it is perhaps poetry, with its lyrical attention to (subversive or rule-based) form, and often subjective telling of experience. Poetry is not neutral. Its ‘aesthetic distance’, marked by the indirectness of language, invites the viewer to actively ‘work out’ the poem (Hanfling, 2003). This quality keeps the medium foregrounded and ensures a degree of critical separation. ⁴

The city text

In thinking the city as a text, as is suggested by the hermeneutic approach of the Data Poets, one must consider the modalities of interaction between urban ecologies and their interpretations by embodied agents (both human and AI agents). Are we reading (Hermeneutics), speaking (Rhetoric), or writing (Poetics) the city? Are we performing or playing the city as in a game or a theatre play? The concept of the city as a text is prevalent in Human Geography literature; however, the limitations of this textual approach have been questioned. B.B. Janz (2017) questions whether ‘textuality exhausts place’ and critiques its anthropocentric lens, particularly when the city is viewed reductively through a direct metaphor. This perspective contrasts with a broader definition of text as trace, as proposed by Derrida (1967). Janz’s critique suggests that viewing the city purely as a text may invoke notions of authorial intent and singular valid interpretations, which is the risk of an overly literal framing of the model. Despite these limits, the metaphor of the city as a text remains compelling. Viewing the city as a text reveals a dual nature: it is both read and written, interpretable and (at times) obscure, participatory and individual. While this approach may not be exhaustive, it serves as an evocative model for reflexive questioning and exploration.

The textual angle invites a reading of the city’s inherent traces, whether intentional or not, and the ways it shapes and is shaped by human practices. Janz (2017) notes that both places and texts involve reading, performing and creating new paths. There are both established conventions and transgressive interpretations. Understanding the city as a text invites us to consider the layers of meaning, the necessity of literacy in comprehending these spaces, and the affective engagements that result from our interactions with urban ecologies.

Anderson (2019) complicates this view of the text in his work on representations of the city; here the text can be seen as both the city object and its representation, its making present by act of writing. For Anderson, words generate a world and affect a relationship between text, reader, writer and world. Representations ‘do things – they are activities that enable, sustain, interrupt, consolidate, or otherwise (re)make forms or ways of life’. (p. 1120). The Data Poets project engages in a dual reading of the city text, examining both what the city represents and the processes by which it is represented. From a semiotic perspective, Roland Barthes (1986) proposes the city as a poem, a structure that ‘unfolds the signifier’. Barthes’ notion of the semiotics of space presents the city as a dynamic text, rewritten by its inhabitants through their interactions and experiences. In this semantic approach, we should understand the play of signs within the city, recognising that it is a structure that should not be filled in or completed. Barthes (1986) states,

For the city is a poem, . . . but it is not a classical poem, a poem tidily centered on a subject. It is a poem which unfolds the signifier and it is this unfolding that ultimately the semiology of the city should try to grasp and make sing. (p. 93)

This perspective suggests the hermeneutic task of interpreting signs within their context. The city, therefore, can be seen as an evolving semiotic system. The pesher genre – interpretive commentaries on scripture (exegesis) – deploys such a context-driven approach to understanding. This method presumes that prophetic texts contain ‘mysteries’ that must be unravelled/interpreted (pesher) by a divinely inspired interpreter to reveal their true meaning (Williamson, 2010: 354). Through a ‘combination of quotation-interpretation’ (p. 339), the text is recontextualised and given new meaning relevant within the interpreter’s timeframe and community; this is often achieved by explicitly linking a scriptural passage to a contemporary event or figure (p. 348). This duality between text and contextual interpretation could be applied to urban spaces: the surface level is experienced phenomenologically, by walking without knowledge or acquaintance with a place. The concealed ‘mystery’ is the personal and interpersonal (social) associations with the place. The place becomes a symbol for memories, events, people, groups and communities.

Barthes (1986) would describe these as signifiers (the material aspects) and signifieds (the meanings). The city involves constant negotiation of those meanings. Barthes further suggests that the city ‘is a discourse and this discourse is truly a language’ (p. 95), a language spoken by its residents through their daily interactions. The city ‘speaks to its inhabitants, we speak our city . . . simply by living in it, by wandering through it, by looking at it’ (p. 95).

This discourse is not metaphorical but a real, scientific (for Barthes) method of analysing the city’s qualitative elements. Henri Lefebvre (1991) expands on this by examining the production of space, arguing that urban environments are shaped by social activities over time. Lefebvre invites us to ask whether a city is best understood as a ‘work’ or a ‘product’ (p. 73). The city is ‘a space which is fashioned, shaped, and invested by social activities during a finite historical period’. Using Venice as an example, he treats it as a work ‘as unique and unified as a painting or a sculpture’. It is distinct, original, it occupies a space and is associated with a particular, contained time. Lefebvre then asks, ‘But what – and whom – does it express and signify?’.

Lefebvre’s interrogation surfaces the inherently expressive and significant aspects of urban spaces. Michel de Certeau’s analogy of walking through urban space as a rhetorical act builds on this notion of urban space as a dynamic and interpretative medium. He likens walking to a form of interpretation and enunciation, suggesting that just as speech acts bring language to life, walking brings the urban environment into being. ‘The act of walking is to the urban system what the speech act is to language’. Walking, he explains, is ‘a process of appropriation of the topographical system’, where the pedestrian appropriates and ‘acts out’ space much like a speaker engages with language. Through this movement, relationships are enacted between different positions: ‘interlocutors’ within the city (de Certeau, 1984: 97). Walking transforms urban spaces into a living text, continuously rewritten by the movements and interactions of its inhabitants. Barthes would likely respond that the signified is necessarily changing and therefore only a moving picture can be sketched, but for de Certeau, the individual signified is to be celebrated. The city understood hermeneutically is both a text to be read and a stage for social performance, a palimpsest of signs and meanings inscribed and re-inscribed by its inhabitants. The role of participants in the study, those using the Data Poets, is not fixed: Who is the reader, the device, or its user? Does the poetic creation change the narrative of the city or simply reveal it? A settled answer is less interesting perhaps than the question space raised. Thorough qualitative research would need to be conducted with participants to reveal how they negotiate the relationship between themselves, the group, the device and the urban ecology.

These analogies focus our attention on different aspects of the walks with the Data Poets. Which is most important to document and analyse? Is it (1) the performance of the live walk, (2) the writing and rhetoric of the traces and captures, or (3) the a posteriori interpretation and meta-hermeneutics of participants’ reactions to the generated poems? Each of these aspects offers a distinct perspective on understanding urban interaction mediated by the Data Poets.

1. Performance: Certeau’s view of walking as a speech act of performance leads us to focus on the live experience of the walk, including participants’ spontaneous comments, reactions and decisions. Documenting these through detailed note-taking and photography provides insights into the immediate phenomenological, lived experience of urban spaces.

2. Writing and Rhetoric: This focuses on analysing the data traces and captures, including sound and image data chosen by the participants, alongside their logic for capturing said data, and the AI-generated poems. Like Barthes’ concept of the city as a dynamic text, a place is rewritten by its inhabitants through the semiotic processes and interpretive strategies of participants.

3. Interpretation and Meta-hermeneutics: This concerns participants’ reflections on the generated poems and their overall experience. As in the hermeneutic Pesher approach, it involves a dual-layered understanding: both the surface-level formal experience and the symbolic associations.

While the Data Poets devices partially document the latter two aspects, a comprehensive study would require additional sociological methods such as photographic documentation and note-taking.

Conclusion

The Data Poets project was conceived as both a reflective critique and an experimental methodology to explore the potential of AI in mediating and representing human engagement with urban ecologies. This article has aimed to reflect on this project’s development, situate it within relevant theoretical frameworks and draw practical insights that may inform future research. A primary focus of this work is the dual role of AI as both an agent for generating socio-ecological knowledge and as a subject of inquiry, with its biases, limitations and perspectives brought to light through critical reflection. The project’s exploration of AI as an ‘other’ (Don Ihde’s ‘alterity’ relation to technology) opens the potential for using technology to generate ‘empathy’ and a critique of non-biological intelligences: researching not just with AI but also on AI.

Inevitably, the evolution in capability of LLMs described in part 1 (The Data Poets) also coincided with a transformation in public perception; while participants in 2021 expressed surprise at the concept of AI-generated poetry, by 2024 (and perhaps most importantly since the November 2022 release of ChatGPT) AI-generated text was now commonplace, and the project coexisted alongside examples of public pushback against outsourcing creative writing (Roeloffs, 2024), and studies documenting aversion towards AI-generated content (Qin et al., 2025; Zhou and Kawabata, 2023). The proliferation and ubiquity of Generative AI content have been categorised as AI ‘slop’ (Hern and Milmo, 2024; Moore, 2024). Anecdotally, public reactions to the Data Poets project can illustrate this ambivalence, as seen in this online review:

Although I am a frequent critic of AI projects, like all good hypocrites, I’m also fascinated by the subject. The poem above was generated by an experimental program [the Data Poets] based on the photo above of Bergun, Switzerland, and a brief prompt from me. (Butler, 2024, emphasis added)

To leverage this tension, upcoming deployments would benefit from foregrounding participant attitudes to the AI methodology. This project and article have demonstrated how Generative AI can provoke reflection on urban environments. Yet, while AI can encourage novel perspectives, it lacks the lived, embodied and contextually rich experiences that define human interactions with place. Through this tension between human and machine perceptions, AI emerges as a partial and situated ‘observer’ rather than an objective ‘tool’.

The ‘Data Poets’ project has aimed to augment our understanding of urban spaces through a poetic and dialogic lens, contrasting human sensorial experiences with AI-assisted interpretations of urban environments. But whose sensorial experiences are we revealing? Initially, the methodology was thought to be of interest for people who have the power to make decisions which affect urban ecologies (such as town planners and local authorities) and recognise the need to gather a wider and more experiential picture of an area by involving inhabitants. Conversely, it may also be used by communities themselves in an effort of counter-mapping. Who ultimately benefits from these reflections, and for what purpose are we collecting this data: professionals shaping urban policies or communities seeking to redefine their spaces?

The initial phase has revealed critical reflections on the methodology and execution. The pilot walks, conducted with few participants, local or familiar to the context, carried the risk of perpetuating a form of ‘helicopter research’ (Haelewaters et al., 2021). Despite the intention to ground the research in the lived, tacit experiences of the urban environment, the lack of direct involvement from local inhabitants raises questions about the authenticity and relevance of the poems. The subversion of tools alone does not seem to make for successful counter-mapping. By involving inhabitants who possess intimate, tacit knowledge of their surroundings, the project could better capture the nuanced, lived realities of urban spaces. This shows that the apparatus surrounding the research tool, including means of recruiting participants and contextualising the project, is key for aligning the Data Poets project with the lived experiences of inhabitants.

The Data Poets project, by capturing participants’ observations and transforming them into poetic outputs, also demonstrates the limitations of AI in fully comprehending the qualitative associations that define human experiences of place. The AI engages in a syntactic reading of the city-text, whereas the human participants perform a rhetorical one, enacting the space through their walking and inscribing it with meaning. In other terms, the poems often discuss what the picture shows, while the direct quotes from participants discuss what the pictures represent and symbolise. The participants have, in effect, created representations of these spaces and their lived experience of them. They have verbalised or written their qualitative assessment in various ways and taken a picture at a given point. The Data Poets project has given them a reflexive moment to process these representations. The act of discussing (or uploading) and sharing these reflections provides others with a glimpse into these personal narratives and may also begin to shape the perception and meaning of these spaces collectively.

In conclusion, the successful future deployment of the Data Poets project hinges on collaborative partnerships with people who have lived experience of the places visited. Future plans include conducting more walks with local participants and ensuring that the project’s outcomes are rooted in the lived realities of those who inhabit the places being interfaced with. This iterative, collaborative approach aims to create a more experiential, interpretative understanding of urban ecologies, challenging objectivist data narratives. Finally, by reflecting on the subjective, phenomenological aspects of city life, the Data Poets project can also critically explore how other forms of intelligence may come to inhabit and shape the interpretation of those environments.