If teachable technologies are to become deeply embedded in our technological ecosystems, economies, activities, and education systems, how can we utilize them in ways that lead to empathy or response-ability towards declining nonhuman others? And who will build these systems and be accountable for them? Machine learning (ML) invites poetic and storytelling potentials for authors, artists, and conservationists alike to imagine and make visible what is perhaps one of the most pressing problems of our time: the sixth mass extinction on planet Earth. Drawing together knowledges from multispecies ethnography and studies in algorithmic and multimedia creative writing, this article has two entangled purposes: firstly, to consider ways in which species extinction is currently being narrated through experimental algorithm-based approaches, and secondly, to consider how ecological mourning, as a reaction to recognizing extinction, can be embedded in such approaches. While the narratives and poetics of ecological mourning have been discussed by authors such as Thom van Dooren (2014a; 2014b; 2019; 2022), Deborah Bird Rose et al. (2017, eds.), and Stuart Cooke (2019; 2020), there is far less scholarship on what digital, algorithmic, or software-assisted creative practices may tell us about how extinction stories permeate into culture.

These purposes are expressed here through a reflective practice-led weaving of process and outcome fragments concerning the still-in-progress development of the Artificial Extinction Archives (AEA). Through the creation of the AEA, I follow multispecies ethnographer, and seminal author in extinction studies, Thom van Dooren’s call: ‘we have not yet found enough good ways of thinking through what extinction is and what it means… just how these extinction stories might, or should, be told requires continual rethinking…. How are “we” called into responsibility here and now, and how will we take up that call?’ (2014a: 147). Based on this call to creative action, it is important to consider ways in which critical creative making, alongside teachable machines and digital media, can help us to consider new modes through which extinction stories are presented.

The AEA is a work of digital art/creative writing in which a TensorFlow audio model has been taught to recognize the vocalizations of fifteen extinct or believed-extinct bird species through supervised machine learning and can display poetic text as a form of real-time translation. These are vocalizations that, outside of limited and aging audio data recordings or recreations, we will never hear again. This model is trained on archival data, as well as digitally filtered archival recordings that reduce background noise and enhance the bird’s calls, and in four cases, the creation of manipulated recordings from related birds in the absence of any known recordings in order to digitally recreate what research suggests those species may have sounded like. When the device believes it has heard one of these trained calls in the surrounding environment, translations appear on-screen in the form of combinatory poetry fragments drawn from a pre-written array. These fragments bare no scientific accuracy to the nature of the call itself, though each is intrinsically tied to a specific bird and its call.1

Why create a real-time audio-to-text translator trained to pinpoint sounds that the device will never hear in the wild? Rather than offering a prescriptive technological solution for how to express extinction, I’m instead looking towards ways in which creative practices involving teachable technologies may help us to connect with, and to potentially mourn, what is no longer there.2 Playing on the ability to ‘fool’ machines (Jospin et al., 2025), the resulting translations are fragmented phrases based on percentage of similarity to the original bird call, rather than requiring 100% similarity. This allows the AEA’s model to interpret sounds in any given sonic environment that appear somewhat similar to the calls of these extinct birds, and display this as a percentage match. There is an interplay between possibility and impossibility in the AEA in that it captures something that is gone and renders it synthetic precisely through the same technological and data limitations that mean it will never capture anything at all with 100% accuracy.

The AEA is a post-narrative that asks the implied reader/user/listener/viewer (hereafter: reader) to cognitively generate their own paratextual narrative around the fragmented visual and poetic on-screen information given by the artwork and the situated environmental sounds that the device is responding to. To take this position on computational fragmentation, this article relies on the history of post-classical inclusivity of both poetry and computational engagements within narrative inquiry, though it does not represent a purely narratological analysis. It follows a trend begun in the 1980s when poststructuralism destabilized narratology’s reliance on clear distinctions between story and discourse, and challenged assumptions about completeness and coherence (Pavel, 1990). This later developed into a series of post-classical narratology approaches that extended the field beyond prose fiction to include poetry, drama, visual art, and digital media such as hypertext, electronic literature, and digital poetry. One such approach is computational narratology, which still represents one of post-classical narratology’s most radical expansions, because it forced narratology to confront the narrative capacities of machines. Computational narratology inherited post-classical narratology’s interest in cognition and affect, which adds further vital history to the AEA project wherein the building of narrative is a partially cognitive activity dependent on how readers build story-worlds from poetic fragments or make sense of fragmented events.

Nonhuman death has been systematically devalued and rendered invisible in Western culture (Butler, 2004; Butler, 2009). As a result, mourning nonhuman death is often viewed culturally as irrational or sentimental (Wrenn et al., 2015). In the industries of animal death, such as factory farming and fishing, death itself is often hidden from public view or sanitized (Twine, 2010) in order to make it appear acceptable. René Descartes’ (1637/2006) assumption that other species were mere automata, incapable of self-awareness, set the foundation for these modern assumptions that other species are not worthy of our grief. Immanuel Kant deepened this assumption by arguing that other species lacked human rationality (1785/1997). Martin Heidegger further claimed that only humans had an authentic relation to, and understanding of, mortality (1927/1996). Each one of these assumptions classified the human species as separate from all others, and set a foundation that presumed no other species should be rationally mourned. Other species were not intelligent enough, self-reflective enough, nor emotionally or physically sensitive enough to be granted the same weight of life that we axiomatically grant ourselves.

This narrative is beginning to be overturned in the 21^st century by ongoing scientific research into other species’ minds, and indigenous and posthumanist frameworks that recognize interspecies bonds and entanglements as world-shaping (Haraway, 2008; Bird Rose, 2022). Many experience the death of a companion animal or animal friend as a deep loss based on care and recognition rather than ownership, the open admission of which is becoming culturally acceptable. Rituals for animal deaths such as pet memorials (Brandes, 2009), vigils by activists for slaughtered animals (Parme, 2024), and conservationists staging funerals for wildlife (Gibbens, 2018) show some of the ways in which humans are readily engaging in more-than-human mourning. These rituals strengthen the notion that in 21^st century Western culture, mourning is becoming more clearly defined by relational engagements as opposed to species boundaries (Redmalm, 2015).

Collectively, these viewpoints move towards an ethics that sees nonhuman life as deeply mournable. However, mourning extinction more broadly can be harder to grasp than the death of an individual animal, as it asks us to mourn for creatures with whom we may have never had relations, and whose extinction may feel comparatively slow and invisible. Extinction, unlike death, can unfold over several human lifetimes. Moreover, the sheer scale of extinction currently being seen, and the vastness of this loss, can create a sense of overwhelming paralysis towards action (Cunsolo and Ellis, 2018). This is where Thom van Dooren argues that to tell the stories, narratives, or poetics of extinction contributes to making a species real in the public imagination (2014a; 2014b; 2016; 2019; 2025). Without doing so, species fast being lost cannot come to life in the minds of human audiences, and thus, we do not fully recognize what we are losing, nor perceive extinctions as losses worthy of grieving. For van Dooren and his collaborators, the storytelling of extinction is an attempt to make space for the mourning of another species, and importantly, becomes an act of mourning in itself (2014a; 2014b).

Extinction constitutes a type of narrative, but one that is almost incomprehensibly large, fragmented, and entangled. Extinction appears, at first, to simply mark a definitive ending, as if it were a terminal point in the classic arc wherein the species’ biological story concludes. Indeed, as van Dooren et al. argue (2025), extinction is an entirely world-ending process. But embedded within that arc is a far more complex fragmented ecology of interdependent micro-narratives that unfold across generations in fragmented, rhizomatic, hypertextual, nonlinear forms. These are stories both of singular protagonists and of distributed, entangled agencies including subplots, detours, and entangled knots (Bastian, 2017) that resist containment within a tidy narrative structure. In this sense, the lifeline of an entire species unfolds as the proliferation of interwoven texts, situated in thick nonlinear ecologies of time and place. Extinction disrupts these tangled webs as a severance of countless fragmented threads at different temporal rates.

Environmental humanities and STS scholars such as van Dooren, Vinciane Despret (2016), Donna Haraway (2016), Michelle Bastian (2017), and the late Deborah Bird Rose (2022), all remind us of the power of storytelling, poetry, multimedia, and art to equip audiences with potential tools to approach vast and incomprehensible concepts such as extinction, climate change, death, and other forms of ending. Extinction stories take up this charge by inviting connection between readers and other species, which can lead to a deeper sense of empathy (van Dooren, 2014a). For example: van Dooren’s tales of Hawaiian crows (2014b) allow the reader to feel an immediacy in the importance of recognizing other species’ lifeworlds and cultures; David Owen’s (2004) non-fiction or Octavia Cade’s (2021) fiction stories about extinct Thylacines help us to recognize what we have lost despite that loss occurring almost a century prior; and by contrast, Bird Rose’s (2022) stories of bats as indigenous kin on the brink of extinction allow us to feel the importance of a relationship that we may still have the ability to preserve.

At the same time, a growing body of narrative theory reminds us that there are limits to what stories can do in the face of phenomena such as extinction. For instance, Hannes Bergthaller (2018) argues that large-scale ecological events like climate change are systemically complex processes that possess no intrinsic narrative properties. They are instead hyperobjects (Morton, 2013) that resists conventional storytelling methods. The extinction of countless individuals emerges from multi-scalar entanglements of land use, capitalism, colonialism, weather systems, and climate change, and perhaps no single storyline, however moving, could possibly accommodate the full complexity of such a distributed narrative. This apparent unnarratability does not mean that no stories can be told about vast ecological phenomena, but that there is a mismatch between the scale of such phenomena and the typically human-scaled narrative templates we traditionally draw upon to make narratives experiential (Bergthaller, 2018). Narrative’s anthropocentric habits inevitably reduce complex systems into more humanly manageable plots (Dürbeck & Hüpkes, 2021; Raipola, 2019). Juha Raipola argues similarly that those patterns that arise from ‘thousands, millions, or billions of small stories’ (2019: 269) should be classified as unnarratable because the relationship between micro-interactions and large-scale effects cannot be rendered as a single coherent storyline while doing justice to the phenomenon’s complexity. Thus, the subject of extinction is always-already a distinct challenge to narrative form that deserves alternative ways of navigating.

Van Dooren (2014a) argues that there are further methods for telling extinction stories that deserve to be explored. Both the apparent unnarratability of ecological crises and the post-classical traditions of narratology would suggest that forms such as fragmented, nonlinear, generative or algorithmic literature are worthy contenders for designing momentary synthetic interactions with extinct species. So, what should we do, then, to connect with the dead, and how do teachable and algorithmic technologies help us? Some inspiration here can be taken from the world of ghost hunting hobbyists: a popular past-time for those who believe that traces of deceased spirits can be captured and interpreted via technologies. Using ghost hunting devices such as so-called ‘spirit boxes’ and ‘ghost boxes’, as well as electromagnetic field readers, heat sensors, and voice recorders, gives readers a thrill of imagined connection with someone, or something, that no longer exists. By creating a narrative in which the deceased are still communicable, and using technology as a conduit to imagine connections between the past and the present, ghost hunting equipment can allow us to imagine the deceased as a close virtual presence. While there are clear problems with aligning extinction mourning with devices for ghost hunting—for a start, as anti-de-extinctionists argue, it is problematic to assume anything dead can still be reconnected with as it may cause apathy towards extinction (van Dooren et al., 2025)—what is striking about the practice of ghost hunting via technology is its philosophy of connection at all costs. A melancholy and desperate plea with both the universe and handheld technologies, to grant communication based on vague belief. Similar desperation is seen in the comments on Robert Davis’s YouTube upload of the Kauaʻi ʻōʻō audio: an unwillingness to accept the hopelessness of extinction, and a thirst for more stories, more media, and more archives that might help us to connect with those who are no longer here.

Technology has long played the role of pseudo-science-sidekick in communication and mysticism surrounding death. In the 19^th century, some were tricked into believing that cameras could record ghostly apparitions and ectoplasm (Cambridge University Library, N. D.). Beyond ghost hunting, contemporary efforts to communicate with the dead are increasingly turning to ML and algorithmic technologies (Higgins, 2023), where algorithms are becoming tools of digital resurrection. In particular, chatbots trained on the digital traces of deceased individuals (Kelly, 2024) offer ‘digital immortality’ and participate in the speculative storytelling of history as a live, interactive archive (Harder, 2024). The grief-stricken are turning to AI to fulfill a need to speak to lost loved ones, such as the chatbot projects Replika and Project December that utilize ML to simulate conversation with the deceased based on the stylization of an individual’s past written communications. At present, these chatbots function less as accurate resurrections and more as tools for navigating the meaning-making of personal data (Park, Kim and Lim, 2025). However, these technologies show a strong and growing cultural investment in computational mourning, wherein grieving users are willing to accept a stylized algorithmic approximation of the mourned in the name of posthumous connection.

Barthes saw visual media as already haunted: to look at a photograph was to confront an image that in some way certified that its contents existed and is now irretrievably past (Barthes, 1981). Recontextualizing ghostly presences through ML, and the broader uses of digital software, AI, and synthetic media, has caused a cultural rethink in media’s indexical relation to its original referent. While Barthes showed that the photograph testifies to the past reality of its subject as the that-has-been, synthetic media loosens this bond because generative systems produce the ultimate in remixed media with no direct causal tie to any one specific trace in the world. This has been described as an indexical crisis for media (Velasquez, 2024), where there is no longer a that-has-been. While the AEA is not a generative AI model and thus does not produce synthetic media, it can be seen as operating in the threshold between indexical trace and synthetic media output. The archival bird recordings that train the model retain Barthes’ indexical charge. They are sonic that-has-beens via audio files in which vocalizations once physically vibrated a microphone diaphragm. But when the AEA listens to its surroundings, it finds patterns learned from these archival traces within contemporary soundscapes and, from this, produces as-if textual and visual translations. Perhaps the AEA, and similar technologies, transform Barthes’ that-has-been into something closer to a that-might-have-been by suggesting traces within contemporary noise.

Fragmented, that-might-have-been texts can be understood as a creative response to the unnarratability of extinction. There is expressive power in multimedia artworks such as AEA precisely because they manage to engage with extinction without forcing it to the mould of a singular story. The work is dispersed into poetic fragmented encounters for the reader to interpret, where each fragment of text functions as a tiny narrative shard. These shards can never add up to some master plot of extinction, instead, they accumulate as a nonlinear constellation of almost-stories that emerges from the reader’s sonic environment. This accumulation of fragments arguably places the AEA closer to the kind of experimental form better suited to the Anthropocene: a structure that accepts fragmentation, incompleteness, combinatory generativity, and nonlinearity as formally appropriate to the unnarratability of large-scale ecological crises.

This fragmentation of narrative gives a strong indication of how the affordances of algorithmic and computational technologies may help us to find new ways of storying extinction. Fragmentation in contemporary experimental writing is already understood to carve narratives into discrete, manipulable units/nodes/events that can be recombined algorithmically, whose overall coherence can emerge through dynamic assembly by the reader, the machine, or both. Beyond presenting nonlinearly, the fragmented text distributes meaning across gaps that the reader must actively bridge, where writers ‘hand over a significant part of the meaning-making to the reader’ (Krauth, 2019). The AEA’s poetics of fragmentation arise from both the artist/writer’s decision making and the affordances of the conditions under which the work is experienced. Meaning is created in the gaps between the two, and narrative is assembled from nodes that can be recombined based on external forces. In contrast to the linearity of classical narratology, the AEA’s algorithmic listening produces a cognitive paratextual narrative that is shaped by temporal environmental contingencies around the reader, as well as machine interference, without authorial sequencing.

Mourning is an active process. One of van Dooren’s central focuses is the way in which his own research and writing are forms of active ecological mourning, including reading surrounding literature as mourning, writing as mourning, and fieldwork as mourning. Van Dooren, along with multispecies ethnographers such as Deborah Bird Rose and Michelle Bastian, and ecopoets such as Stuart Cooke and Travis Mason (2013), all arguably engage in acts of what could be called creation-as-eco-mourning. Through the process of creating the AEA I have come to a similar realization, but wish to extend upon it that digital creative practices of making, coding, writing, and multimodal mixing can also serve as acts of ecological mourning.

Creative practices have long played a vital role in active mourning, wherein the bereaved may take up a hobby-craft or decorate the graves of loved-ones (Brennan, 2015; Dreifuss-Kattan, 2016). Creative, active mourning is broadly seen as a healthy coping mechanism (Nimroody, 2014) with the potential for feelings of transformation (Melgar, 2007). Mourning is not necessarily a space in which nothing gets done, instead it can be an invitation to turn to flow-based actions that are self-soothing. Further, creative practices are ways of processing emotions that may be difficult to convey in other forms of communication, for example if we struggle to express how we feel through speech, we may be better able to do so through drawing. Grief often resists direct articulation, and creative expression offers a parallel syntax by allowing what is difficult to say to be materialized otherwise.

For me, ecological grief does not feel like an overwhelming personal loss, but it is certainly an ever-present and anxiety-inducing drumbeat; a slow-moving splinter that cannot be removed, a Heideggerian dwelling in a place I have never experienced directly. I often give in to a nihilism of feeling that I cannot positively contribute. I feel the exasperation of realizing so many of my actions have negative environmental consequences that affect the lives of other species. Every day there are more stories of terror from my local community: Today the bats were driven out of their roost again; the heat kills them; a mob of kangaroos were shot rather than relocated; a group of koalas were shot rather than helped; an unknown paralysis disease kills the birds and the bats. I feel myself constantly wanting to be smaller: To drive less, travel less, eat less, buy less, use less, waste less, need less, want less.

I have found the creative process of making the AEA to have been, at least in some very small part, a form of self-soothing of these anxieties. Just as repetitive bodily motions can regulate the nervous system, so too have the repetitive motions of digital creative processes offered me a sensory salve. Such self-soothing clearly does not resolve ecological problems for our collective future. However, it is an attempt to turn feelings of exasperation into an energy source. In my case, making is thus a deeply embodied form of ecological mourning in which the making process becomes a reflection on the haunted traces of referents that are no longer alive. For example, training the audio model required hours of listening to archival recordings, at times adding digital distortion and removing background noise, cutting and looping to create data classifications, and hearing patterns that could inspire my fictional translations. Each clip I worked with carried weight through its traces of a past life. Teaching these clips to a machine became a way of knowing that this data could be carried forward, recognizable by a nonhuman entity who simultaneously acts as an archive that preserves knowledge of these bird calls. It felt like I was like teaching an extinct language. It gave me some comfort to recognize that I was building a machine whose goal was to listen for what is lost, and to make space for that listening in the lives of readers.

My practice is complicated by the fact that I also consider it research. In other words, the very act of writing this article complicates practice-as-mourning as it positions it as a methodology, wherein methods must be repeatable or at least transferable. This is important to consider for how other creative practitioners and creative practice researchers can engage in creative ecological mourning in the future. To consider ecological mourning as a methodology for creative practice, we must recognize it as a non-systematic approach to knowing that is transferable but beyond containment, in so much as mourning is done and shared for reasons that are usually outside of research. While despair has often been associated with a concept of artistic genius in Western culture, it is in many ways fundamentally incompatible with organized research, the scientific method, and the researcher’s supposed objective gaze. Not only does mourning-as-practice come from a place of subjective inquiry, it cannot be forced into the times and spaces academia may usually reserve for active fieldwork or other methods. It is not a methodology that can be placed upon an individual at a time that is convenient, instead mourning while making comes when it does; it becomes at certain moments, spurred on by events usually outside of the maker’s control. So, it is not research design and planning, but rather misfortune that aligns ecological grief with creative practice—an accidental methodology born from simply existing at the same time as occurrences of extinction, climate change, and habitat destruction, and noticing the ways in which the tendrils of these occurrences sow seeds into making processes.

This research has put forth combinatory fragmentation and unnarratability as central narratological conditions for understanding how algorithmic technologies can express extinction. While unnarratability suggests the difficulty of representing distributed, multi-scalar ecological processes through narrative expression, digital nonlinear fragmentation as a literary form, with its dispersed and gap-filled structure, connects metaphorically to the discontinuity and ecological ruptures of extinction. Together, these challenges to traditional narrative forms suggest that extinction stories cannot always depend on completeness, or perhaps even coherence. In emphasizing combinatory fragmentation and unnarratability, algorithmic works such as the AEA point toward narrative modes perhaps better fitted to the realities of ecological loss.

In devising an almost-impossible act of listening for voices that no longer exist, the AEA has aimed to reframe the extinction of species as a loss that continues to be readable in the everyday. Algorithmic technologies increasingly mediate our relationship to finality and our ability to keep connections alive, where technologies such as ghost hunting devices, AI replicants, and the AEA, offer imagined proximities that reveal our desire for contact. Further, the act of building such a system doubles as a ritual of ecological mourning: an attempt to recognize lost life through creative processes and use technologies as speculative ways of being-with the dead. Through this device, I have attempted to take up van Dooren’s call of rethinking how extinction stories are told by implementing the affordances of algorithmic digital media that highlight fragmentation and unnarratability. If, as van Dooren suggests, the telling of extinction stories requires continual rethinking, then creating algorithmic devices is a way of experimenting with how we might continue to listen, even when nothing is speaking, and promote a willingness to remain with the trouble of extinction and feel its heaviness in our immediate surroundings.