Tweeting the cosmos: On the bot poetry of The Ephemerides

Contexts: Interpreting Twitter bots

The technical beginnings of digital text generation can be found in a 1952 program developed by the mathematician and engineer Christopher Strachey (see Link, 2016). This program printed a series of comic love letters, operating on a template message by replacing its verbs, nouns and adjectives with arbitrary selections from a stored vocabulary list. Although the resulting letters were maudlin, repetitive and otherwise absurd, they had the effect of making distinctly visible the procedural, permutational operations that drove their production, as facilitated by the still-novel paradigm of digital computing (see Strachey, 1954, for examples and source code).

Strachey’s efforts were pioneering in the digital realm, but the procedural recombining of written language as a compositional technique, whether at the level of creation or reception, has a much longer history. Simanowski (2011) describes a genealogical trajectory reaching back into early Baroque experiments with Ars Combinatorica, such as in Quirinus Kuhlmann’s short poem XLI. Libes-Kuß (1671) featuring individually selectable line endings, before emerging prominently in Dadaist efforts at cut up and found poetry in the 1920s, and then, much later, in Raymond Queneau’s Oulipian effort Cent Mille Milliards de Poèmes (1961), with its 14 groups of 10 lines that can be selected and recombined at will. Specific early attempts at harnessing the speed and flexibility of digital computing to generate poetic forms include the ‘stochastic’ poems of Theo Lutz (1959–1960), which recombined grammatical excerpts from Kafka’s The Castle (1926). Following soon after was Nanni Balestrini’s Tape Mark I (1962), which used a random number algorithm to sequence stored extracts from Michito Hachiya’s Hiroshima Diary (1955), Paul Goldwin’s The Mystery of the Elevator, and the Tao Te Ching by Lao Tzu – the initial selections being filtered by a series of grammatical constraints in order to output an intelligible poem.

These texts represent only some of the more well-cited instances of many such experiments in digitally generated prose and poetry over the past 60 years (see Funkhouser, 2007, for a detailed overview). If a contemporary successor to these historical efforts could be cited, then it might be noted how, amidst the fluctuating constellations of user accounts that constitute the social media platform Twitter, there exists a sizeable population of automated entities, ‘bots’ – between 9% and 15% according to Varol et al. (2017). These bots generate a stream of tweets using fixed algorithmic rulesets and stored databases of material, either according to a periodic schedule or in response to the actions of other users. In recent years, a significant degree of popular infamy has accrued around bots for their role in influencing online political discourse (see Woolley and Howard, 2017). Nevertheless, outside of the extensive media commentary that is associating Twitter bots almost exclusively with varied strains of automated political activism, there exists a subset whose outputs can be characterised as expressly aesthetic in aspiration and orientation. There are myriad such bots available, encompassing a wide range of idiosyncratic topics and thematic concerns, and deploying a variety of expressive techniques by which to represent them – such as the production of generative images and writing, or the remixing of existing works of art and literature, or creatively re-presenting information drawn from extensive online databases, including Twitter itself (see Giles, 2016, for a typological overview).

To concentrate specifically on bots involved in the generation of text, a relatively typical and accessible instance is poem.exe (@poem_exe) by Liam Cooke, which is introduced simply as an exercise in ‘micropoetry’. Characteristic outputs from this bot over the years include:

night traffic against a pink sky in darkness lingering. 4:00 p.m, 1 Dec. 2015. climb mount fuji under the snow even among the stars 6:29 am, 31 Dec. 2016. clouds will separate us the bitter parts clouds will separate us in the autumn wind 9:20 am, 17 Oct. 2017.

In his comments within the bot’s source code, Cooke (n.d.) describes how each ‘poem is constructed by selecting the first line of a random haiku, the second line of another and the third line of yet another’. The source does not reveal the specific haikus (all of which are already translated into English) from which these poems are assembled, but indicates that every poem generated through this process is filtered to ensure its phrasing is seasonally appropriate – that is, references to autumn winds do not collide with those of summer sunshine and so on.

Other instances of Twitter bots deploying procedural techniques for the generation of text are many and varied, ranging from the relatively conventional to the highly abstract. Nevertheless, the key question confronting these pieces is the same as that detailed in Simanowski’s (2011) examination of generative writing more broadly: In what ways can linguistic structures that have emerged through chance (pseudorandom algorithms) be read and interpreted? This question becomes especially pertinent in cases where the level of aesthetic provocation and interest, at least on the part of the reader, is deemed significant (potentially in line with the conventions of established literature and poetry), or, conversely, the level of abstraction is sufficient as to entirely disrupt normative reading strategies, with the concomitant risk of suggesting merely ‘nonsense’ (whether for comic effect, or because of insufficient programming).

Simanowski himself debates some extant possibilities of interpretation, moving beyond a simple rejection of these texts as meaningless, without a sense of authorial intention (at the level of composition) in which to enter dialogue, to acknowledging, but ultimately discarding, the potential for chance alone to become meaningful, especially as it provokes the reader. In place of these approaches, Simanowski favours an exploration of how meaning emerges in the relationship between the generative process itself and the original themes of the literary sources utilised. In other words, this approach considers how the generative process either animates or crystallises the themes and concepts suggested by its source material, which is then reread in the light of the transformations made.

A conception of generative writing as staging a creative dialogue with its source represents an engaging critical approach – one that might be applied readily to instances such as poem.exe and its haiku foundations. However, Simowski’s rejection of chance as meaningful, though made in a literary critical context, appears to elide the possibility of considering the machinic aspects of a generative poem as being expressive in their own right. Simowski’s argument here is that the search for meaning in chance invalidates any sense of a ‘message’ being conveyed by its varied outputs, for if chance is equated with the uncertain nature of being, then, in the case of generative poetry, ‘[life] has now been centered in the algorithms of a machine’ (2011: 100). He continues:

The line of text that “happens” is both more and less representational […] It is more representational because it is not abstract; it is less representational because it does not represent a human intention. One may say that the seemingly representational line in fact only (re)presents its own occurrence. It is pure code. (2011: 100)

While an unalloyed focus on generative contingency may well be incommensurable with the search for meaning at the level of textual composition – and this article will not be debating such assertions further – it does draw attention to the status of Twitter bots as specific forms of software. Like all such artefacts, bots emerge out of complex, interlocking genealogies of engineering logics, cultural imperatives and social contexts, which define subsequently the possibility spaces of function and expression they afford (see Fuller, 2008). Operating within a social networking environment, Twitter bots exist in association with technical and cultural factors that reach far beyond their immediate thematic bounds, and so give cause for considering how these aspects might characterise the way a given bot can be viewed as a whole – or at least be accounted for when assessing its expressive operations, rather than treating this software infrastructure as a relatively neutral vehicle for the dissemination of ‘chance’ textual outputs, which are then read in isolation.

A technically informed approach towards the analysis of digital textuality is not in itself a novel development and has been conducted at length previously in the work of numerous scholars, following Aarseth’s (1997) injunction that the close reading of ‘born-digital’ textual forms must acknowledge the agency inherent within the machinery itself, in its capacity to resist, reshape and recast the reading encounter.

Instances of this approach include the paradigm of ‘expressive processing’ outlined by Wardrip-Fruin (2009), which describes the ‘operational logics’ of software as encoding specific, culturally contingent vectors of human thought and expression. These embed subsequently particular assumptions about how certain aspects of the world can be modelled in a computational environment, and it is in tracing their effects that a rich context of reception is provided for reading the acts and outcomes of textual generation. Hayles (2008) develops and expands this perspective by outlining a model of ‘intermediation’, in which the computer and the human user interpret, adapt and reconfigure one another within the space of their mutual entanglement. Here, different material structures and processes mediate the feedback cues generated at each level of the exchange, whether technical, biological or cognitive in origin. It is within the resulting matrix of intermediating actions between human and machine that, for Hayles, is where meaning is crystallised in the digital literary encounter – which, in helping to make explicit this fact, encourages a detailed consideration of these interlocking aspects on the part of reader and critic alike.

It is along similar lines that Kirschenbaum (2008) characterises digital texts as embodying two distinct modes of materiality: formal and forensic. The latter encompasses the specific structures constituting digital artefacts as physical phenomena, ranging from the ‘micron-sized residue of digital inscription’ on hard disks, to the globalised networks of modern computer manufacturing, up to the politics of e-waste management in third-world nations (Kirschenbaum, 2008: 10). Following on from the specificity of forensic materiality is that of formal materiality, which Kirschenbaum defines as the ‘imposition of multiple relational computational states on a data set or digital object’ (2008: 12). That is, how digitally encoded objects can take on different forms depending on the software protocols and processes applied to their actualisation. It is in exploring the relations between forensic and formal materiality that Kirschenbaum evaluates the material and logical specificities of the matrix from which all digital artefacts emerge, are sustained and are actualised in practice, while acknowledging also the range of different possibilities this matrix affords in each instance of digital inscription and expression.

Resonant with the work of the above-mentioned writers are the approaches documented under the broad field labels of ‘software studies’ and ‘critical code studies’, as outlined by Fuller (2008) and Marino (2014). These are concerned similarly with how software articulates different logics, epistemes and attitudes. Due to its infrastructural status in the contemporary environment, it is recognised that software plays a key role in enabling and constraining different possibilities of thought and action across culture, society, politics and artistic endeavour. The analyses presented in the context of software studies often work at the level of the code itself – its algorithmic structures and how they function, both individually and in sequence, and what their implications might be. Although not concerned exclusively with the reading of digital texts, such approaches do nevertheless treat the code of all software artefacts as worthy of close analysis, and so offer suggestive paths for reading the code in conjunction with the textual forms it makes available.

Given the range of perspectives encompassed in the above, it should be noted that none engage explicitly in the close-reading of Twitter bots. Nevertheless, at the time of writing, these artefacts still represent a developing topic of academic criticism, with extent commentaries offering diverse reflections on what they represent and how they might be engaged. Much of the open-access material online, for example, is confined to discussing the software needed to realise creative bots in the first instance (see Compton, no date; Graham, 2017). More scholarly treatments can be found, however, with Flores (2014) conducting close readings of specific bots, Giles (2016) assembling an overall genre typology, Sample (2014) assessing a distinctive class of protest bots, or ‘bots of conviction’, and Pipkin (2015 and 2016) treating bot making practice as a contemporary form of folk art. One extended analysis of Twitter bots is provided by Lampi (2017) who investigates the functioning of varied bots in the absence of any available source code, identifying the recurring patterns in their output. This approach offers a distinctive window into the creative influence of a bot’s author, in terms of their writing of the vocabulary lists, structural templates and grammatical rulesets that are fundamental to the character of a bot’s output. A more recent investigation is by Veale and Cook (2018), which details the processes involved in crafting Twitter bots, characterising them as miniature laboratories for experimenting with the nature of meaning in an age of algorithmic expression.

Records of the developing academic discussion around Twitter bots can be traced also in online databases such as ‘Electronic Literature as a Model of Creativity and Innovation in Practice’ (ELMCIP), which lists abstracts from varied ‘Electronic Literature Organisation’ conferences, as well as short entries detailing the functioning of specific bots. Listed among the current ELMCIP records is an article by Rodley (2015), which evaluates bots and other generative online writing as enacting a post-literary mode of ‘Boolean poetics’, centred on the manipulation of data structures.

First-hand accounts by bot authors concerning their work are also available. The technical reflections offered by Cooke (n.d.) on poem.exe are one instance, as are the meditations of Kazemi (2016) and Compton (2016), who ascribe the aesthetic and political value of bots to the often bizarre and unexpected contrasts they establish against the flow of curated, corporate tweets that border them. Whitson (2016) considers his BillBlakeBot (@autoblake) as a vehicle to reflect on the digitisation and reconstitution of the works of William Blake more widely. On this point, he describes BillBlakeBot as exposing the latter’s ‘algorithmic condition’, in which ‘Blake’s images, words, and ideas are broken up, remixed, channeled [sic], transported, shared, and transformed’ in the contemporary digital environment, reshaping fundamentally how they are received and understood. This reading resonates with Simanowski’s own approach, in focusing on the generative reworking of literary sources, and the critical insights these might afford.

All of these authors provide different ways of assessing the relationship between the generative operations of Twitter bots and the specific outputs they yield. As this article emerged primarily in response to Simanowski’s critiques, and his arguments against the machinic as a source of expressive agency, the remainder of this discussion will focus on the viability of reading Twitter bots along this aspect. The emphasis here will be on situating the thematic framing of a specific bot, The Ephemerides by Alison Parrish, within the technocultural contexts that are intrinsic to its functioning and networked milieu more broadly. In the reading that follows, the software and its enabling infrastructure are treated not simply as neutral conveyers of contingency, or as cultural formations offering a provocative context of reception, but as articulating a distinctly machinic mode of expression. It is this which is presented as a key attribute of this particular bot, but one that raises useful points of enquiry for the future reading of others.

The question of where the specific outputs of The Ephemerides, or of other bots more generally, fit into this critical approach has a twofold answer. Practically, analysis of these outputs is rendered ancillary to assessing the overarching technical and thematic concepts that motivate and characterise a bot as an expressive artefact. This again stems from their forming a key basis on which bots can be shown as emerging from, and engaging with, greater technical and cultural contexts, and it is this which constitutes the critical locus of this discussion. However, this stance also reflects their subsequent value in providing a comparatively stable context in their own right by which to consistently frame and interpret otherwise diverse outputs.

Many bots can generate texts that traverse a broad spectrum of linguistic coherency or abstraction, and each iteration is often singular in scope, bearing no immediate relation to that which came before or after. In her assessment of early generative writing, Van Dyke (1993: 180) finds the pleasure of these texts in their ability to challenge the reader’s capacities for reencoding – to ‘establish different boundaries among the obvious, the unusual, the hallucinatory, and the hopelessly unintelligible’, and so realise their own ‘idiosyncratic triumphs over incoherence’, enjoying ‘the pull of multiple codes even on stubbornly anomalous passages’, as well as ‘the text’s occasional gifts of unquestionable coherence’. Nevertheless, when assessed in terms of what a bot is presented as trying to achieve, or the concepts it is exploring, divergent instances can be brought together and read as instantiating the shared technical or thematic basis that motivated their production – even when they fail to cohere or appear meaningful in isolation. Therefore, to concentrate on these overarching aspects is not inappropriate for evaluating the expressive capacities of Twitter bots. In this regard, a key goal of the following reading will be to reveal the significance of generative bots as expressive, machinic agents more broadly, and thus offer another demonstration of how this art form, however tightly delimited, can afford a rich catalyst for critical reflection.

The Ephemerides – The poetry of space probes

Allison Parrish is a prolific author of Twitter bots founded on the procedural generation and remixing of text. Of the many she has coded that would be valuable for this discussion, it is perhaps The Ephemerides (@the_ephemerides) that is most apposite. As befitting its title, The Ephemerides is a bot that selects imagery of the outer planets from the NASA Outer Planets Unified Search (OPUS) catalogue – which collates masses of digital data from a variety of unmanned space probes – and mixes it with algorithmically selected extracts from two source texts: Astrology: How to Make and Read your own Horoscope (1920) by Sepharial and The Ocean and its Wonders (1874) by RM Ballantyne.

Having processed these source texts in order to separate them out into their constituent sentences, grouped according to their standalone clauses and syntax, The Ephemerides selects and recombines different clauses at random, producing juxtapositions that are then affixed to an arbitrary image selection from OPUS (see Parrish, 2015a). The resulting outputs are then posted onto Twitter before being mirrored onto a separate Tumblr account. Some typical examples are presented in Figures 1 to 4.

OPEN IN VIEWER

Figure 1.

The accompanying tweet reads: ‘This long intense/darkness was most/depressing. Some/cases/showed all the colours/of the attrition/as well. 5:18 am, 15 Oct 2017’.

OPEN IN VIEWER

Figure 2.

The accompanying tweet reads: ‘Its thin edges/represented/in a long day/are only wavelets/of the good aspects/of the scarred slopes. 1:18 pm, 12 Sep 2017’.

OPEN IN VIEWER

Figure 3.

The accompanying tweet reads: ‘The sea may be the/sun itself without/the perfect astronomical/paradigm of stirring/billows. 9:18 am, 20 Aug 2017’.

OPEN IN VIEWER

Figure 4.

The accompanying tweet reads: ‘Whales were afterwards/learning, but registering/of a knowledge above/the truth. 1:18 pm, 16 Oct 2017’.

If these outputs were to be viewed in isolation, following Van Dyke’s (1993) observations, then there are many potential lines of interpretation, based on the varying thresholds of connection and association brought forward by a given reader. The above citations were made in the light of their linguistic coherency and an apparent (if entirely arbitrary) resonance with their accompanying photographs. For example, in Figure 1, there is a darkness shared between the textual and the visual imagery, a greyscale ‘attrition’ of the otherwise colourful planet Saturn, as framed against the stark blackness of deep space. In Figure 2, a scientific account of Saturn’s rings might be discerned, their fineness and reflective properties being contrasted against their constitution out of rough boulders of ice. In Figure 3, a loose association could be noted between the varying waves and lines in Saturn’s rings, and those of ocean waves and currents on Earth – as illuminated and influenced, in both cases, by the sun. In Figure 4, the text and image are the most enigmatic of all, but perhaps suggest an allegory concerning all the profoundly non-human ways of being (whether found at the bottom of the ocean or the furthest reaches of the cosmos) that exceed the epistemic frameworks of modern technoscience – knowable, but residing nevertheless beyond human experience.

In order to read these idiosyncratic meditations as expressing an overarching theme, it can be noted that Parrish (2015a) herself characterises The Ephemerides as an imaginative take on the poetics of deep space probes – of how these machines, if able to reflect on their regimented encounters with austere, alien bodies, would do so through poetry:

[S]pace probe poetry would be minimal, contemplative, turning suddenly from the technical to the lyrical and back. The poetry of Basho, Gary Snyder and Rae Armantrout came to mind as potential points of reference. I proceded [sic] with these poets in mind and tried to produce something stylistically similar.

Parrish’s speculative depiction of space probes as distinctive agents, rendering their experiences in forms other than that of numerical data, gives cause for reflection concerning the nature of the observable as mediated through such devices. It might be considered how these machines, like any form of scientific apparatus, function not as neutral, facilitating links between subject and object, but as particular actors that are entangled with, and thus leave their mark upon, the phenomena they perceive. It is in following this line of thought that an opportunity emerges to develop a reading of the thematic and generative operations of The Ephemerides, treating its textual outputs, and the images associated, as bearing out its operations as a software artefact – one that is attuned to actualising, and meditating upon, non-human, machinic modes of expression more broadly.

Data space – Contingency, agency, interpretation

Reiterated in the abstract, following on from Parrish’s own description (2015b), the functional premise of The Ephemerides is a bot documenting the space of expressive possibility across a pair of digital databases: one that is textual, curated by a human author working with varied online sources, and another constituted by entries from multiple robotic spacecraft, whose explorations of the outer solar system actualise its evident potentials of discovery. What connects the divergent technical, environmental, and interpretative contexts represented by Twitter bots and robotic spacecraft are the various signal processing operations – of reception, encoding, storage, and transmission – that enable the radio networks that relay sensory inscriptions made by distant celestial objects back to the Earth and into an online database, prior to their rebroadcasting on a social media platform. In short, the functional premise of The Ephemerides makes explicit the globally and celestially dispersed network architectures that draw the material remoteness of outer space into the empirical locality of the Internet, a ‘data space’ that is navigated, read, and expanded on by human and machine alike.

To begin evaluating the aesthetic significance of these infrastructures on which The Ephemerides relies, it is worth turning initially to its most evident manifestation in the form of the raw images that accompany every generated poem. As noted earlier, these images are derived from the online OPUS catalogue, which provides access to a comprehensive body of instrumental data captured by various NASA missions to the outer planets, namely Voyager 1 & 2, Galileo, Cassini, and New Horizons. These raw images, as present on the database, are strikingly different from the polished, full colour depictions offered in more public-facing outlets. Here, the images are invariably black and white, sometimes poorly aligned with the target, and are usually replete with visual artefacts, including blurring, distortion, noise, and black banding. Many are simply calibration and orientation shots, depicting little more than background noise, distant stars or fuzzy horizons. Parrish (2015a) herself reads these images as ‘warmer and more personal […] which fits into the “fiction” of the bot: that these poems are a personal documentation of the probe’s own experience’. An alternative view, however, would be see these images such as taking on a much more remote and fragmentary aesthetic, presenting partial, tenuous glimpses of forbiddingly remote objects, as compared to the lustrous vistas more popularly associated with planetary photography.

In the latter case, it is important to note how this final imagery is constituted through mosaics of raw images edited into a seamless perspective, combining the otherwise separated red, blue, and green filters to depict, with the aid of some careful tuning, ‘natural colour’ views of what the unaided human eye would ostensibly perceive – even if this would rarely be the case in actuality, given the poor lighting conditions of deep space. Such editing erases the status of these final images as deriving from highly engineered processes, in which spacecraft alignments, exposure times, and intricate automatic sequences combine to deliver the requisite raw shots, whose varying quality is evidence of how challenging such conjunctions are to achieve. Moreover, the editing process removes those visual artefacts that are indicative not just of the imaging procedures themselves, but also the crucial networks through which the data gathered is transmitted back to the Earth across the electromagnetic cacophony of the celestial medium, which induces a significant degree of error and equivocation in the downloaded signal.

It is this noise inscribed into each raw image, the origins of which are multiple, that accentuates their status as material, mediated artefacts. Far from being direct, Cartesian windows into the cosmos, traversing lightly over the material barriers of space and time, they are shown instead to be the product of concrete infrastructures and contingent processes. As a bot program, The Ephemerides draws on its capacity to interface readily with the Earthly nodes of these celestial networks, and so facilitate a creative and conceptual exchange that goes beyond simply juxtaposing astronomical images with generated poetry. That is, in its use of raw spacecraft imagery, The Ephemerides draws attention to how the instruments and networks through which outer space is documented leave their mark on the data gathered, before staging a dialogue into how such contingency is emulated within, and is of significance for, the inscriptive operations of generative expression.

To help consider this latter aspect, it is worth following Goriunova and Shulgin (2008) who explore the cultural reception and aesthetic potential of computational glitches and noise. The authors contend principally that these emergent eruptions of unexpected and undesired behaviour constitute ‘a manifestation of genuine software aesthetics’, revealing how the normative enactment and reception of human–computer assemblages are never straightforward expressions of technoscientific functionalism, but are carried through the entangled exchanges between human and machinic actors (2008: 111). Both authors highlight especially the contrasting imagery of a functional, knowable and ultimately controllable technological domain – one that, not coincidentally, was integral to the military imperatives that gave rise to modern space exploration – as opposed to one that is interrupted by dysfunctional, alien, and resistive agency:

Functionality, as a characteristic of established machine aesthetics is always chased by dysfunctionality (if not preceded by it). […] Glitches are produced by error and are usually not intended by humans. As a not-entirely human-produced reality, its elements are not one-hundred percent compatible with customary human logic, visual, sound, or behavioral conventions of organizing and acting in space. […] A glitch is the loss of control. (Goriunova and Shulgin, 2008: 114–115)

Goriunova and Shulgin’s contentions here can be redrawn in the light of Pickering’s (1995) essential point that the processes of engineering and scientific discovery can be characterised as a ‘dance of agency’: a performance of mutual resistance and accommodation between human and non-human (material, technical) actors. This dance represents a process of interactive stabilisation, in which the agential capacities of intersecting actors impinge upon one another, foreclosing and facilitating certain vectors of possibility, until they become tuned reciprocally so as to perform consistently certain actions – such as the development and operation of a new piece of scientific apparatus, which binds together different physical processes in such a way as to be instructive as to their properties and potential. Contingency here is critical, for as apparatuses break down, and hypotheses prove incorrect, the resulting disruption promotes fresh interpretative manoeuvres and practical interventions on the part of the human interlocuter, before establishing, eventually, a repeatable, stabilised performance that can be formalised as empirical scientific knowledge.

Pickering takes this image as the starting point for a much broader characterisation of an agential universe, one that is enacted continually through differential modes of material agency that exceed frequently the thresholds of human observation and intention. This performative, processual domain is changing and evolving continually, and so at no point can human endeavour be characterised as having subordinated entirely the agency of the non-human. Instead, only a small portion of the world’s material becomings can be arrested within a very particular set of interactively stabilised contexts. Hence, while a technological artefact may function correctly across a wide range of conditions, the agential dynamism and unpredictability of its surrounding milieu ensures there is always the potential for it to be confronted by liminal scenarios, disrupting and dissolving the entanglements that underpin its normative operations.

It is in this context that a computational glitch demarcates a visible rift in the fragile sociotechnical ecology in which all digital devices operate, and showing it to be an illusion sustained whenever the smooth, reciprocal predisciplining of human and machinic agency holds sway. Echoing the emergent behaviours expressed by generative programs such Conway’s Game of Life (1970), while the concrete operations of digital systems may be comparatively stable, knowable, and predictable at a local level, the intersection of multiple layers of software, hardware, and enabling infrastructure, as well as their surrounding environmental conditions, can crystallise behaviours that are impossible to foresee.

This reading has important consequences for understanding what is represented by the technical and poetic operations of The Ephemerides. As material artefacts, the precision and predictability of digital systems are the product of many interlocking and interactively stabilised performances, borne through years of technical experimentation and applied experience, seeking to filter the disruptive noise of an agential universe into the pristine, discrete signals that are the functional basis of all digital computing systems and communications networks (see Link, 2016). Nevertheless, this hard-won stability represents a challenge for the many instances of automatic text generation which require a source of stochastic values to provide the dynamo behind their varied outputs. Most generators, including The Ephemerides, deploy a random number algorithm for this end, but in a computational context, this system is described more appropriately as being ‘pseudorandom’, for the signal processing architectures that enable digital computing ultimately foreclose the native fostering of noise within the system, and so, unless a hardware-based solution is implemented, only an algebraic emulation can be utilised. Effectively, the goal becomes the intentional mimicking of the very contingency that the foundational performances of digital computation otherwise strive to eliminate.

Although the texts generated represent the intentional outputs of specifically crafted software, their chief aesthetic draw comes from their potential to make manifest the unexpected, especially given their origins from within a deterministic machine. It is this unpredictability that draws generative outputs into a resonance with the types of error yielded by the unravelling of normative technological operations, which Goriunova and Shulgin characterise as a productive breakdown in established patterns of stabilised, constraining, behaviour, and thus affording novel insights into the expressive potential of digital systems more broadly: ‘An error [is] a sign of the absence of an ideal functionality, whether it be understood in the technical, social or economic sense’ (2008: 115–116). It is in this context that the relationship between the grainy imagery of The Ephemerides, as situated alongside its poetic operations, and overall thematic focus, can be understood fully. In its speculative staging of the poetry of space probes, The Ephemerides articulates how the glitches seen in the genuine probe imagery are reflected in its potentially anomalous poetry – with both juxtaposed against pristine planetary photography and the passivity of the obedient digital computer – and so bearing out the status of all the elements involved, from the target objects, the spacecraft sensors, the Earthbound receiver arrays, the databases, image processers, and social networking environments, as interlocking agencies, with their own expressive capacities, rather than empty conduits of human intention.

As a bot program, The Ephemerides represents another such entity involved, and another step in an ongoing dance of agency that takes systems of knowledge and technology developed in one context and redeploys them in another, creating an assemblage that catalyses a not-dissimilar exchange on the part of the reader, whose expectations and understandings of written expression are resisted and recast by their arbitrary remixing, yielding new interpretive vectors in the process. Even in cases where no meaning is immediately forthcoming, an important aspect of generative Twitter bots is the way in which their iterative outputs circulate across the social network, resonating variously with specific observers, and being forwarded along by its embedded sharing mechanisms. Such dissemination highlights not simply a kaleidoscopic splintering of subjectival perspectives, but how the expressive power of these bots is symbiotic with the fact their outputs propagate in precisely the environment in which they are most likely to have their interpretative potential realised – that they are enacted collectively at the level of their reception, as well as their operation, within the networked milieu in which they manifest.

More poignantly, amidst the intensity of concern around the use of clandestine bots in shaping online discourse for politically subversive ends, the openly curious activities of generative bots, inserting themselves into tightly contested timelines, does gesture, however faintly, towards a form of creative disruption that reveals the network, and its algorithmic architectures, as offering potentials for activity and expression beyond facilitating various threatening agendas. Admittedly, such interpretations are far from guaranteed, for just as political bots present a facade of authorial intentionality, a cursory reader could ascribe to generative bots similar such qualities, even in the knowledge that they are artificial, and so treat their operations as but a beguiling curiosity. This recalls the so-called ‘ELIZA effect’ (Hofstadter, 1995: 157–158), named after the pioneering ELIZA ‘chat bot’ developed by Joseph Weizenbaum in 1964, which managed to sustain remarkably intimate, typed exchanges between itself and its human users, despite being presented as a machine (see Weizenbaum, 1966; Bassett, 2018). Such effects can undermine the disruptive self-reflexivity of artifice, and while it is beyond the scope of this article to debate this issue in-full, it is perhaps worth recalling Compton (2016) and Kazemi’s (2016) observations concerning the simple value of generating strangeness and surprise, in at least showing that bots are not always the obedient tools of corporate or political actors.

To draw these different observations together, it can be stated that The Ephemerides offers a meditation on the founding conditions of its technical being, redeploying both the technologies and the thematics of space exploration to reveal the expressive potentialities of systems and infrastructures that would appear far removed from the goal of creating art. In so doing, they are recast as performative agencies, whose actions bear out the intersection of different material, technical, and cultural imperatives.

Despite the note of caution that preceded this analysis, it can be observed that such a technically oriented reading is not alien to the specific outputs generated by The Ephemerides. Here, Simanowski’s characterisation of generative writing as animating and extrapolating themes and imagery depicted within the source text remains especially pertinent, for in Sepharial’s book on astrology, and Ballantyne’s dissertation on the far reaches of the Earth’s oceans, The Ephemerides derives its poetry from sources that are engaged explicitly in questions concerning worlds beyond the human, and how these might be depicted and understood.

For Sepharial, such questions and their implied relations, however arcane, can still be cast in practical terms, with his book detailing the interpretative vectors necessary to procure a personal horoscope that aligns the movements of planets with personal fortunes. Leaving aside the scientific veracity of such endeavours, the task is nonetheless one involving a flattening of any established hierarchy between human and non-human modes of agency, characterising daily life as a negotiation between both, whether explicit or otherwise. In the case of the voyages documented by Ballantyne, these negotiations with the non-human, in the form of oceanic currents, trade winds, ice floes, and hidden reefs, are especially concrete, and his book notes a range of scientific endeavours that seek to measure and model these phenomena, as well as participating itself in the challenge of how they might be captured subsequently within the frame of written language.

As evidenced in the earlier excerpts listed in Figures 1 -4, the mixing and reformatting of these two texts into poetry produces imagery that is often enigmatic. However, in the terms outlined above, it is apparent how such anomalous compositions are an apposite way of depicting a world whose richness of becoming far surpasses the carefully turned performances of the human technocultural domain – with strange juxtapositions of words, like the grainy images of the cosmos, suggesting not just the contingency of technical formations, but of an agential universe more broadly. Just as space probes are deployed to negotiate and interprete this domain in ways that prompt human observers to review their understandings, reading the poetry of The Ephemerides implies a reimagining of the operations of meaning to encompass the potentialities of language recast by machinic modes of composition, and of what these modes represent in-themselves as expressions of the contemporary environment.

Conclusion – Future readings

The reading conducted in this article has concentrated on a single instance of bot-generated poetry and has oriented its questions and observations to the particularities of this instance. Nevertheless, in tackling Simanowski’s contention that the functioning of code is of limited significance for the question of meaning in generative textuality, it has outlined a range of perspectives that are of relevance for future such analyses.

Foremost is the entanglement of a bot’s output with the network infrastructures in which it is produced and disseminated. Although other instances of bot-generated poetry might not afford the explicit opportunities presented by The Ephemerides for engaging these aspects as part of a thematic analysis, the networked milieu in which they operate still demands a consideration of the potential relationships and resonances established. Indeed, there are many avenues of enquiry here that have not been explored extensively in the above discussion, but will bring to bear useful perspectives concerning the wider cultural status of bot generated art and poetry more widely – that is, the question of what it means to convey automated poetry within the context of the Twitter social networking environment specifically, or what it might mean especially in light of contemporary fears concerning bots engaged in propagandistic actions.

Another aspect of critical import concerns the status of Twitter bots as distinctive software artefacts embedded within a larger set of digital infrastructures. While the source code of The Ephemerides has not been made available publicly, other instances, such as poem.exe, permit this information to be accessed and analysed. Certain approaches, such as those headed under ‘software studies’, would encourage a forensic engagement with such material, so as to unearth the ways in which it structures the outputs generated, and to consider how the generative process itself is of significance for what these bots represent as a form of software art (at least within the constraints and affordances of the Twitter platform). Nevertheless, in the absence of any source code – including, indeed, that of the commercially propriety environment in which it operates – the above reading still demonstrates that a purely empirical approach that is sensitive to the basic operations common to much generative software, and the fundamentals of digital infrastructure more generally, can provide important avenues of consideration when assessing the varied interrelations between a bot’s thematic framing, its key enabling systems, and those of the contemporary digital environment.

Finally, The Ephemerides is not unusual for a bot in its juxtaposition of the visual and the textual, with the former often derived from generative operations or drawn from extensive online databases. As demonstrated by the above reading, these multimedia aspects can provide important interpretative cues regarding how either aspect can be assessed in terms of the broader themes and attitudes they evoke, and future analyses will thus have to be cogent about the particularities of digital visual culture, such as Goriunova and Shulgin’s noisy ‘software aesthetics’, in order to evaluate both the images themselves and their significance for the textual outputs that accompany them.