Is this text written by a machine learning agent or a human scholar? Even though I assure you that the latter is the case, you will likely be aware of the legitimate possibility of the former. The fact that this question does seem unwarranted points to the intensity by which machine learning agents – known as large language models, such as OpenAI’s ChatGPT – are able to mimic humans and produce humanlike language. These systems can generate poems, song lyrics, and stories that often emulate the styles of well-known authors and are generally recognizable as being of a literary kind. In short: they imitate human language with a hitherto unseen efficacy. The advent of large language models has triggered excitement, fear, and discussion across contexts – in the technology industry, the academy, popular culture, and public debate. Their unprecedented capability to imitate human writing seems to usher in a new situation, wherein questions of computational creativity, artificial intelligence, and the very concept of humanity seem crucial. In this situation, however, one question often goes unnoticed and unanswered: what does it even mean that machine learning imitates human writing? What does imitation do, what does it infer?
The research project Machine Mimesis takes on the question of what imitation means in the context of machine learning. The project focuses on what it means to mimic in an age of machine learning, which is a subset of the larger question about what happens to literature and literary culture when the production of language, and especially linguistic imitation, has become algorithmified by machine learning. To that end, the project anchors the notion of imitation in the concept of mimesis, which has roots in Ancient Greek philosophy and still today plays a central role in our conceptions of similarity, representation, and imitation.
As its main contribution, the project coins and develops a notion of machine mimesis: an emerging literary practice that exists at the intersection of human and computational imitation. As a central feature, machine mimesis is based on human participation in the mimetic loop, which means that we are not just dealing with computers that seek to imitate human writing, but with humans that imitate computers’ imitation of themselves, which creates an at times dizzying recursivity that is the primary source of literary and political agency of machine mimesis. Machine-mimetic practices negotiate and, at times, resist the drive towards indistinguishability in machine learning by sustaining an imperfect, friction-filled, and poetic kind of mimetic practice that reworks the imitative logic of machine learning from within.
The project identifies and analyzes the operation of machine mimesis in three instances, with a view to many more. First, machine mimesis plays out as the imitation of software that imitates humans in online communities, conceptualized as bot mimicry. Second, machine mimesis operates as an emerging yet distinct poetics of misrepresentation in contemporary digital poetry and prose alike, where authors collaborate with language models in reciprocal mimetic ensembles. Third, machine mimesis instantiates a mode of practice-based knowledge development that utilizes the epistemic qualities of prototyping, which also affords a view to alternative modes of designing the language models themselves. In sum, the project sheds light on the often unnoticed mimetic aspect of machine learning, with a special view to its literary dimensions.
Machine mimesis opens up the imitative relation between humans and machine learning and enables difference, critique, reflection, and renegotiation. This stands in contrast to the otherwise dominant assumptions about machine learning and imitation alike, both of which are often considered unilateral forms of assimilation – concerned with optimizing indistinguishability. Against such optimization, machine mimesis illuminates trajectories by which we can learn to read, write, and live with machine learning through reciprocal imitation, allowing us a more prosaic relation to machine learning and allowing us to affect its propagation in digital culture.
This collection of platform poems were based on a corpus consisting of Zoom transcripts, Zoom chats, abstracts, and conference papers from the 2021 Electronic Literature Origanization Conference and Festival, co-organized by DARC.
Platform poems constitute a literary approach to data analysis, using machine learning to transform datasets into poems thus allowing for literary and material engagement with text-based datasets.
A platform poem consists of (1) a title that is also the query we fed to the machine in order to generate the poem and (2) a poem that takes the form of an array, that is, the output for the query. The relation between title and poem is governed by a K-Nearest Neighbors model, which is optimized to look for the closest, so to speak, vectors in a dataset based on a machine learning-based statistical operation of myriad data points.
With this approach, data becomes poetic and poetry becomes a literary way of knowing in data culture.
*** Based on and inspired by code by Allison Parrish.
Participation in event hosted by the Machine Vision in Everyday Life (University of Bergen) project group.
In November 2021, around 50 people (approximately 20 researchers) participated in Sivilisasjonens Venterom - a live action role play (LARP) focusing on machine vision, surveillance, and ethics. The LARP centered on a fictional future, where a post-apocalyptic world is riddled with environmental and military damage. Few places remain habitable. One such place is Sivilisasjonen [Civilization], in which Intelligensen [The Intelligence], an advanced AI, is responsible for all major decisions, and where personal scores (continuously calibrated by Intelligensen) determine the value of individual humans. The plot of the LARP takes place in the 'waiting room', where outsiders are evaluated and possibly granted entry into Sivilisasjonen. MSE Participated in the LARP, playing the role of Trinidag, an Observer whose eyesight was directly linked to Intelligensen.
The LARP functioned as an autoethnographic study as well as a design experiment and gave valuable insight into the dynamics of decisively using mimesis (in the form of roleplaying) as an approach to machine learning design.
Collaboration Anders Visti.
This work was commissioned for the 2021 ELO Conference and Festival: Platform (Post?) Pandemic. It was also a direct continuation of the efforts made in the MOBBOT workshop series (see below).
Aarhus Urban Operating System (AaUOS) is situated as a parasitic ‘flipside’ of the ELO 2021 conference website. On the AaUOS website, you’ll find a chatroom populated by e-literary bots that are trained to be connoisseurs of certain aspects of the city of Aarhus. The bots of AaUOS are based on equal parts handcrafted conversation trees and recurrent neural networks (RNN). Each bot is a character in a metropolitan drama, from the head of the tourist department to the local bog body, the Grauballe Man.
The RNN models are trained on texts about Aarhus as it is in its presents and pasts, as well as urban development plans that represent an increasingly gentrified future of Aarhus. The work furthermore entails multimedial content, from images to sounds. The sounds were recorded locally in Aarhus as part of an earlier version of the work, which this new chatroom-inflicted version is based upon.
By creating an interface that connects these generative corporate visions and city-imaginaries with a virtual conference on electronic literature, AaUOS sustains an e-literary encounter with emergent imaginaries of Aarhus, stemming collaboratively from local city developers, a machine learning algorithm, playful imitative writing practices, and an international community of scholars.
Collaboration with Christian Hagelskjær From and Anders Visti.
A collaborative investiation of the dynamics of mimicry between humans, AIs, and everything in between. On August 29th, 2020, the workshop series "Mob Programming for Bot-mimicry" initiated, setting the code&share[ ] collective on a course to design and develop a platform for curious experimentation with bot-mimicry, be it a conversational game or a performative setting. We ended up hosting three sessions which led to further development in a smaller group.
Taking collaboration to the next level, the MOBBOT project approaches both design and development of the platform through the practice of mob programming - the practice of programming collectively in a group (or mob), taking turns operating the mob's single keyboard.
The workshop series segued more or less directly into the creation of AaUOS (see above).
Literature in Digital Transformation was a project collaboration between Roskilde Libraries, Herning Libraries, Helsingør Libraries, Aarhus Municipality's Libraries, and Litteratursiden.dk.
As part of the project, MSE interned for three months, taking part in the the development of a new iteration of the Poetry Machines, and furthermore partaking in the development of a new teaching platform for digital literature. The teaching platform is the first widely available of its kind, and was based directly on previous research-based experience as well as Erslev's ongoing research into the potentials of digital literature to sustain the development of post-digital literacy in K-12.
Collaboration with Mitra Azar
"DoppelGANger.agency believes that every human on Earth needs to find their algorithmic double – a first step towards a new idea of privacy concerning facial recognition and biometric technologies at large." (from project website).
The project points at raising the question of humanity’s aesthetic and emotional extinction, attempting at finding the humans in the midst of the latest technological disruption. The project is a radical and ironic gesture which mixes algorithmic art and street art, questioning the relation between online and offline world, the human and the technological.
MSE partook in early concept development, and produced the initial iterations of the project, including the iteration that was exhibited at the 2019 Havana Biennial, as part of !!!Sección A R T E (cf. http://www.nestorsire.com/act-no-22_12-04-2019/).
Collaboration with Søren Pold and Jakob Fredslund, co-produced by CAVI
The Oracle from Selphie is a new layout for the Poetry Machine. The Oracle instantiates a close resemblance between ancient oracles, horoscopes, machine learning, and social media. The Oracle from Selphie lets the reader create new and unique texts from a corpus of available sentences, all of which parody the style and content of horoscopes; some sentences give statements about the reader’s current mood and thoughts, while others predict future events.
With the graphic design resembling a kitsch-like version of a social media feed, and an introductory text referencing machine learning and computational statistics, these horoscopes are put into a post-digital context. Here, it becomes evident that most of the texts we read, write, share, etc. online are highly pre-defined. The connection to ancient oracles highlights how these strange predictions often require interpretation and specific action from the recipient – the question of whether social media is mainly documentation of or template for our lives becomes evident.
MSE's interest has mainly been on the practice of invoking references to machine learning in a piece which is in no way based on machine learning techniques. In particular, MSE am interested in the way we attribute functionality to the system based on shared cultural conceptions, as embedded in tech narratives around machine learning.
f-ah-n-eh-t-ih-k_m-ih-r-er [Phonetic Mirror] is an experiment into natural language processing (NLP). It investigates the relation between corpus text and output, and it troubles notions about ‘learning’ present in machine learning discourse. Phonetic Mirror lets you build a corpus by talking to your computer – the computer only ‘knows’ the words you say to it. In addition, the Phonetic Mirror only ‘learns’ the words based on their phonetic structure – the structure individual syllables present in each word you say.
Phonetic Mirror then talks back to you, creating words (or sound poetry) based on the learned relations between phonemes. As such, Phonetic Mirror operates closer to an extracted sound-similarity than to any grammar when creating new words based on you input.
Phonetic Mirror is an investigation into the mirrored relation between person and interface: who mimics whom in this phonetic dance? To what extend are we inclined to label the program ‘natural language processing’, given that what it produce is quite far from anything we would usually consider to be NLP. The output seems to make close to no sense – though the algorithm is based on NLP-processes and ‘learns’ in a way which seems closer to that of human language acquisition: by listening to and copying the phonetics, not the grammar, of language.
*** Based on and inspired by code by Daniel Shiffman, Daniel Howe (the RiTa library for p5.js), and R. Luke DuBois (the p5.speech library).
