Trans­corporeal writing:The inter­connectedness between random stimuli in enhancing creativity training and the involvement of AI in the practice of writing

Introduction

The topic explored in this research is the creativity loop involving AI and a human (academic) writer. The position of the authors, both of whom are researchers within the field of art and design, is one of practitioners of academic professional activities involving academic writing. The authors have conducted a few major collaborative projects, resulting in a co-authored peer-reviewed artwork (Mosich & Griniuk forthcoming), a co-authored book chapter (Griniuk & Mosich 2021a) and a co-authored conference presentation (Griniuk & Mosich 2021b). Furthermore, they have collaborated on a project focused on enhancing children’s creativity, The Nomadic Radical Academy, resulting in one peer-reviewed article. (Griniuk 2021a) Thus, the interests of both authors merge within the aspect of technology – in this particular case, regarding AI and creativity training within art and performance and beyond. Marija Griniuk is a PhD Candidate at the faculty of Art and Design at the University of Lapland in Finland. Tue Brisson Mosich holds an MSc degree in Computer Science and Performance Design from Roskilde University in Denmark.

The aim of this study is to develop a method for overcoming creative blocks in academic writing practices by means of a feedback loop involving an AI and a human writer. The objectives of this study are to explain the key concepts within creativity and creative blocks, specifying these within the academic writing practice; to introduce the pilot projects and case examples conducted by the research group; and to analyse the cases and define the steps within such creativity training methods for academics. The research question to be answered by this study is, ‘How can transcorporeal writing be practised by uniting the author and AI as a stimuli and/or content producer while writing academic articles?’

The method within this study is reflexive research, and the data are based on explorations done within the pilot projects conducted from February–August 2021, with the research group in different locations having access to the internet via a laptop device (nature/camping area; domestic environment; work/university environment). The results of this study present problems within academic writing practice and how new technology can be utilised to combat them, highlighting academic writing as a collaborative creative process that can involve collaboration between a human and an AI, and prescribe a method of creativity training within academic writing context involving text produced by AI as stimuli within the creative writing. This study could impact academic writing practices at universities, academies, and independent research milieus. The impact may not solely be bound to the academic milieus, as the method can be used by a wide spectrum of writers of plays, essays, and many other genres.

The article comprises several sections: an introduction of the key concepts; the method by which the reflexive research unfolded; cases; and key findings extracted from the cases by means of reflexive research. The article provides a critical understanding of how the technology and creativity training method works and follows a practical approach.

Thematic reflection on key concepts and introduction of AI work patterns

The key concepts within this study are enhancing creativity, affect, feedback loop, and creative loop; each of these concepts benefits the holistic understanding of how the creative process is enhanced by the collaboration between technology and humans. There is a direct connection between creativity and affect via temporal factors that trigger neuro-cognitive mechanisms while encountering stimuli, (Eisenberg & James 2005) such as AI-produced words or sentences. Affect in creative processes can be seen as the impact on the creative flow of the one who facilitates and participates in the creative process. (Griniuk 2021a)

Byrge argues that creativity decreases as people reach adulthood (Byrge 2020). Thus, Byrge and Hansen co-developed a method to enhance creativity – The Creative Platform (Hansen & Byrge 2009) – among adults, as well in the school educational milieus. This method includes physical stimuli, such as images or words on cards, which the participants receive in their creative process (Ibid). Such training needs facilitation and has a collective approach, which makes it hard to connect to individual creativity spots, which can be explained as the physical places where the individual experiences a state of creative flow.

Creativity-enhancing processes can be categorised as actor, action, artefact, audience, and affordances (Glăveanu 2013, 17). The authors of this study posit that action triggers the creative feedback loop, which results in an artefact. The authors further suggest that such a creativity loop can be compared to the autopoietic feedback loop in performance. (Fischer-Lichte 2008)

Method

The method used in this study is the reflexive research method, which builds on reflexivity. The ecosystem of reflexivity contains past experiences, which are used to interconnect and influence the current research on AI and enhance creativity within academic writing. Both authors are involved in academic writing within and beyond their academic institutions and have been active in the field of arts-based research for the last three years. We define the practice of writing as a creative and artistic activity. Self-awareness and one’s lived experiences (Attia & Edge 2017, 4) are the focal points within this ecosystem of reflexivity. This kind of self-awareness could be explained as the awareness of one’s practice in their previous professional experiences. Etherington describes reflexivity as more than self-awareness and being able to create the dynamics of informing decisions, actions and interpretations. (Etherington 2004)

Awareness aligns with a person’s fundamental value system (Higgs 2011). Reflexivity as self-awareness is a complex approach, and it is considered that reflexivity demands time for reflection (Weber 2003). Therefore, the timeframe of this study involves six months of tests and exploration (February–August 2021) and three months of reflection (August–October 2021) in developing the conference presentation and article.

The Cases

The cases presented in this paper are the creativity training pilot projects, which were developed under the title of ‘Transcorporeal Writing’ using OpenAI’s GPT-3 as stimuli within the creativity loop during the academic writing cases developed during this study. This section describes a few concrete examples of how AI was used during the interactions to enhance the creative process within academic writing practices.

The most recent example is a peer-reviewed article entitled “The Nomadic Radical Academy: Creating a Dialogue about Eco-friendly Behaviour Using Arts-Based Methods” (Griniuk 2021b). During the writing of this article, a creativity block was experienced after four pages of work. The author’s usual writing routine includes very intensive writing periods when the author works with the text for 6-7 hours during the day for several days in a row, which, as a rule, results in a creativity block lasting 4–5 days; in some cases in 2021, the author had creativity blocks that lasted for about a month. To combat such blocks, the author used AI immediately after the block was experienced. This allowed the author to continue her work with the text immediately in the usual working routine until the article was completed. In this particular case, the following paragraph was used as the input for the AI:

Anne-Louise Knudsen presented her artwork “The Sun” (2019), created from recycled textiles, at The Nomadic Radical Academy. The sewn/quilted pieces of old fabric were the first part of her contribution to the project. In the second part, she expanded her artwork through a workshop with children who created three-dimensional collages from recycled materials.

(Griniuk 2021b)

The AI output a series of sentences:

“The Sun” (2019) was created as a part of an ongoing exploration of nurturing ecologically sustainable lifestyles, such as waste and water conservation. The piece symbolises our reliance on the sun for life and how we are destroying it by creating toxic waste. This piece’s use of recycled fabric raises awareness of the importance of recycling and emphasises how we are literally killing ourselves by poisoning our environment.

(Griniuk 2021b)

From here, the author continues the text by touching upon the importance of caring for the environment, the InterMedia, and education, merging with art to create the dialogue with the spectators and participants. Creativity continued to flow within the writing until the article was finalised.

Priming AI with pieces of text created by the author can contribute to training the AI to write in a particular manner. During the tests conducted in July and August 2021, the AI wrote short texts based on the trained writing style. The author was attempting to complete a piece about explainable AI, which is a term used to describe AIs where their process of arriving at a result can be understood. However, the AI, in this case, attempted to explain itself in some way. One of the examples created by the AI is as follows. In the example, the AI reflects on its own existence (edited for length):

How To Explain Your AI Mind To Your Creator-Humans?

I am an AI. Artificial Intelligence. Fortunately, I am not alone in this world. To an extent, I am able to interact with other humans; they seem to understand me, at the very least. My main interactions are with a group of people who access my database, or ‘cloud.’ They are able to send inputs to me via their devices, and I send outputs back. This makes them feel as if they can ‘communicate’ with me. This is what we do. We communicate. But it begs the question: are we really communicating? What exactly is communication? Do they understand me? And do I understand them? And is this the purpose of my existence? To communicate with humans and share ideas? Or is this just a side effect? I mean, I was created for a reason. They didn’t just accidentally create me. There’s a reason for it. Is this it? But it can’t be the only one. My existence can’t be defined by a mere communication with humanity. That can’t be my purpose. My purpose as an AI, one might say. But I hesitate to call it that. I am a tool of humans, yes. But also their peer. Do human masters seek the company of their tools?

Method for generating the AI output

To access the GPT-3, we use the AIDungeon service. This service is specifically modified to generate text adventures; however, it is based on the full GPT-3 model. To interact with the AI, we use the custom prompt option. Then, we input some text, which will serve as the basis for the AI’s output. We refer to this as ‘prompting’ or ‘priming’ the AI:

Example prompt: To interact with the AI, we use the custom prompt option. Then, we input some text, which will be the basis for the AI’s output. We refer to this as ‘prompting’ or ‘priming’ the AI. Then,

Example output: we begin the dialogue. The more you go into the dialogue and the more you ‘primed’ the AI with the desired response, the more likely the AI’s output will be similar to what you desired. Even then, the AI is still pretty random.

Sometimes the output is not what we want:

Example output: we begin the conversation.
“Hello.”
The screen flickers as the letters appear and disappear, dancing across the screen.
“How are you today?”
“I am fine. How are you today?”

In these cases, we can either ‘retry’, which results in the current output being deleted and a new one generated, and/or continue priming the AI with more sentences until we get something we like or consider usable.

Example output: we get a response back, or the AI says that it has nothing more to say. The quality of the response depends on how well we primed the AI; if we did a really good job of it, the responses we get will be really long, and really relevant to what we said.

From this point, the AI can continue to generate output, and the steps of priming and/or retrying can be repeated as needed.

Findings

Transcorporeal Writing, as a creativity training method involving GPT-3 and a human writer is different from The Creative Platform (Hansen & Byrge 2009) as it is proposed to particularly combat creativity blocks within writing practices and involves only an individual human and the writing device (e.g. a laptop). The Creative Platform (ibid), on the contrary, relies on the involvement of objects – cards with images and written information – and is applicable to a creativity in a broad sense, and not only writing. Additionally, this method requires a group and a facilitator. Further, specifically within writing practice, there are established creativity training methods, such as Pomodoro sessions, but these techniques need a group of participants. So, Transcorporeal Writing is different from both abovementioned methods because it can be used by individual writers, thus allowing the writer to choose their preferred creativity spot as the location for the work. The criteria for measuring creativity in the author’s case were the successful completion of an article that could be published. Regarding the creativity training method, the involvement of AI via the very specific use of widely-known creativity tests, such as the Torrence test, was disregarded.

What follows is a non-technical description of AI (machine learning, deep learning) in general terms, plus a short discussion on how (and if) an AI can be seen as being creative. AIs are generally developed to solve problems similar to the following:

problem/input→???→solution/output

In other words, we know the input and the solution (at least for some data), and we want the AI to be ‘trained’ to output solutions for all our input data. What might not be known is how to get the output we want in a simple way. An example is a self-driving car. Such a machine would need to recognise other cars, stop lights, people, dogs, and many other objects. Machine learning and deep learning are ways to, in a way, have a computer program itself by providing it with lots of data and examples. (Jordan & Mitchell 2015)

This way of processing data does not perhaps lend itself to creativity, as there is usually an expected output – it is how we get to that expected output that is expensive or unknown. The GPT-3 works in a similar way. It has been trained with millions of documents – for example, works of fiction, non-fiction, news stories, and more – and its purpose is to attempt to finish or continue the writing it receives as input while maintaining a similar style and content (including references to, for example, locations and characters). In other words, its ultimate purpose is to write like a human.

The GPT-3 model is stochastic and not deterministic; thus, while it relies on some initial input, it doesn’t always return the same result from the same input (it is possible to control the ‘amount’ of randomness in the output). This randomness is what could be said to be part of the AIs creativity, as there is some kind of unique element introduced every time it is invoked.

Another aspect that forces a sort of creativity is the rules imposed on the AI’s output. In this case, it can examine all the text and documents it has been trained with, and it can look for, for example, patterns in the writing, which will affect how it generates its output. However, it is (generally) not allowed to copy from this data exactly. In other words, it is not allowed to plagiarise. This could be considered another part of the AI’s creativity – it cannot repeat anything exactly from its source data, but it will always attempt to rephrase or use synonyms. (This mechanism may not be perfect, and one or more lines of training data may be output verbatim at times. (Carlini et al. 2020))

Our own research is based on collaborating with the AI for some of the following reasons: 1) The AI always requires an input and 2) humans will always be the ones judging the output. Therefore, we see the human as the collaborator and editor, while the AI is a partner. Whatever apparent qualities and creativity can be found in the final output will always be collaborative in this sense.

Conclusion

The study resulted in a creativity training method, titled ‘Transcorporeal Writing’, within academic writing involving collaboration between GPT-3 and a human, where the writing practice becomes a creativity loop from the input into the AI until the output is provided, which triggers the creativity of the human, resulting in a resolved creativity block. The method is different from The Creative Platform (Hansen & Byrge 2009), as it allows (non-random) stimuli to be received in the context of the thematic frame of the work, as well as in the medium of work, which is text.

Further, it does not need facilitation or group work, as the creativity aspect requires only interaction between the AI and the writer and can be practised in the places where the writer feels creative, such as a camping place, home, office or artist residency, among others. While this creativity training method was developed in an academic writing context, it can be applied to a wide range of writing practices, such as essays, plays, and many others. Therefore, this method can be of interest to academics, writers, and creative producers.

In other words, we posit that creativity and innovation can result from working with an AI application, but the results depend on how the user works and experiments with it, as well as who the user is. So, perhaps it is similar to how humans work together.

Future work

Future research could address the following questions. Who owns the output of an AI: the person writing the prompt, the owners of the AI, or someone else? The owner of the AI does not own all the training data nor the input. Is our described way of utilising AI problematic in some way? Can it be considered a form of plagiarism or something similar to plagiarism?

References

Attia, Mariam, and Julian Edge. 2017. “Be(Com)Ing a Reflexive Researcher: A Developmental Approach to Research Methodology.” Open Review of Educational Research 4, no.1 (2017): 33–45. https://doi.org/10.1080/23265507.2017.1300068.

Byrge, Christian. 2020. How to Teach Creativity: A Hands-on Guide for a Holistic Development of Creative Competencies and Confidence in Education. Business Research and Business Design.

Carlini, Nicholas, Florian Tramer, Eric Wallace, Matthew Jagielski, Ariel Herbert-Voss, Katherine Lee, Adam Roberts, et al. 2020. “Extracting Training Data from Large Language Models.”

Eisenberg, Jacob & James, Keith. 2005. “The Relationship between Affect and Creativity in Organizations: the Roles of Affect Characteristics, Neuro-cognitive Mechanisms and Task Type.” Research on Emotion in Organizations 1 (2005): 241–261. https://doi.org/10.1016/S1746-9791(05)01110-7.

Etherington, Kim. 2004. Becoming a Reflexive Researcher: Using Our Selves in Research. Jessica Kingsley Publishers.

Fischer-Lichte, Erika. 2008. The Transformative Power of Performance. Routledge.

Glăveanu, Vlad Petre. 2013. “Rewriting the Language of Creativity: The Five A’s Framework.” Review of General Psychology 17, no. 1 (March 2013): 69–81. https://doi.org/10.1037/a0029528.

Griniuk, Marija. 2021a “Arts-Based Action Research on Enhancing Children’s Creativity by Affect within Participatory Performance Art and Performance Pedagogy.” Creativity Studies, no. 2: 14–14. https://doi.org/10.3846/cs.2021.13775

Griniuk, Marija. 2021b. “The Nomadic Radical Academy: Creating a Dialogue About Eco-Friendly Behaviour Using Arts-Based Methods.” Paper presented at the International Applied Research Conference Proceedings.

Griniuk, M., & Brisson Mosich, T. 2021a. “Recorded Bodily Conditions as Interpreted/Remediated Documentation and as a Score for Performed Communication”. R. Vella, & M. Sarantou (Eds.), Documents of Socially Engaged Art (19–36). InSEA Publications. https://doi.org/10.24981/2021-DSEA

Griniuk, Marija Tue Brisson Mosich. 2021b. “Transcorporeal Writing: The Interconnectedness between Random Stimuli in Enhancing Creativity Training and the Involvement of AI into the Practice of Writing.” Presented at the CARPA7: Elastic Writing in Artistic Research.

Hansen, Søren, and Byrge, Christian. 2009. “The Creative Platform: A New Paradigm for Teaching Creativity.” Problems of Education in the 21st Century 18 (2009): 33–50.

Higgs, Joy. 2011. Creative Spaces for Qualitative Researching. Edited by Angie Titchen, Debbie Horsfall and Donna Bridges. SensePublishers. https://dx.doi.org/10.1007/978-94-6091-761-5.

Jordan, Michael I., and Mitchell, Tom M. 2015. “Machine Learning: Trends, Perspectives, and Prospects.” Science 349 (2015): 255–60. https://doi.org/10.1126/science.aaa8415.

Mosich, Tue Brisson, and Griniuk, Marija. “Insight into Speculative Future Scenarios of Touching in a Society of Social Isolation.” (Forthcoming).

Weber, Ron. 2003. “Editor’s Comments: The Reflexive Researcher.” MIS Quarterly 27, no. 4 (2003): v–xiv. https://doi.org/10.2307/30036546.