Composing with AI

Themes of the Collection

This collection is designed for scholars and instructors in the field of Computers and Writing, as well as those in adjacent disciplines such as Rhetoric and Composition, Digital Humanities, and Communication Studies. It delves into the multifaceted intersection of technology and writing, offering insights into the evolving landscape of digital communication and its impact on pedagogy and research. While the chapters in this collection can be read in any order, we have organized the collection into six main themes: histories, policies, applications, style, multimodal composition, and theoretical frameworks. The corresponding navigation helps readers choose which topics to visit.

By addressing the dynamic interplay between technology, writing, and education, this collection seeks to facilitate a nuanced understanding of the digital age and its transformative potential for the future of scholarship and instruction. We are grateful to the contributors for the wide range of topics. Given the rapid pace of change in AI technologies, our aim was for this collection to be a valuable resource for diverse audiences including faculty, researchers, administrators and policymakers who are grappling with the implications of AI for teaching and learning. The chapters help with perspectives necessary to navigate the complex implications of AI for teaching and learning, fostering informed decision-making and responsible implementation of AI-driven tools and strategies in educational settings.

Histories: C&W Approaches to Disruptive Technologies

We decided to start the collection with a historical approach that situates our current moment within the context of how the field of computers and writing has taken on the challenges of new technologies in the past. In 100 Years of New Media Pedagogy, Jason Palmeri and Ben McCorkle argue that instructors have always used various multimedia technologies in the English Studies classroom, even before the advent of the Internet. Using the findings from analyzing a huge corpus of over 700 articles over 100 years (1912-2012) and providing examples such as audio-visual aids and typewriters in the classrooms, they highlight how varied levels of technological engagement shaped the history of writing. Some popular technologies that have aided the evolution of writing pedagogy include MOOs, OWLs, webtext, hypertext, computer programming (HTML), multimodality, and social media (Palmeri, 2012; Palmeri & McCorkle, 2021; Marlow & Purdy, 2021). Tracing this history helps us understand the evolution of the field with AI.

Some early concerns about technology in writing spaces came from a critical view of technology that defines technology in terms of human–computer interaction and a contextualist view of writing--a scenic view that focuses on production and effects of writing in its political, social, and rhetorical context (Porter, 2002). Early works by Donna Haraway (1991), Jim Porter, (1998) and Pat Sullivan (1991) described the relationship of humans with technology, especially the networked and social nature of work that results from its uses. Writing studies research tended to focus on the intentions of writers and outputs generated through the use of technology rather than focus on the technology itself.

Policies: Programs and Publications

New technologies produce tools for writing that often disrupt prior pedagogical and social norms. The chapters in this section focus on institutional responeses to the disruptions generative AI poses. These policies focus not on what of AI is or does, but on how we can responsibly use these new tools. Rhetoricians have traditionally given more weight the “how” questions over the more instrumental "what" questions or the more philosophical "why" questions. We ask questions such as "how do we use this technology?" and "how will the technology impact writers and their audiences?" The chapters in this section will answer aim to provide some answers to these questions in concrete terms.

Policy making for generative AI has become important for every sector. We have noticed that some institutions have prioritized policies around acceptable uses for students, while others have decided to wait until we have more certainty about how these new tools will actually work in practice. Both students and faculty have been asking for instutional policies to guide generative AI use; the chapters in this section provide a policy model for teaching and an evaluation of policies developed for academic publications.

Reports from the Field: Classes & Students Using AI

Across all levels of education there have been debates about whether AI could be beneficial to students' learning or detrimental to the development of critical thinking skills and creativity. Chapters in this section provide examples of both faculty-focused and student-led examinations of how generative AI might be used in writing courses. We take the position that trying to ban AI use outright is not a helpful response to the challenges posed by these technologies; rather, students must be taught to use them effectively and ethically. Too often we see faculty operating at one of two extreme positions: no AI on one end or extolling an anthropomorphized, nearly sentient system that can be used to do literally any writing task on the other. The aim of chapters in this section is to show how careful consideration of appropriate use in the teaching of writing can help us better understand AI, and, in turn, better teach it as one among many writing tools.

Style: Comparing AI and Human Approaches to Style

Chapters in this section both compare human and AI approaches to a rhetorical undrestanding of style and how these systems might lead to collaborative partnerships between human and AI writers. Because LLM-based generative AI tools are designed to output words in a human-like manner, we have an opportunity to better understand the operationalization of style as interpreted through a system that generates it purely via statistcal analysis. As current systems don't have a way to fully incorporate the rhetorical situation (that is, they cannot understand audience, purpose, or context), the best approximation comes through in the form of stylistic choices. Using style as means to assess the effectiveness of generative AI output provides a very promising mechanism for better understanding its limitations and also for designing pedagogical approaches to teaching AI in writing classes.

Multimodal Composing: AI Text-to-Image Applications

Most chapters in this collection focus on textual production via generative AI based using natural language processing algorithms. These systems, however, also make it possible to create images from text. Text-to-image generation uses computer vision algorithms that simulate the human visual system so that computers can "see" and comprehend the content of still images and films. Many AI applications now can use sound for input and output, but at the time this collection was formed, sound capabilities were far less common. Chapters in this section evaluate the effectiveness of generative AI for multimodal composing, providing an explanation of how these systems work as well as examples of more and less successful multimodal composition cases.

Theory: How AI Impacts Rhetoric and Ethics

While humans have a distinct edge in the layered, nuanced complexities of communication, AI writing systems certainly have the edge on processing huge volumes of data. But even with seemingly unlimited data points, many AI writing systems are built on an information transfer model of communication that assumes text production is a simple matter of converting raw data into sentences and paragraphs. This model generally obscures the critical role of audience and context and excludes ethics as an element of textual production (McKee & Porter, 2020). Chapters in this section analyze the gaps between human abilities of context detection as compared to those of AI applications and the impact these gaps in rhetorical practice have on the content produced. Chapters in this section provide theoretical frames that can help guide AI users as well as suggest approaches for AI pedagogies.

A Note on Image Headers

For the main header images of each chapter, we provided imagine.art the chapter abstracts we received from their respective authors. We generated images using several different models (labeled in the system as SDLX, Copax Timeless XL, Creative). It has been interesting to see which abstracts prompted images of people and which tried to produce textualized images (most models can't produce actual texts, instead creating text-like glyphs). Interestingly, academic text prompts seemed more likely to produce black and white images. We typically generated beween eight and twelve images and then selected three to combine into the header image for each chapter. We opted to not edit the images (aside from re-sizing), and we did not tweak the prompts to be more effective, as we were curious to see what the system would do with the kinds of texts we provided.