The broad rollout of automation technology is not simply a question of technological development, but also economic calculation, and thus it always exists in price competition with human labor.Moritz Altenried, The Digital Factory: The Human labor of automation
Last January I began teaching an experimental “Automation & Writing” course in our Professional & Technical Writing Program. The aim of the course was to theorize and train alongside the many forms of automation facing technical writers today. Such automation spanned digital technologies that prompt design automation (e.g. Microsoft 365’s Designer), established PTW practices like Component Content Management (Batova & Andersen, 2017; Batova, 2018), and, what proved more fascinating to the class, platforms that promise content written with AI through Large Language Models, like OpenAi’s ChatGPT or Jasper.AI. As a teacherscholar who remains theoretically invested in our practices of valuing labor (e.g. Horner, Inoue, Trimbur), the affectual and economic implications of automation rose to become a primary focus of the course.
Automation, as Altenried (2022) remarks in my epigraph, is always in competition with human labor. Despite the repeated assurances of some of my senior colleagues who at the time insisted that they “can always just tell” when something is LLM generated (I’ll have what they’re having): in a contemporary digital atmosphere that seemingly embodies “move fast and break things” as a mantra, the slow divination of a select few is a paltry guarantee to the graduates entering the workforce tomorrow. Instead, what was useful in the course, and indeed, in my own thinking about the future of writing in a world of content-generating machines is thinking about writing work as an extension of the factory or, as Altenried calls it, the digital factory that is “an apparatus and logic for the ordering of labor, machinery, and infrastructure across space and time.” (p. 6).
Approaching these emerging technologies of automation through the lens of the digital factory clarifies at least two different pedagogical approaches. On one hand, there is the approach toward theorizing and acting on labor and activism within the technologies themselves. Teaching the potentials of Platform Cooperativism and Activist Design, if we can make inroads with future technologists, is one way of acting this out. On the other hand, nevertheless, there is some value in treating these platforms as the newly articulated technological regimes of the workplace. This is not an endorsement of AI but, barring some impossible moratorium on automation technologies (because you can’t put the magic of AI back in the bottle), an immediate practice for graduates of tomorrow.
This summer I’ve experimented with various ways of putting LLM generation technologies to use in the context of Professional and Technical Writing work. I want to preface this section by saying that I remain skeptical of the potential for high-paying “prompt engineering” jobs (note, engineer rather than writer or author). If such machine-generated content exceeds the value of human-generated content then it’s just as likely that the prompt-engineering jobs will quickly be supplanted by the sort of procedurally generated algorithmic-prompt engineers. Nevertheless, if the boom-bust cycle of LLM plateaus at a new AI Winter then there may be some value in learning the tools. Toward this end, what follows is a practical application of using Chat-GPT to produce User Personas for an upcoming faculty development retreat.
With the fall semester nearing, I’ve been preparing faculty training for our Composition program. As the former director of our PTW program and the new director of our Composition program, I have wanted to apply a few PTW methods to conceptualizing our composition program. In particular, I am using User Personas to prompt instructors to think about student differences with regard to the classroom (design and practice). In order to do this I created a thread with Chat-GPT 3.5 and fed it student demographic information from Weber State University. Additionally, I provided the model with University Writing Program goals and a summary of the University’s strategic plan. In three separate instances, I then prompted the model to create user personas for a traditional student, a non-traditional student, and a student who primarily speaks Spanish at home. I summarized these personas and used MidJourney to create a likeness for each of these hypothetical students.
With very little training, ChatGPT was able to quickly produce the following (unedited) three personas. I spent less than a few hours of an afternoon revising and clarifying the outputted personas to direct them more toward our composition program’s specifics (including a heavy digital emphasis and bringing in the context of our commuter student population).
In general, I found the user personas to be easy to generate and revise. Nevertheless, there were downsides too. In particular, the personas rely on stock characteristics that in some situations can, when carefully approached, be useful prompts in understanding students but also threatens to bake in various forms of hegemony. For instance, when prompted to create a User Persona for a student from a Spanish-speaking household, the model created Ana and shaped her goals and challenges around cultural integration and linguistic difference as a barrier rather than a resource.
I want to offer two takeaways from this experience. I’ll first begin with how these personas, even while imperfect, will guide an upcoming faculty development program, and then I’ll discuss the potential of using machine-generated personas in our PTW courses. Even though they’re far from perfect, these machine-generated personas materialize potential student goals, attributes, and challenges. In my experience, it’s easy for a room of teachers to discuss pedagogy as a purely abstract practice. Additionally, the competing (and often contradictory) goals of a university program and our personal (or field-specific) pedagogical aims, when unstated, tend to flatten in the direction of institutional legibility. In a faculty training session, these personas may offer points of friction to bring such tensions to light. For instance, Ana’s persona may prompt: is Composition a locus of enculturation/cultural integration?; and what are the consequences of tacitly approaching Academic Written English instruction as cultural integration?
Lastly, I want to touch on the potential for using machine-generated personas in PTW course contexts. I still believe that humanistic creativity, analysis, and ethical reflection remain the capstone skillsets of a PTW degree. With that said, there is a lesson in using an LLM to generate a user persona or, perhaps, other PTW genres. Although these technologies currently create a good-enough genre performance of a user persona I think it’s worth using them as a foil. For instance, we might ask what would improve the output. Is it more data and increased quantification of our users?; I certainly hope not. Instead, I hope that such machine-generated genre performances illustrate what is missing in many unreflective genre performances: human creativity, weirdness, and willingness to say something a little different.