Inventionless Invention with Grammarly Go
Once more, it's "not bad" if you already know what you're doing as a writer.
This is the second of a Prose and Processors series on the explicit and implicit pedagogies embedded in various generative AI technologies. As I wrote in my overview of the series, I aim to consider what these platforms are trying to teach us, especially about writing, via their theoretical, practical, material, embodied, ethical, and ideological dimensions. Previous posts:
In a recent post,
questioned the current, widespread preoccupation with AI chatbots as the primary means of interacting with AI:It is as if the only way for us to do work is through texting an intern… except you are texting a different intern in every chat, one who forgets everything you had previously discussed, and whose memory starts to fail after just a couple of pages of text.
He called attention to other ways of interacting with AI, including AI devices like the Rabbit R1 and AI copilots, which are “narrowly focused to help with a specialized task.” I think he overstates the case when he says a copilot “just completes a task for you” but potentially undersells the case when he says, “They make AI easy to use, but in doing so, may distance users from understanding the underlying LLM’s capabilities and limitations.” I see this as underselling because copilots can also shape, and potentially misshape, our understanding of other kinds of tasks, like writing. Enter: Grammarly Go (hereafter, GG). (Disclosure: I accessed GG via an institutional license.)
GG uses Azure OpenAI as its LLM. There are three ways of interacting with it: online using Grammarly’s web word processing interface, when using web apps, or as a plugin for Microsoft Office and numerous other desktop apps. For this post, I will focus on Microsoft Word, where it functions like a copilot. The plugin appears as a floating widget. It includes a count of grammar and style suggestions (also available through the basic Grammarly tool) and a green lightbulb, which accesses GG’s interface. The widget thus visibly but unobtrusively separates “grammar” (error detection, stylistic suggestions, and clarity suggestions) from its generative AI chat interface.
The green GG lightbulb suggests that the broad purpose of the tool is to get ideas, but what you get when you click the lightbulb is a mishmash of ideas, genre suggestions, writing process tasks, and rhetorical actions.
For example, with a blank page for a new document, the options are “I’m not writing for school” and “Brainstorm topics for my assignment.” When I click “more ideas,” the options include additional ways of engaging school-based work; help coming up with ideas for non-school writing; and “Write,” which includes some genres (e.g., “Write an engaging introductory email”) and some actions (e.g., “Politely ask for an update”). Below these options is a text box where a user can request help from GG with tasks not represented in the suggestions.
A screen shot of some default menu items in Grammarly Go.
Pedagogically speaking, GG’s interface thus rests on a simple “school/not school” binary, which is undoubtedly due to the company’s two main customer bases—schools and business—rather than a principled pedagogy. The school-based (and even some of the non-school-based) options suggest a linear writing process: topic selection, research plan, and outline. Mostly absent is the notion that users might be writing in response to an authentic communication need, which risks reinforcing decontextualized, school-based writing pedagogies. That said, some default options are at least superficially rhetorically oriented, such as “Ask for a favor.” These options just aren’t the school-based ones. Thus, as with most of these technologies, instructors must fill in the rhetorical gaps in their assignment designs.
Like Packback Writing Lab, GG uses suggested prompts to help users interact with the chatbot. Unlike Writing Lab, GG allows users much more freedom in how they interact with the AI. For example, when I clicked “Generate ideas for a blog post,” it produced three options:
An organic gardening lifestyle
Exploring the benefits of remote working
Insight into the power of automation
Curious, I copied the last one into the textbox, and it suggested the following:
Exploring the Symphony of Efficiency: How Automation Enhances Music Production
The Art of Automated Cuisine: Revolutionizing the Culinary World
Beyond Science Fiction: Real-Life Stories of Automation Changing Lives
Gardens of the Future: How Automation Can Green Our Urban Spaces
Secrets of the Automated Classroom: Shaping the Next Generation of Learners
I want to pause here and echo what I wrote above: this part of the process felt distinctly arhetorical. Although I write about AI, this interaction felt like pulling ideas out of the blue rather than responding to a specific need for my audience or a situation I believe needs addressing. Pedagogically, it feels like “inventionless invention.” In simple terms, “invention” is the process of developing arguments or ideas for communication. Invention always happens in context, and it is shaped by audiences, genres, occasions, and disciplinary or professional norms. This particular use of GG felt like it was shaped by nothing whatsoever except maybe the genre of the blog and my general interest in AI.
Nevertheless, I was intrigued by the last bullet. I mean, holy cow. The automated classroom shaping the students of the future? Talk about foreboding. I couldn’t wait to see what GG had to say about this.
I retyped it into the textbox and GG generated an outline with several sub-sections (some of which I have omitted for brevity). The basic structure was:
Introduction
Understanding automated classrooms
The benefits of automation in education
Personalized Learning: Discuss how automation allows for individual learning paths, catering to the unique needs and pace of each student.
Increased Engagement: Explain how interactive and adaptive learning tools can make education more engaging and fun.
Efficiency and Accessibility: Talk about how automated systems can provide education to a wider audience with less need for traditional resources.
Challenges and concerns
Address concerns regarding the lack of human interaction and the potential impact on social learning.
Discuss data privacy issues and the ethical use of student data.
Consider the challenges of ensuring equitable access to technology for all students.
Case studies and success stories
Preparing for an automated future
Conclusion
Once again, this felt like “inventionless invention.” With no sense of audience or purpose, how could I decide what my readers might need to understand? For whom might there be benefits, challenges, and concerns, and would those be the same for all stakeholders? What kinds of case studies would I need? The answers to these questions have to come from me, as does the ability to know which questions to ask in the first place. And without answers, the suggestions here read like a wikipedia article, not a blog. The suggested structure supposes a neutral, informative stance, which is at odds with what I take to be a highly voiced and opinionated genre. In other words, GG flattened its own pedagogical potential to teach processes, genres, rhetorical actions, and what it means to have an idea.
Thus, as with other generative AI tools, you have to know something about what you’re doing to get the most out of GG, and the more you give it, the better it will respond. To extend this example: after receiving the above outline, I clicked “Identify gaps” in GG, and it responded, “What are some examples of schools, districts or countries that have successfully implemented automated classrooms?” Beats me, I thought, so I put the question back to GG, and it suggested educational technology journals and magazines, government and educational organization reports, academic research databases, and even education technology conferences. To its credit, it also provided examples of these. Still, I would have to know how to find these venues, access them, search them, and read and use the relevant information, which amounts to a heavy information literacy task.
GG thus presumes users have a lot of composing and literacy knowledge ready at hand. Of GG, Grammarly boasts:
Unblock your ideas and accelerate team and individual productivity. Click the Grammarly button to compose, ideate, rewrite, and reply with a ready-to-go AI co-creator that understands your communication context and goals.
However, novice learners might struggle to use it in these ways without deliberately scaffolded interaction and learning experiences with the platform—which can be provided by live instructors and tutors. It may be helpful for more experienced writers who simply need a sounding board to jumpstart their work, but by the time they get to that point, they may already have a set of writing habits that make anything but suggested rewording mostly irrelevant anyway. After all, ideas and genres arise out of rhetorical situations, not copilots.
Your observations on the "arhetorical" nature of the tool-and I would say LLM writing assistance in general-is the key for me because that rhetorical situation (message, audience, purpose) along with the underlying audience analysis (audience needs, attitudes, knowledge) is the underpinning of the thinking I do as I write something and the GG generation is absent any of those concerns, of course, because LLMs have no capacity for that kind of reasoning. It's always going to be a simulation. The simulation may be more or less convincing, but if I can discern between the good and bad stuff from the LLM, as you say, I probably already have a robust process where it isn't helpful. If I don't have that capacity, I'm just letting the tool throw stuff at the wall and hoping it sticks.