6 More is Less?: Using Generative AI for Idea Generation and Diversification in Early Writing Processes

Introduction

Brainstorming in groups is a widely used technique for idea generation both in higher education and beyond. Indeed, existing research suggests that collaborative brainstorming can lead to better and more productive ideas than ideas generated by an individual (Paulus & Kenworthy 2019). As writing teachers, we often harness the benefits of collaborative brainstorming in our students’ early writing process. At this stage, students benefit greatly from exploring their initial ideas with others to refine their position and argument before moving on to the mid-process/organization and drafting, and late process/review and revision stages. In fact, process writing works best when students expend more energy on the early-stage assignments, such as collaborative idea generation, as that is often what will provide the foundation for longer and more sophisticatedly argued pieces.^[1]

While we are strong proponents of process writing, we are also aware that early process work can be time consuming and repetitive, two aspects that can contribute to student anxiety, frustration, and disengagement. In fact, despite its various benefits, collaborative brainstorming, like any collaborative learning activity, is not without its potential pitfalls. For instance, various studies suggest that some students hold back when sharing their thoughts in groups fearing that their peers might be critical of their ideas (Paulus & Murdoch 1971; Diel & Stroebe 1987; Isaksen & Gaulin 2005). In addition, some students might forget what they wanted to share with the group if another group member talks excessively about their own ideas, a problem that is commonly referred to as “production blocking” (Paulus & Kenworthy 2019).

That this problem of self-or group-censorship might not equally affect all student populations becomes apparent when thinking about group brainstorming activities in relation to the existing literature on collaborative learning. Several studies have shown that social privilege affects students’ contributions to collaborative learning scenarios (Hargittai & Walejko, 2008; Voss, 2018, Grover, Ito & Park, 2017). For instance, female students often get assigned secretarial rather than managerial duties during collaborative activities (Aguillon et.al, 2020). On the one hand, collaborative early process work like brainstorming and workshopping ideas seems to promise the diversification of ideas and approaches to a topic. On the other hand, it can also be quite overwhelming for some students to make their voices heard in their groups, particularly if they are students of color, international students, second-language learners, queer, or otherwise ethnically, racially, socio-economically, or culturally marginalized. Those students may easily feel spoken over or be worried that their particular ideas are not important or ‘good’ enough to share. Needless to say, students holding back in group working scenarios negatively impacts the overall quality of the group process including the end result.

Attempting to improve both the quality and quantity of ideas generated during collaborative brainstorming sessions, researchers have explored the benefits of nominal brainstorming, a technique in which group members first generate ideas individually before bringing them back to the group rather than sharing their ideas all at once in real time in a traditional brainstorming scenario. Especially beneficial seems to be the incorporation of various electronic tools such as PollEverywhere or Mentimeter in the nominal brainstorming process that allow team members to share their ideas via a digital platform before presenting them to the group. In this way, teams work individually and yet collaboratively (Paulus & Kenworthy 2019; Al-Samarraei & Hurmuzan 2018). However, while anonymous idea generation via a third-party electronic platform may solve the problem of production blocking and ease some of the pressure on diverse students, those students might still opt for self-censorship in the process. For instance, De Rosa et.al. (2007) suggest that it is inconclusive if anonymity during the nominal electronic brainstorming process actually helps team members produce better or more ideas despite the fact that production blocking should no longer be an issue in these scenarios.

This is where we see some potential uses for generative AI (e.g. ChatGPT and Google Gemini). Generative AI like ChatGPT makes writing processes more efficient in a manner analogous to a calculator. Generative AI in its current state cannot think or argue, which is why it cannot replicate the kind of evidence-based argumentation featured in college writing classes. Our efforts thus focused on AI’s organizational strength as a compilation tool. We were driven by the following questions: should students use generative AI as a collaborative tool to move through early-stage brainstorming more efficiently; would collaborating with generative AI help students create more specific, sophisticated ideas faster than peer-to-peer models of brainstorming; and would using AI to ‘test’ individual ideas before sharing them with a group increase student confidence and improve group brainstorming efficacy.

Drawing on the process of nominal, electronic brainstorming, we created two different prompts, one that improves the efficiency and efficacy of individual idea generation, and the other designed to help increase student confidence in their ideas in collaborative brainstorming situations. Our hypothesis is that this might lead both to more productive, refined group results while helping ease students’ hesitancy in sharing their ideas while amplifying voices by students who would not share their ideas readily otherwise. Students were invited and encouraged to incorporate AI in both activities and to use that work as a springboard for their upcoming projects. Students included all AI use attestations as a footnote on the first page of the final drafts with provided language (“(Name of AI used) was used in the drafting stages of this project. All work presented herein is that of the author(s) alone. Please See Appendix ______ for AI prompts and responses”). Students had to include all AI prompts and responses in an appendix attached to the final product.

Background

The following exercises were administered to students enrolled in a STEM-focused large R1 university in the American Southeast. The student body is more racially diverse than other Research I universities in the region. At least fifty-one percent of the undergraduate body is Asian (31.3%), Black or African American (7.6%), Hispanic/Latino (7.9%), or multiracial (4.4%). White students account for 37% of the undergraduate population. Ten percent of its undergraduate students are international. That diversity does not necessarily expand to gender, as 60.7% of the students are male. These particular demographic breakdowns were reflected in the class under discussion. The class was 53.4% male, 46.6% female. Ninety-two percent of students were people of color, and several were second-language learners. While the groups worked well together, concerns arose about students who are secondary or tertiary English speakers or who are part of traditionally marginalized communities being less willing to share their individual ideas, or who may have felt less secure in their ideas because of language or cultural obstacles.

Both exercises were assigned to a summer 2023 technical and professional communication (TPC) class as process assignments for a technical report and a multimodal final project that included written, recorded, and video elements. While the TPC class is not a traditional composition class, it is a rhetorically aware college writing class. The underpinnings of the class—that all nonfiction writing is, at heart, a rhetorical act responding to and reflecting its cultural context—are the same as those of the first year composition classes we also teach.

The class was designed on the creative-collaborative model similar to those found in some sectors of the tech and creative fields, so all major assignments were completed in randomly assigned groups. Students worked in the same groups throughout the semester for consistency and to develop deeper collaborative relationships. Students were made fully aware of the experimental nature of both activities through verbal instructions and written affirmation (see Appendix A), and students were given the option to opt out of sharing their work. All student work referenced and quoted in the following sections is done so with the written permission of the student.

Intervention

Prompt I: Effective Idea Generation (EIG)

The EIG activity (Appendix B) focused on audience engagement and was broken down into two parts. The first part asked students to examine their university’s strategic plan and assess its rhetorical efficacy for an audience of undergraduate students. The students felt that, while the strategic plan was composed well for faculty, staff, research, and graduate student audiences, it was less successful engaging its undergraduate community. Students were then given ten minutes to individually brainstorm and write down three recommendations to strengthen the plan’s messaging for an undergraduate audience. The second part gave students half the time (five minutes) to ask ChatGPT or Google Bard^[2] to brainstorm ideas on the same topics and asked them to compare the quality of the ideas they had generated without AI against those they had come up with using AI, and to use the resulting material to create one or two new ideas out of that combined work.^[3]

Students’ initial responses were good, if a little vague, and they fell within expectations for quick brainstorming. In the following example, a student examined the strategic plan’s mental health and wellness measures. The student then worked with ChatGPT as one would with a collaborator and came up with a more detailed outline in five minutes of work (Table 1). The language provided by ChatGPT allowed the student to return to their original ideas and revise them so they were more specific. We also feel that working with AI early allowed students to more quickly see just how much potential their own ideas contained, which can help build confidence. Students also commented in the class discussion that followed the exercise that, while they likely would have come up with many of the same ideas ChatGPT did, it would have taken them much longer than five minutes to do so. Using AI to speed along the idea generating and compiling process allows students greater time to evaluate, elaborate, and build upon their ideas.

Table 1: Effective Idea Generation Activity: Student and ChatGPT Responses

Prompt II: Collaborative Confidence Building (CCB)

The CCB AI activity (Appendix B) was designed to help students better generate and more confidently share their ideas in a group setting by asking students to essentially ‘peer review’ their initial ideas with AI before sharing them with their project team. This exercise was also split into two parts. The first part began in a manner similar to the EIG exercise: by asking individual students to spend five minutes brainstorming ideas for their final project, a crisis communications project that included recorded, visual, digital, and written elements. Students were then asked to run their initial ideas through ChatGPT or Google Bard to refine those concepts and to revise and/or grow their original ideas accordingly, which they would share with their groups immediately after revision. Our hope was that this step would allow student writers the opportunity to quickly and efficiently develop ideas in which they had minimal confidence by running them through an AI ‘peer review’ before sharing. Students were subsequently placed in their groups, and each group member shared their revised ideas. Groups were also invited to repeat the process with their collaborative ideas, as well. Students were also asked to write down brief reflections on the activity, focusing on the efficacy of integrating AI into brainstorming activities.

Students liked the immediacy of ChatGPT’s augmented revision notes, which are more thorough than any student could have composed in five minutes (Table 2). ChatGPT’s expansion was witnessed and remarked upon by every participating student, who all enjoyed seeing the almost immediate expansions of their original ideas, which many saw as testaments to the usability and durability of their original concept, noting, “AI helped to add more nuance.”

Students unanimously agreed that running their own initial, sometimes vague ideas through generative AI not only gave them more material to work with, but also allowed them to see the strength in their own ideas. ChatGPT was not giving students new ideas to share with their groups, prompting several to reflect, “honestly, I don’t think my ideas were really that far off from what ChatGPT recommended,” and “my ideas mostly remain the same, but include more of the specific things that should be done when this is a real situation.” Using AI to expand upon their ideas instead provided students the necessary evidence to see the potential in their ideas, thus making the inherent value of sharing their individual ideas with their peer collaborators evident. Students also commented that they thought AI was best suited to this type of early intervention, with many including reflective statements like, “If you have no idea where to start, ChatGPT can provide a really good starting point.”

Most students also felt that generative AI was incapable of composing the final project. One student noted: “The AI helped sort of, but it didn’t really give anything complex” suggesting that successful projects rely on the expertise of individual students and the collaborative knowledge of the group. Several students indicated that using AI to quickly peer review individual ideas made the following team brainstorming meeting more focused and efficacious because potential issues had already been spotted, prompting one student to add, “AI has continued to locate blind spots in my content. Therefore, I find it quite helpful.”

Table 2: Collaborative Confidence Building Activity: Student and ChatGPT Responses

Discussion

While students generally appreciated the activity and felt more at ease sharing their ideas, there are several areas of potential concern for college writing instructors. Although generative AI can quickly augment ideas and provide additional information, that enhancement can be generic, especially in a classroom setting, where most of the prompts being used are derived from the same assignment. Students may be writing prompts that are closely related to their subject, but those prompts are still constrained by the processes and required elements of their corresponding assignment. While it could look like more content has been produced, the nature of that content is not particularly diversified. For instance, in Table 1 we see that Chat GPT proposes several ideas for improving mental wellness in the university’s strategic plan. However, most of the responses generated ultimately advocate, quite formulaically and predictably, for the promotion of “mental health awareness.” In fact, it seems that AI is responding to multiple related prompts with similar approaches and answers requiring students to weed through those responses on their own. One potential solution to this issue could be to ask students to use more than one AI for brainstorming. We would also strongly recommend that instructors lead the class through a similar exercise so students can recognize just how similar AI responses can be to different student-generated prompts for the same assignment.

Another major concern for us as writing teachers is the way generative AI standardizes language. AI, like ChatGPT, is multilingual to an extent, but has significantly less functionality in non-English languages, which is well documented both in popular and scholarly conversations (Kohnke, Moorhouse, & Zou 2023, Dave 2023). In 1974, the Conference on College Composition and Communication, the major organization in the field of writing and rhetoric, adopted a resolution on students’ “right to their own language” that remains a central guideline for practitioners in the field. Ironically, the incorporation of generative AI in the university writing classroom can pose a significant threat to this right, a fear that CCCC has recently voiced in their “Writing and AI Working Paper” (2023).

In our classrooms, we noticed that multilingual students working on the brainstorming activity often used translation tools, such as Google Translate, to generate a workable prompt for AI. In fact, prompts for AI need to be as specific as possible, as precision is directly linked with the quality of output, including the diversification of ideas that we were aiming for in the brainstorming activity. This specificity can sometimes get lost in translation in its dependence on fine nuances such as the placement and usage of small syntactic units such as prepositions, or the precise usage of field-specific jargon. Users of non-standard versions of English have to conform with these linguistic requirements if they want to generate content that is as workable as that generated by their native-speaking counterparts. Not only does this issue pose a threat to students’ original languages, but it effectively creates an unfair advantage for students who, as speakers of Standard English, already enjoy privilege in our classrooms effectively excluding other users, such as speakers of postcolonial varieties of English or African American Vernacular English. Paradoxically, although students might feel more comfortable sharing their ideas as a result of brainstorming with AI, the ideas they are sharing have been translated into another voice away from the students’ own language and vernacular, privileging one type of language usage over another. This clearly creates an ironic scenario in which the amplification of voices that we were aiming for when designing the activity, might make precisely those students voiceless whose voices were more precarious to begin with.

Recognizing this potential pitfall, we designed a reflective activity that asked students to discuss their process and analyze how Chat GPT not only helped them develop and refine their ideas but also standardized their language. We would recommend instructors assign a popular article on the subject, like Wired’s “ChatGPT Is Cutting Non-English Languages Out of the AI Revolution” for class discussion, followed by an activity where students attempt to create an AI prompt using a non-English language or Google Translate to compose a non-English prompt to better understand the difficulties and obstacles facing many digital users. A similar activity would ask students to quickly draft a thesis on a subject related to the course, then ask them to run those arguments through generative AI to see first-hand how AI-produced arguments privilege certain phrases or sentence structures, especially if generated from the same assignment prompt. Students could then write reflections on the implications of such generalization inside and outside the classroom. Whatever activity is chosen, we strongly recommend that practitioners make room for critical reflection on the process when adopting this type of activity in the classroom.

References

Aguillon, S. & Siegmund, Gregor-Fausto & Petipas, Renee & Drake, Abby & Cotner, Sehoya & Ballen, Cissy. (2020). Gender differences in student participation in an active-learning classroom. CBE Life Sciences Education, 19(12). https://doi.org/10.1187/cbe.19-03-0048

Al-Samarraei, H., & Hurmuzan, S. (2018). A review of brainstorming techniques in higher education. Thinking Skills and Creativity, 27, 78-91. https://doi.org/10.1016/j.tsc.2017.12.002

Conference on College Composition and Communication (1974). Students’ right to their own language. CCCC. 25. https://cdn.ncte.org/nctefiles/groups/cccc/newsrtol.pdf

Dave, P. (2023, May 31). ChatGPT is cutting non-English languages out of the AI revolution. Wired.

DeRosa, D. M., Smith, C. L., & Hantula, D. A. (2007). The medium matters: Mining the long-promised merit of group interaction in creative idea generation tasks in a meta-analysis of the electronic group brainstorming literature. Computers in Human Behavior, 23(3), 1549–1581. https://doi.org/10.1016/j.chb.2005.07.003

Diel, M., & Stroebe, W. (1987). Productivity loss in brainstorming groups: Toward the solution of a riddle. Journal of Personality and Social Psychology, 53(3), 497–509. https://doi.org/10.1037/0022-3514.53.3.497

Grover, S. S., Ito, T. A., & Park, B. (2017). The effects of gender composition on women’s experience in math work groups. Journal of Personality and Social Psychology, 112(6), 877–900. https://doi.org/10.1037/pspi0000090

Hargittai, E., & Walejko, G. (2008). The Participation Divide: Content creation and sharing in the digital age. Information, Communication & Society, 11, 239 – 256. https://doi.org/10.1080/13691180801946150

Isaksen, S., & Gaulin, J. (2005). A reexamination of brainstorming research: Implications for research and practice. Gifted Child Quarterly, 49(4), 315–329. https://doi.org/10.1177/001698620504900405

Kohnke, L., Moorhouse, B. L., & Zou, D. (2023). ChatGPT for language teaching and learning. RELC Journal, 0(0). https://doi.org/10.1177/00336882231162868

MLA-CCCC Joint Task Force on Writing and AI (July 2023). MLA-CCCC Joint Task Force on Writing and AI Working Paper: Overview of the issues, statement of principles, and recommendations. https://hcommons.org/app/uploads/sites/1003160/2023/07/MLA-CCCC-Joint-Task-Force-on-Writing-and-AI-Working-Paper-1.pdf

Paulus, P. B., & Murdoch, P. (1971). Anticipated evaluation and audience presence in the enhancement of dominant responses. Journal of Experimental Social Psychology, 7(3), 280–291. https://doi.org/10.1016/0022-1031(71)90028-X

Paulus, P. B., & Kenworthy, J. B. (2019). Effective brainstorming. In P. B. Paulus & B. A. Nijstad (Eds.), The Oxford handbook of group creativity and innovation (pp. 287–305). Oxford University Press. https://doi.org/10.1093/oxfordhb/9780190648077.013.17

Voss, J. (2018). Who Learns from Collaborative Digital Projects: Cultivating Critical Consciousness and Metacognition to Democratize Digital Literacy Learning. Composition Studies, 46 (1). 57-80. https://www.jstor.org/stable/26497806

Appendix A: Student Permission Language for Effective Idea Generation (EIG) and Collaborative Confidence Building (CCB) Activities

The IRB at our institution does not consider work products from a course as human subject research and does not require the submission of a protocol. Instead, according to the IRB, the use of student work in scholarly publications is regulated by FERPA. In compliance with FERPA, we sought the explicit permission of all students whose work is represented in this article using the form below.

*Please note: I will be using the material in this exercise for an article. Please include the following information:

You may use my name and my work in your research. Name as you’d like it to appear: __________________ Pronouns: __________

You may use my work but not my name:

Please do not use my work or my name:

Appendix B: Effective Idea Generation (EIG) and Collaborative Confidence Building (CCB) Activities

Effective Idea Generation (EIG)

Please do the following in the order presented

1. Brainstorm recommendations for the following three areas in our university’s strategic plan and write down at least three ideas
2. Now ask AI (ChatGPT or Google Bard) to brainstorm ideas on the same topic

Collaborative Confidence Building (CCB)

Please do the following three steps in the order presented:

1. Brainstorm one or two ideas to include in part I of the Crisis Communication project and write them down below:
2. Run those ideas through Chat GPT or Bard or something else (you must identify which one) to refine those ideas and copy below:
3. Revise your original idea(s) or create a new one based on the AI intervention and write it down below. You will share these with your group later in class:
4. Thoughts on what AI produced/helpfulness of using AI to help refine or expand ideas in this type of work

About the authors

name: Franziska Tsufim

institution: Georgia Institute of Technology

Dr. Franziska Tsufim is a Marion L. Brittain Postdoctoral Fellow at the Georgia Institute of Technology, where she teaches first-year writing courses in the Writing and Communication Program. Her work on generative AI in the college writing classroom is part of her larger research project that studies help-seeking behaviors and self-perceived agency in historically marginalized student populations.

name: Lainie Pomerleau

institution: Georgia Institute of Technology

Dr. Lainie Pomerleau is a Marion L. Brittain Postdoctoral Fellow in the Writing and Communications Program at the Georgia Institute of Technology, where she teaches composition and technical and professional writing and communication. Her current work focuses on AI’s potential and ethical impact in the college writing classroom. Additional research explores the impact of popular beliefs about history and their impact on medievalism and speculative fiction, and as well as the tension between memorialization and commemoration in pre-modern life writing.