Writing can play an important role in STEM courses by providing opportunities for students to learn the conventions of writing in a discipline and by enhancing student learning and engagement. The way that students engage with writing, however, is changing rapidly as students now have easy access to a range of chatbots and other GenAI tools. 

Chatbots and GenAI pose unique challenges, as well as possible opportunities, for writing in STEM courses. The tools can hinder or enhance learning. To address both the risks and opportunities of student use of chatbots, this guide considers how writing assignments might be designed to reduce students’ uncritical use of chatbots and, alternatively, how to engage students with GenAI to enhance their learning, writing process, and writing products. 

Writing Prompts and GenAI

Writing is a recursive process that occurs within complex social networks. Writing is important for student learning, and studies have demonstrated that the more students write, the more they learn and the more they grow as individuals. The role of writing in the learning process is being questioned, however, as GenAI and associated chatbots are increasingly able to respond well to a range of writing assignment prompts. This can both help student learning when used constructively–and hurt student learning when students rely too much on a chatbot’s output, and avoid doing the conceptual work that comes with writing. 

As GenAI models become more sophisticated, they will likely be able to generate better responses to writing prompts. In addition, students are becoming increasingly adept at using GenAI to assist them in their writing assignments (see Owen Terry’s piece in the Chronicle of Higher Education). The widespread availability and use of chatbots by students, provides instructors with an incentive to revisit the goals and approaches to writing assignments. 

The role that writing assignments play in a course depends on the learning goals for the course as well as how assignments advance these goals. Prior to modifying assignments because of chatbot availability, it can help to revisit course goals and consider if they need to be updated. Instructors can then decide whether or not it is appropriate for students to use chatbots in their learning process–whether it be for exploring or understanding concepts and content better or for assisting in the writing process. In some cases, it might help to refine writing prompts to dissuade students from relying solely on chatbots to generate responses; in other cases, it might make sense to help students learn how to use chatbots as part of their writing process. 

Regardless of the approach, students will need clear guidance from instructors about course policies related to chatbot use as well as open discussions about how these policies align with the instructor’s pedagogical philosophy. U-M provides ideas for syllabus language in its Handbook for Faculty and Instructional Staff (see page 11) and the Sweetland Center for Writing has provided some ideas about syllabus language along with examples from U-M instructors. 

Ethical Considerations for Chatbot Use

There are several important considerations for the ethical and appropriate use of chatbots and GenAI. These ethical issues are covered in detail elsewhere - such as the report from U-M’s GenAI committee as well as MLA-CCC. Users of chatbots should know that their output is sometimes inaccurate as content, information and sources can be fabricated and/or falsified (Elali and Rachid, 2023). The training of the LLMs that run chatbots make it impossible to verify the source (and therefore accuracy) of the information so the content needs to be fact-checked. The use of chatbots can also present concerns related to linguistic justice as the training sets used for GenAI platforms are based largely on the language of Standard (White Mainstream) English. Chatbots unnecessarily “correct” some dialects of English, which potentially compromises writers’ use of all their language resources.

If students are allowed to use chatbots or other GenAI programs for writing assignments, they should learn both how to use the technology in a responsible and appropriate way as well as how to remain in control of the process and output. Chatbots generate content through algorithms based on word probabilities from training datasets. Students who use chatbots are responsible for the veracity, accuracy, and quality of chatbot content. In addition to the need to verify the accuracy of the material generated by chatbots, students should also be taught ways to maintain their own voice and agency throughout the writing process, to ensure that they remain fully in control of the ideas and writing style they produce in their assignments. Some ideas for how to do this are presented below (see Writing with Chatbots section). 

Writing without Chatbots

This section provides suggestions for revising writing assignments that might help reduce the reliance of students on chatbot responses. Any revisions in assignments should align with main learning goals for students.

  • Focus on process over product: One way to reduce students’ overreliance on chatbots is to focus assignments on a writing process that involves multiple stages of invention, composition, and revision. Since chatbots can often produce text that sounds good, consider de-emphasizing the end product of writing assignments and placing more emphasis on the writing process instead. This emphasis on the writing process can involve scaffolded assignments that ask students to show how their thinking and writing on their topic is evolving by sharing lab notes, objectives and hypothesis statements, primary sources, outlines, drafts, peer feedback, revisions, and reflections on the research and writing process. Students who engage in a scaffolded writing process show benefits in their learning as well as their personal development (see Kramer et al. 2019). 

    • Ideas: Emphasize writing process by allocating more time and points to scaffolded assignments. For example, have students share their research questions or objectives, their sources, their outlines, and several drafts along with their revision strategies. To understand how students experienced their writing process, consider having students submit a writer’s letter with their final submission that explains their writing process and their thoughts on their final piece and/or incorporate metacognition and reflection throughout the writing process (see next item). 

  • Emphasize metacognition: Metacognition and reflection are important ways to help students grow and develop as writers. Reflective metacognitive tasks can engage students in examining their own writing process and can reduce their incentives for using chatbots. Consider adding these elements to existing assignments and assign adequate points to reflect the cognitive importance of this work. 

    • Ideas: Consider adding reflections and writer’s notes at different points in the writing process. One way to integrate reflection into assignments is to have students annotate their writing to share their thoughts about the content and format of their piece. For example, students can add comments to their drafts that share their thoughts or questions about a particular section of text or that explain what changes they made or plan to make to their piece to improve it. Students can also use these annotations to ask for feedback from their peers or instructors on particular parts of their work. Another way to integrate reflection into assignments is to have students submit short reflections with the different stages of their writing in which they explain what changes they made to their piece and what they thought went well or where they struggled with their process or their piece. Finally, consider using writer’s notes or writer’s letters to go along with revised writing, and ask students to reflect on their experience in writing a particular piece and how they approached revising and improving their work in response to the feedback they received. 

  • Create assignments that ask students to engage in discipline-specific reasoning and higher-order thinking: The current generation of chatbots (e.g., GPT-4) appears to not be adept at engaging in discipline-specific reasoning. This limitation can manifest as a lack of details or specifics when chatbots are asked to respond to discipline-specific content and is due to the ways that the LLMs underlying the chatbots have been trained (on copious amounts of more general information). Helping students recognize and develop how disciplinary subject-matter knowledge is expressed can help them participate in the discourse community. Their discursive competence will enable them to use chatbots advisably and with a recognition of technological limitations.

  • Ask students to engage rigorously with sources, especially journal articles: Most chatbots, including ChatGPT, are not well equipped to work from sources in a rigorous way and are not reliable in their use of sources and in-text citations. This weakness often manifests as fabricated or erroneous sources, especially for peer-reviewed articles (McGowan et al. 2023). Consider structuring assignments around rigorous use of sources and appropriate citation practices to disincentive students from relying too much on chatbots for their responses. The goal in making these changes is to disincentivize students from overly relying on chatbots to generate their responses, while also engaging students in careful reading, analysis, and synthesis from sources.

    • Ideas: Revise writing assignments to require students to integrate in-text citations into their assignment and to have accurate references listed in their reference list. Alternatively, have students read and work from pre-assigned (or pre-approved) sources and use these for the basis of their writing assignments. When evaluating student work, give importance and weight to this rigorous use of sources and ask for in-text citations of specific information and facts.

  • Create opportunities for primary research: Consider developing opportunities for students to conduct and then write about their own research. Providing primary research opportunities for students can help disincentivize reliance on chatbots. These primary research assignments can include written reports of lab work or mini-research projects based on research conducted over the course of the semester. Such efforts will need scaffolding, but can also result in students’ engaging in their research and writing processes. For examples of how primary research might be integrated into a course or departmental curriculum, see Morrison et al. 2020. 

  • Lean on class-specific content and context: Create assignments that rely on class-specific discussion and content. Chatbots are trained on large textual data sets and have no access to class discussion or specific lecture information; they aren’t able to generate responses to class-specific discussions or localized contexts. Ideas: When possible, tie writing assignments to lectures or class discussions, design work, or lab work completed by students in the course. This material will be difficult if not impossible for GenAI to reproduce or reference without substantial effort from the student writer. 

Writing with Chatbots: Using Chatbots to Support Student Writing

There are several ways in which students can benefit by using GenAI and chatbots in their writing process. Part of this learning process will need to include how to write prompts for chatbots that elicit the most useful and appropriate content from this technology - see Kansas University Prompting of AI Chatbots for ideas on how to refine chatbot prompts. 

Writing Process

  • Help with brainstorming and structure: Chatbots can be used to help engage students with brainstorming about their topics. In this way, chatbots can be good sounding boards or idea generators for students before they start writing, even for STEM discipline-specific pieces. Chatbots can also help students experiment with outlines to get ideas for structuring their pieces. When used in the earliest stages of writing, chatbots can help students get started on their work. 

    • Ideas: Have students input sections of notes and ask a chatbot to generate an outline of possible content. Continue to coach the chatbot with ideas and revisions until the outline is a good structure for the writing assignment. 

  • Experiment with ideas and content: Some writers find that chatbots can be helpful when used in a dynamic manner, such as when a writer comes with some content and ideas about a writing piece and then uses the chatbot to probe for additional directions or other considerations. This approach can be helpful as the writer first comes with a direction and purpose and then uses the chatbot as a type of research assistant to explore directions and possibilities to their piece. Students can also use chatbots in this way to better understand a particular topic for their writing and explore general information on a subject. 

    • Ideas: Have students brainstorm topics, ideas, and content in class or on their own, and then think about what they might still need to understand or explore about their topic. Have students then use a chatbot to find out more information that might help them expand, refine, or narrow their topic and associated content. 

  • Help with Revision and Editing: Chatbots can also be used to generate suggestions for improving existing text and providing suggestions for correcting grammatical errors. Consider helping students find ways to use chatbots to identify ways they might improve their text, while still remaining in control of how they choose to edit and refine their work (and to avoid allowing chatbots to make these changes for them).

    • Ideas: Consider having students use chatbots to create a reverse outline for an existing draft or section of a text to help students see how their piece is currently structured and to help think about ways to improve the structure of their work. Many students (and writers) find that reverse outlines are useful for clarifying the key point of each paragraph and for checking for redundancy and flow in a piece. Students can also share their text with a chatbot and ask for suggestions for ways to improve their prose by identifying and explaining possible grammatical mistakes.

Writing Products

  • Summarize from Specific Sources: Some chatbots can be directed to summarize key content, ideas, and themes from specific sources and articles (for example, see Humata.AI). Students can use this feature to explore if their reading and understanding of a research article aligns with a chatbot summary of the piece. They can critically evaluate the chatbot summary for accuracy and emphasis of main ideas. Students can also use this feature to help them better understand the key findings of an article, especially if it is written for an expert audience and/or uses a large amount of technical or difficult text. When used in this way, chatbots can help students gain a better and deeper understanding of discipline-specific material. 

  • Write in a Specific Genre and for a Specific Audience: Chatbots can be especially useful when students need to explore ways to write in a different genre and/or for a different audience. If students aren’t familiar with a particular genre, they can use chatbots to generate examples of the genre as a model. Chatbots are helpful for writing an abstract or a summary of a piece, as they can help the writers see how an algorithm is interpreting the key themes in their piece. Chatbots can also be helpful for exploring writing for different audiences, especially in terms of making content and text less technical and/or more engaging for a broader or more general audience. 

Other ways chatbots might be useful in a writing classroom:

  • Use Chatbots to Generate Examples to Critique: Chatbots can be helpful in generating example responses that can serve as a sample text to discuss and critique. This can be especially helpful with STEM content as students often benefit by seeing examples of what they are asked to generate and can then critique the material to evaluate accuracy and quality of the response. 

  • Compare Chatbot Responses with Student Papers: To explore the strengths and weaknesses of chatbot output, consider having students compare their own writing and analysis on a topic with the responses produced by chatbots. Have the students evaluate the facts, evidence, and logic of their own writing and the chatbot text.

The citation conventions for how to acknowledge the help and use of GenAI are changing rapidly. Sweetland has summarized a few of the conventions outlined by MLA, APA, and The Chicago Manual of Style in our Citation Conventions for GenAI and Chatbots document. If students are allowed to use chatbots, there are programs that can capture and record the prompt and associated text for easy reference -  A.I. Archives and ShareGPT (a Chrome extension).

Summary

The widespread availability of chatbots will likely change how we approach our own writing as well as how we teach students to write. These changes present opportunities to re-envision how we use writing assignments in our curriculum. We can require  students to learn how to use the technology in ways that facilitate conceptual, disciplinary learning and writing development. The GenAI landscape is changing quickly, and the material in this guide will continue to be revisited and updated as appropriate. For more individualized support in designing or revising writing assignments with chatbots in mind, please reach out to sweetlandinfo@umich.edu.

References

Elali, F. R., and Rachid, L. N. (2023). AI-generated research paper fabrication and plagiarism in the scientific community. Patterns 4(3). https://doi.org/10.1016/j.patter.2023.100706

Kramer, T. J., Zeccardi, J., Concepcion, R., Emhoff, C. A., Miller, S., and Posell, K. V. (2019). WID course enhancements in STEM: The impact of adding ‘writing circles’ and writing process pedagogy. Across the Disciplines 16(4), 26.

McGowan, Alessia, Yunlai Gui, Matthew Dobbs, Sophia Shuster, Matthew Cotter, Alexandria Selloni, Marianne Goodman, Agrima Srivastava, Guillermo A. Cecchi, and Cheryl M. Corcoran. (2023) ChatGPT and Bard exhibit spontaneous citation fabrication during psychiatry literature search. Psychiatry Research 326: 115334.

Morrison, M. E., Lom, B., Buffalari, D., Chase, L., Fernandes, J. J., McMurray, M. S., and Stavnezer, A. J. (2020). Integrating research into the undergraduate curriculum: 2. Scaffolding research skills and transitioning toward independent research. Journal of Undergraduate Neuroscience Education 19(1), A64.