Source-linked AI summary
Delegating Before Learning: Where Generative AI Sits in Students' Professional Communication
Jared Ren, Soobin Cho
TL;DR
Students use generative AI most for pressured professional messages, especially email to instructors and administrators. Through interviews and paired process models, the paper finds that delegation removes learning opportunities, weakens recipient-specific writing, and shifts confidence toward the system, creating risks for capacity, authenticity, and trust.
Problem
The paper asks where generative AI sits in students’ academic communication, a question left unanswered by prior evidence of varied AI use.
Method
The authors interview twelve students and model instructor-email writing with AI at the highest observed involvement level, comparing it with an unaided model based on participant accounts and classic writing-process theory.
Results
AI use was highest in institutional and professional communication, and comparison of the models identified removed learning loops, generic recipient targeting, and confidence returning to system use.
Takeaways & Limitations
Delegation may improve messages in the short run but risks preventing individual capacities from forming and making authenticity and trust additional work.
Takeaways & Limitations
The confidence feedback loop beyond participants’ directly reported reasons for AI use is an inference requiring evidence that follows students over time.
Abstract
from arXiv · showhide
We conducted an interview study with twelve students on their use of generative AI in academic communication. Students delegated professional messages to AI most where the pressure to sound professional is highest: email to instructors and administrators. AI involvement ranged from correcting the writer's own text to working out and writing the message outright, and students checked AI-written text against two criteria: whether it looks like AI and whether it sounds like them. Building on these findings, we model the AI-mediated process of writing a student--instructor email at the highest level of involvement we observed, and compare it with an unaided model of writing the same messages, built from participants' accounts and a classic model of the writing process. Three differences emerge: the learning loop that builds writing skill is removed, the message is no longer written for its specific recipient, and the confidence a successful exchange returns goes to using the system rather than to the writer's own ability. From these differences we derive two risks, that individual capacities never form and that authenticity and trust in communication become work. Design can respond to both but is unlikely to be enough, so the risks also need research and policy attention.
1 Introduction
The paper asks where generative AI sits in students’ academic communication and models that process alongside unaided writing. It contributes findings, comparative process models, and three differences involving learning, recipients, and confidence.
- 1 Introduction: The study examines generative AI in messages to instructors, advisors, peers, and administrators, whose demands differ by recipient.
- 1 Introduction: The paper models student–instructor email writing with and without AI to locate AI within the communication process.The unaided model draws on participant accounts and a classic writing-process model.
- 1 Introduction: The paper contributes research findings, two process models, and three differences concerning learning, the recipient, and confidence.
- 1 Introduction: It concludes by discussing two risks, possible design responses, and the need for research and policy attention.
2 Where and How Students Use Generative AI
Twelve students reported using AI most for institutional and professional communication, with involvement ranging from correction to deliberation. They evaluated generated text for both machine-like appearance and personal voice.
- 2.1 AI use concentrates on instructors: 62 of 80 institutional or professional recipient–channel groups (77.5%) reported AI use, compared with 6 of 30 peer groups (20%).
- 2.1 AI use concentrates on instructors: Students cited channel professionalism, recipient expectations, and high-stakes messages as reasons for using AI, alongside doubt about their writing ability.
- 2.2 Roles for AI in messages: AI involvement ranged from correcting spelling and grammar to deliberating, composing, advising, and writing messages outright.As autonomy shifts toward the model, writers can become reviewers of system-authored text.
- 2.3 Two checks on AI-written text: Students checked AI-generated text for whether it looked like AI and whether it sounded like them.The first check concerns perceived origin; the second concerns the writer’s voice.
3 Modeling AI-Mediated Student–Instructor Email
The paper models the highest observed level of AI involvement in professional academic email, where the system writes and the student revises. Prompting absorbs planning and framing, while editing remains optional and the exchange returns confidence in system use.
- 3.1 Scope and assumption: The model represents email from students to instructors and teaching staff, a setting treated as representative of professional academic email.The paper assumes the deliberator end of observed AI involvement because the contrast with unaided writing is clearest there.
- 3.2 The AI-mediated model: The prompt absorbs deciding what to ask for and how to frame it, potentially handing over that work even in a short request.
- 3.2 The AI-mediated model: Students check generated output against origin and voice criteria, then may edit selectively or send it unchanged.No participant reworked the generated message as a whole.
- 3.2 The AI-mediated model: After the message is sent and answered, the outcome returns to the writer as confidence in using the system.
4 An Unaided Model for Comparison
The unaided model represents writing as an iterative process of discovering a need, composing, checking, and seeking information when checks fail. Repeated loops improve both the message and the writer’s communication confidence.
- 4 An Unaided Model for Comparison: The unaided model combines participant accounts with a cognitive writing model in which planning, drafting, and reviewing loop back on one another.
- 4 An Unaided Model for Comparison: Writers check whether the message communicates professionally, fits the relationship, and uses a professional format.
- 4 An Unaided Model for Comparison: When a check fails, writers seek examples, templates, peer messages, or saved emails, then return with new information.
- 4 An Unaided Model for Comparison: The message is sent after the loop turns until it passes, and the reply returns as a small gain in confidence in the writer’s communication.
5 Three Differences
Comparing AI-mediated and unaided email writing reveals three differences: delegation removes learning opportunities, genericizes the recipient, and redirects confidence from the writer to the system.
- Learning drops out: A failed check in unaided writing sends the writer to find an example, while AI-mediated writing turns the failure into deletion or replacement without learning what would have fitted.The unaided loop yields both a better message and a slightly better writer; the AI-mediated process does not convert failure into genre learning.
- Learning drops out: AI delegation also removes planning and some checks, leaving only narrow surface judgments about whether text sounds right for the writer and reader.The prompt absorbs planning, while writers may stop checking errors the system reliably avoids.
- The recipient drops out: The generated message is written for a generic professor rather than the actual recipient’s history, closeness, and prior exchanges.Editing can trim sentences that do not fit, but relationship-based construction must happen from the start.
- Confidence goes to the tool: In unaided writing, a successful exchange builds confidence in the writer’s ability; in AI-mediated writing, it builds confidence in using the system.The contrast follows from what each process returns to the writer after the exchange.
- Confidence goes to the tool: Because confidence grows most strongly through succeeding at a task oneself, delegated success can improve message production without increasing the writer’s ability to produce one independently.Prior work cited in the paper reports more confidence in writing among students using scaffolding rather than generation.
6 Risks and Responses
The model comparison identifies two risks of AI-mediated student communication: writing capacities may never form, and authenticity and trust become additional work. Design can address these risks, but their costs also require research, policy, and curriculum attention.
- Risks: AI-mediated writing removes practice in planning, information seeking, drafting, and critical judgment, risking that individual capacities never form.The authors note that low confidence can prompt delegation, while the resulting loss of practice may reinforce future delegation; the longer-term cycle remains untested.
- Risks: AI-mediated communication makes authenticity and trust work by requiring writers to manage perceived sincerity, machine origin, and recipient-specific relationship cues.Generated messages target a generic professor, so emotional connection survives only when editing restores relevant history and closeness.
- Responses: Scaffolding can restore learning by treating deletions as signals, eliciting the recipient relationship before drafting, and supporting writers rather than generating messages outright.These responses correspond to the missing outward learning step and the loss of recipient-specific message construction.
- Responses: Design alone may not suffice because teaching-oriented tools cost more at the moment of use, while students choose tools message by message under pressure.Stakes, email permanence, and doubts about writing ability all favor ready-to-send messages each time a student faces a blank email.
- Broader implications: The risks need research and policy attention because their costs emerge only in aggregate and over time, including questions about sustained unaided ability and curriculum.The authors propose studying whether heavy delegators can recognize effective professional emails but become less able to produce them, while professional register becomes a curriculum question.