DC Elective 2: Ethical Issues in Digital Communication

Digital Communication Certificate · Elective 2

Ethical Issues in Digital Communication

How to apply a structural AI literacy framework to ethical questions in digital communication — using AI-based academic integrity tools as the anchor governance case — and how to make principled operative decisions about digital tools in your own instructional context. Estimated time: 3–4 hours.

Section 1

The AI Literacy Framework as Ethical Procedure

Ethical questions about digital tools are not primarily questions of individual intent. A faculty member who deploys a plagiarism detection system with the intention of promoting academic integrity is not thereby relieved of responsibility for the system's distributional effects. Ethics in digital communication requires attention to what systems do — whose interests they serve, whose they harm, and how they distribute risk — not only to what users intend.

The AI literacy framework (Cole, 2026) offers a three-level procedure for exactly this kind of structural ethical analysis. The levels are sequential and prerequisite: conceptual literacy must precede structural analysis, and structural analysis must precede operative decision-making. Skipping to the operative level — asking "should we use this tool?" before asking "what does this tool do and whose interests does it reproduce?" — produces decisions that mistake good intentions for adequate ethical analysis.

Conceptual literacy establishes accurate naming and description. Before asking whether a tool is ethical, you need to know what kind of system it is, what its training data and objective function are, and what it claims to detect or produce. Many institutional deployments of AI tools rest on conceptual misunderstandings — treating probabilistic systems as definitive, treating pattern-matching as semantic comprehension, treating training-data outputs as ground truth.

What this elective is not

This is not a module about individual-level digital ethics — privacy settings, password hygiene, or citation of AI-generated content. Those are important but belong elsewhere. The ethical questions this elective develops are structural: what systems are deployed, by whose authority, to whose benefit and whose detriment, and on what grounds.

Section 2

Structural Analysis: Three Questions

The structural level of the AI literacy framework applies three questions to any digital system under consideration. These questions are designed to surface what the system reproduces, not only what it claims to do.

Question 1: What hierarchies does this system reproduce? Every AI system is trained on data that encodes existing social arrangements — linguistic norms, institutional practices, patterns of who gets heard and who does not. A natural language processing system trained primarily on Standard American English academic prose will systematically underperform on texts written in other linguistic registers, dialects, or by writers whose first language is not English. That is not a bug to be patched; it is an architectural consequence of the training data, and it reproduces the hierarchies encoded in that data.

Question 2: Who bears the costs when the system fails? Failure costs in AI systems are rarely distributed equally. When an AI content moderation system produces false positives, the cost falls on the person whose content was wrongly flagged — not on the developer, the institution, or the majority of users for whom the system performs well. Understanding who bears failure costs is an ethical question, not a technical one: it requires knowing whose position makes them most vulnerable to the system's errors.

Question 3: Whose interests does the design serve? AI systems are not designed by neutral parties. They are designed to meet the needs of those who fund and commission them. An academic integrity system is designed to meet the needs of institutions seeking to detect misconduct — not the needs of the students it processes. That asymmetry does not make the system illegitimate, but it does mean that claims about what the system is "for" must be evaluated against whose interests shaped its design.

Section 3

Operative Decisions: Three Paths

After structural analysis, the operative level requires a decision. The framework identifies three legitimate outcomes, each with its own conditions.

Principled Refusal

The structural analysis reveals harms that cannot be mitigated within the constraints of the deployment context. The decision not to deploy is itself an ethical act — grounded in analysis, not avoidance. Refusal requires stated grounds and, in institutional settings, documentation.

Conditional Adoption

Adoption proceeds under explicitly specified conditions: what additional checks are required, what populations require additional protection, what data will be monitored for distributional harm, and what threshold of evidence would trigger withdrawal. Conditions are written down and binding, not aspirational.

Deferred Decision

The structural analysis is incomplete — evidence about failure rates for specific populations is unavailable, or the technology is changing faster than the evidence base. Deferral is not indecision; it requires specifying what evidence would resolve the deferral and setting a timeline for revisiting the decision.

The framework insists that all three are legitimate. The ethical failure is not choosing any particular path — it is proceeding without analysis, or treating adoption as the default requiring no justification while refusal requires justification.

Section 4

Governance Case: AI-Based Academic Integrity Detection

AI-based academic integrity tools — systems that claim to detect AI-generated text, plagiarism, or contract cheating — are among the most rapidly adopted AI tools in higher education and among the least structurally analyzed before deployment. This case applies the three levels of the AI literacy framework to them as a governance problem.

Conceptual literacy. These systems are probabilistic classifiers, not semantic comprehension systems. They assign a probability score to a text based on its statistical similarity to distributions they have learned to associate with AI-generated or copied text. They do not "know" whether a given student wrote a given text. A score of 85% does not mean there is an 85% chance the text was AI-generated; it means the text's statistical features resemble the training distribution at a threshold the vendor set.

Structural analysis. The hierarchies these systems reproduce are well-documented in the research literature. False-positive rates — classifications of human-written text as AI-generated — are significantly higher for multilingual writers, for writers from non-dominant linguistic backgrounds, and for writers whose register differs from the training distribution. The institution that deploys these systems and accepts their output as evidence in academic integrity proceedings concentrates failure costs on already-marginalized student populations. The interests the design serves are institutional — reducing faculty workload in evaluating academic integrity claims — not the interests of students, who bear the costs of false accusations.

The distributional evidence

Liang et al. (2023) found that Turnitin's AI detection system flagged essays written by non-native English speakers as AI-generated at a false positive rate of up to 61.3%, compared to near-zero rates for native English speakers writing in dominant academic registers. This is not an edge case. It is the system performing as its training data would predict.

Operative implications. An institution that deploys AI detection tools without accounting for these distributional failure rates is not making a neutral pedagogical decision. It is making a governance decision that systematically disadvantages multilingual students and students from non-dominant linguistic backgrounds in academic integrity proceedings. The three-decision framework asks: given this structural analysis, what decision is defensible?

Activity 1

Structural Analysis of a Digital Tool

Individual analysis · 45–60 minutes

Select one digital tool currently deployed or under consideration in your teaching or institutional context. This might be a learning management system's analytics function, an accessibility checker, a plagiarism detection tool, a content moderation system, or any AI-assisted communication or assessment tool. Apply the three questions of structural analysis from the AI literacy framework (Cole, 2026).

Identify and describe the tool. (Name, function, who deployed it, at what scale, and what it claims to do. Prioritize conceptual accuracy — what kind of system is this, technically?) Question 1: What hierarchies does this system reproduce? (Consider the training data or design choices. Who or what defined the "normal" against which this system measures? Whose linguistic, cultural, or behavioral patterns are centered?) Question 2: Who bears the costs when this system fails? (Identify the populations most exposed to false positives, misclassifications, or system errors. What is at stake for those populations — grades, access, reputation? Who is not exposed to these risks?) Question 3: Whose interests does the design serve? (Who commissioned this system, and what problem were they trying to solve? Whose interests shaped the design priorities, and whose interests were secondary?) Structural synthesis: What does this analysis reveal about the system's ethical stakes? (This is not yet an operative decision — it is a statement of what you now understand about the system that you did not understand before applying the three questions.)

Contribute to the repository

Structural analyses of tools deployed in specific disciplinary contexts — especially where failure cost distributions are not well-documented — build the evidence base that institutions need for governance decisions.

I am willing to share this structural analysis (with my name and discipline) in the C²TC teaching repository.

Activity 2

Governance Case and Operative Decision

Two parts · 60–75 minutes

This activity has two parts. Part A asks you to apply the operative decision framework to the AI detection governance case from Section 4. Part B asks you to make and justify an operative decision for the tool you analyzed in Activity 1.

Part A

The AI Detection Case: Your Operative Decision

The structural analysis of AI-based academic integrity detection (Section 4) established: probabilistic classifiers with known false-positive rate disparities concentrated among multilingual students; failure costs borne by already-marginalized student populations; design interests oriented toward institutional workload reduction rather than student protection.

Given this structural analysis, what operative decision would you recommend for your institution or department? Select a decision path and develop the full grounds for it.

Decision path. (Principled refusal, conditional adoption, or deferred decision — and the primary reason you identify this as the defensible path.) Full grounds for your decision. (The structural analysis points directly to what justifies this path. Be specific: which features of the structural analysis — hierarchies reproduced, failure cost distribution, design interests — ground the decision? A good grounds statement makes your reasoning followable even by someone who would choose a different path.) Conditions or contingencies. (If you chose conditional adoption: what conditions are binding and how would you monitor them? If you chose principled refusal: what institutional or evidentiary change would make adoption defensible? If you chose deferral: what evidence would resolve the deferral and what is your timeline?)

Part B

Your Tool: Operative Decision

Return to the tool you analyzed in Activity 1. Using the structural synthesis you developed, make an operative decision — principled refusal, conditional adoption, or deferred decision — and write the full justification as you would present it to a department or institutional governance body.

Your operative decision for the tool from Activity 1. (State the decision path and the primary structural grounds.) Full justification. (Write this as a governance statement — precise, grounded in the structural analysis, addressed to an audience that has authority over deployment decisions. What would need to be documented, monitored, or changed for this decision to be defensible?) Assignment design: How would you develop students' capacity for this kind of structural ethical analysis? (Design one assignment or course activity that teaches the three structural questions as an analytical procedure, not as a values statement. What would students analyze, and what evidence of structural analysis would you look for in their work?)

Contribute to the repository

Governance statements and assignment designs for teaching structural AI ethics — especially those grounded in specific disciplinary contexts — are among the most practically useful resources for the C²TC community.

I am willing to share my governance statement and assignment design (with my name and discipline) in the C²TC teaching repository.

Your Artifact

Build Your Structural Ethics Assignment

Design an assignment that teaches students to apply the three structural questions as an analytical procedure. Complete the fields below to generate a formatted, print-ready assignment sheet grounded in your governance work from Activity 2.

Assignment title (Name the analytical task specifically — what will students analyze and what procedure will they use?) Course and audience (Course number and title, and the student audience.) Assignment overview (2–4 sentences for students. Frame structural analysis as an analytical procedure, not a values exercise.) Task description (What tool will students analyze, how will they apply the three questions, and what will they produce — a written analysis, a governance memo, a presentation?) Operative decision component (Will students make an operative decision — principled refusal, conditional adoption, or deferral? What format will that decision take — a governance memo, a recommendation, a policy draft?) Assessment criteria (3–5 criteria that distinguish excellent structural analysis from generic critique. Precision, specificity, and grounding in evidence are hallmarks.) AI use policy (State your policy and the pedagogical reasoning.)

Reading List

Selected Sources

Cole, K. (2026). AI literacy for communication instruction. CWSP/C²TC. cwspwolf.com/ai_literacy_unified.html
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7).
Noble, S. U. (2018). Algorithms of oppression: How search engines reinforce racism. NYU Press.
Eubanks, V. (2018). Automating inequality: How high-tech tools profile, police, and punish the poor. St. Martin's Press.
Broussard, M. (2018). Artificial unintelligence: How computers misunderstand the world. MIT Press.
Crawford, K. (2021). Atlas of AI: Power, politics, and the planetary costs of artificial intelligence. Yale University Press.
Benjamin, R. (2019). Race after technology: Abolitionist tools for the new Jim Code. Polity Press.
Andrejevic, M., & Gates, K. (2014). Big data surveillance: Introduction. Surveillance & Society, 12(2), 185–196.

Complete Elective 2

When you have finished both activities, submit your responses. Your work is auto-saved in your browser.

Submitted. Your Elective 2 responses are received.

Submission problem — responses saved locally. Try again or email kkcole2@ncsu.edu.

Return to

← Digital Communication Certificate