Beyond Surveillance and Capitulation: Compassionate Learning Design and Critical AI Literacy in Open and Distance Learning

A/Prof Daniela Gachago, Centre for Innovation in Learning and Teaching, CHED, UCT

daniela.gachago@uct.ac.za

Caveat

It’s all moving very fast…

Change is inevitable…

Don’t get too wedded to anything…

Overview

Where do we stand in relation to genAI

UCT approach

Assessment Redesign and Critical AI literacy focus

My own response

ODL unique concerns and possible approaches

​

3

4

Where do we stand in relation to genAI?

6

Where do you stand in relation to AI?

UNISA context - what are you doing at the moment? What works? What are pain points?

7

My perspective today...

• Academic staff developer in the Centre for Innovation in Learning and Teaching promoting and supporting genAI in T, L & Research
• Teaching PG programmes through the School of Education (online)
• PGDip in EdTech, Blended and Online Learning Design, Med in EdTech and Higher Education Studies
• Faculty of Humanities MED and PHD Supervisor
• External examiner
• Research ethics committee member
• Managing Editor of CriSTaL (Critical Studies in Teaching and Learning in Higher Education)

8

Who are our students

9

• PGBOLD, PGdip EdTech, MED EdTech, MED HES

​

• Online learners, but small classes (max 30)

​

• Different context: HE, schools, NGOs, EdTech companies, corporate world, government

​

• Adult learners, juggling multiple responsibilities

​

• Highly different academic and digital literacy skills

How does AI help our students?

• Thinking partner
• Brainstorming
• On demand positive feedback
• Unpacking difficult concepts / theories
• Establishing gaps in the literature
• English language support
• Disciplinary language support

​

10

11

https://padlet.com/uct/2026_05_12-exploring-practical-ways-to-support-ai-use-in-pos-opevcqjyzf98h4j0 thank you to Shanali Govender!

12

What are the risks?

• postgraduate work is fundamentally about developing ones intellectual judgment
• risk of jumping important steps of grappling with writing – sitting with uncertainty and discomfort
• “do I belong here, does my thinking count? “

​

​

• The danger isn't plagiarism. It's something quieter: students who never find out what they actually think or how their voice sounds

​

13

Some traps I have seen…

• Hallucination as a structural problem
• The decontextual trap
• The confidence trap
• Outsourcing the struggle
• Dependency without capability

​

Additions by Prof Chaka:

• Intellectual, cultural, affective, ethical … offloading

14

With a few exceptions, the scientific and scholarly community has reached a broad consensus that

1. AI tools can be used in research and to assist scholars in specific tasks, such as copyediting and data analysis;
2. AI tools cannot be named as authors because they cannot be held morally or legally responsible for their work; and
3. the use of AI tools in research should be fully disclosed to promote transparency (COPE 2024; ICMJE 2024; Kaebnick et al. 2023).

Resnik, D. B., & Hosseini, M. (2026). Disclosing artificial intelligence use in scientific research and publication: When should disclosure be mandatory, optional, or unnecessary? Accountability in Research, 33(2). https://doi.org/10.1080/08989621.2025.2481949 

15

https://www.dailymaverick.co.za/opinionista/2026-05-28-ai-is-eating-the-academic-publishing-industry-alive-but-some-good-might-come-of-it/

The UCT context

UCT - academic misconduct policy

Cheating is the practice of attempting to gain an unfair advantage. This includes accessing prohibited materials in an examination, making use of ‘essay mills’, language models such as chatbots (eg. ChatGPT and other large-language models or generative Artificial Intelligence), and any service or software that provides answers to assessments, or writes or re-writes assignments or parts thereof, other than software that detects and corrects spelling and grammatical errors. The only permissible instances of such practices are where they are explicitly permitted by the terms of the assessment instructions. In the case of group work, cheating includes students indicating that they have participated in group work when in fact they have not.

17

High Error Rates: AI detection tools often have high error rates, leading to false positives and false accusations of misconduct.​

Lack of Reliability: No AI detection tool can reliably distinguish between AI-generated and human-generated content.​

Easily Defeated: AI-generated text can be easily rephrased to evade detection.​

Inconsistent Performance: Different AI detectors yield varying results, making them unreliable.​

OpenAI’s Admission: Even OpenAI, the creator of ChatGPT, has acknowledged that their AI detection tools are not effective and have discontinued their AI Classifier due to poor accuracy.​

​

18

19

20

21

https://cilt.uct.ac.za/sites/default/files/media/documents/cilt_uct_ac_za/2486/uct-ai-in-education-framework-june-2025-final.pdf

22

The traffic light system

23

Discursive vs Structural Changes

“We find that educational frameworks frequently borrow the language of socially familiar structural systems (like vehicular traffic lights) while lacking their actual enforcement capabilities, creating an illusion of assessment security. In place of this, we argue for a shift towards structural assessment redesign that builds validity into assessment architecture rather than attempting to impose it through unenforceable rules.”

24

Structural changes to assessment

“Modifications that directly alter the nature, format, or mechanics of how a task must be completed, such that the success of these changes is not reliant on the student’s understanding, interpretation, or compliance with instructions. Instead, these changes reshape the underlying framework of the task, constraining or opening the student’s approach in ways that are built into the assessment itself.”

25

Focus: redesigning assessment

What are my learning outcomes?

​

What do students need to do on their own?

​

Where could genAI be useful?

26

Determining your position

McCorkle, S. (2025). Designing an Aligned Generative AI Course Policy: An Equitable and Transparent Learner-Centered Approach. International Journal of Designs for Learning, 16(2), 96-110.

27

Institutional policies or guidelines

Faculty/ department policies or guidelines

Course AI policy or position statement

Task AI requirements and guidelines

• What are the steps for the Individual Assessment?
• What might we be learning at this step?
• Is this a core learning outcome?
• How could we use AI to support this outcome?
• What would be gained?  What would be lost? If we did use AI? 

The importance of critical AI literacy

• Somewhere between techno-pessimism and techno-optimism is the position this piece is arguing for. Not refusal. Not uncritical adoption. Literacy. Sovereignty. The capacity to engage deliberately with a medium you are already inside. To understand its grain, its tendencies, what it does to you when you engage without awareness. Rather than being used by it in either direction: enchanted into dependency or shamed into secret use.�Individual abstention inside a society already saturated with AI infrastructure functions more symbolically than structurally. The medium is infrastructural now. People are already inside AI-mediated systems whether they consciously use AI or not. The struggle shifts from avoid all contact to preserve agency under conditions of contact.�That is why literacy is resistance

29

Maha Bali

As a human, I am critical of the colonialist, capitalist, racist, Eugenicist roots of AI and the way it is overhyped and also has been historically used to harm humans and exacerbate inequality under the illusion of neutral automation. However, I have also believed for a long time that I cannot control or even attempt to control how students use AI. My overall approach is to spend a lot of time nurturing their critical AI literacy so they’re aware of the ethical considerations, hallucinations, biases, and roots of who and which ideologies lie behind AI tools, show them some alternative tools that hallucinate less.

30

31

UCT adapted AI literacy framework

32

https://cilt.uct.ac.za/sites/default/files/media/documents/cilt_uct_ac_za/2486/uct-ai-in-education-framework-june-2025-final.pdf

A value-based approach to critical AI literacy (Syska, 2024, p. 17)

​

GenAI values-based literacy will help us decide what learning and writing mean to us so that we can activate our values and make the best decisions about both its role in this process and its impact on our identity as learners and writers

​

33

Gachago, Bali & Pallitt (2022): Compassionate learning design. In Jhangiani, S., J. Quinn & M.Burtis (Eds.). Toward a Critical Instructional Design. PressBooks.

PARTICIPATION:

The desire to increase level of participation of learners

JUSTICE:

An understanding of power and history and how that affects our ability to participate: our positionality and intersectionality and how they influence our critical pedagogy

CARE:

A recognition of importance of affect and how that impacts on learning: humanising, care and trauma-informed pedagogies

PRAXIS:

The aforementioned dimensions resulting in a commitment to act, to take responsibility and move towards more socially just learning design

So what do I do in relation to genAI?

• Modelling / scaffolding
• Process learning (reflective process journals, iterative drafts with visible reasoning, viva-style checkpoints)
• Conversations
• Relationship building
• Openness and transparency
• Making AI part of the curriculum
• Critical AI literacy
• Grace

39

“The only way to know what your students are thinking and if they are doing their work properly, and to help them do it properly, is to spend time with them. It is too late to do this during an end of semester oral exam.” Tim Fawns, https://www.linkedin.com/feed/update/urn:li:activity:7465510261269893120/

EDN5511 Researching HE

40

41

42

43

Students’ final assignments

Rather than relying on AI tools to generate substantive content, I used them primarily to review, critique, and provide feedback on material that I had developed myself. This enabled me to retain ownership of the research process and ensure that the study remained grounded in its specific context while still benefiting from AI-assisted support. In this way, AI functioned as a supplementary research and writing aid rather than a substitute for critical thinking, contextual analysis, and scholarly judgement.

​

44

Students’ final assignments

At the same time, I do not believe it is possible, or even responsible, to think about AI as a neutral or objective tool. The literature increasingly argues that technology and human activity are deeply entangled in what is often described as a postdigital era, where the boundaries between technological systems and human thinking are

increasingly blurred (Bearman, Ryan and Ajjawi, 2023). This became visible in my own engagement with AI. While AI often helped me refine and extend my thinking, there were moments when I felt it subtly persuaded or redirected my thinking in ways that did not fully align with my original interests, intentions, or research focus. Sometimes AI responses appeared authoritative, coherent, and convincing, even when they steered my thinking way from what I was genuinely trying to explore.

​

45

Throughout this course I have often purposefully not used AI, especially in instances where I have given my opinion. My grammar might be terrible, and my thoughts jumbled, but I do not want to lose my ability to think or articulate my ideas. My hope is that we can shift the focus of research from technical proficiency (good academic writing, data gathering, etc.), which are the aspects I think that many people have traditionally struggled with during the research process, to a focus on what you are thinking, what you understand and can find out about your context and what the broader implications might be of these finding

​

46

47

48

The ability to say no / refuse AI

Focus on UNISA - genAI and ODL

In collaboration with my thinking partner Claude…

Unique concerns in the ODL space

Verifying that a piece of writing reflects this specific student's actual language development requires three things at once:

​

• Supervised or synchronous production — you watch (or otherwise control) the writing happen
• Confirmed identity — you know who is actually at the keyboard
• Known, bounded tools — you know what assistance was and wasn't available

​

-> recognising the impossibility of assessing of learning

​

50

UNISA's format — large classes, no physical contact, fully text-based, asynchronous submission — cannot deliver any of the three at scale. Not "struggles to," cannot, structurally, as currently designed. That's the bind: it's not a solvable engineering problem, it's a category mismatch between what the format can supply and what the claim ("this grade certifies this student's language proficiency") requires.

The impossible bind…

• Move from individual certainty to system-level assurance. Stop trying to verify every submission; audit a statistical sample, moderate across markers, track cohort-level anomalies. This is what actuarial/audit professions do when 100% verification is impossible — they don't pretend otherwise.
• Selectively reintroduce synchronous contact where the stakes justify the cost. Not every assessment needs it, but a capstone or a professional-accreditation-relevant module might — triage rather than blanket redesign.
• Be honest about what's actually being certified now. If individual authorship can't be verified, the degree may need to shift what it claims to certify — demonstrated engagement, applied performance, in-the-moment response — rather than silently continuing to imply a guarantee it can no longer make.

51

Benefits for ODEL and English Language Learning

• promote student autonomy (Van Horn, 2024),
• collaborative learning / conversation partner (Ghafouri, 2024)
• content curator
• personalized learning (Xiao & Zhi 2023), different proficiency levels
• scaffolding
• automated feedback/marking

52

53

54

Assessment for learning

55

Examples of assessment redesign in the ODL space

• Shifting weight toward invigilated or proctored exams, oral exams/vivas, and live presentations
• Less high stake assessments, towards more smaller assessments / more feedback
• More process-based assessment (drafts, reflective logs, version histories)
• Authentic, personalised, or locally grounded tasks that are harder for a chatbot to answer generically
• Requiring students to declare and acknowledge AI use, explicit instruction on "AI allowed / limited / not allowed" labels per assignment
• Multimodal assessments, ie oral presentations, transcribed/written up with genAI, reflection on what was gained/what got lost
• Making genAI explicit part of assessments - critical AI literacy becomes part of the curriculum

​

56

Critical AI literacy becomes part of the curriculum

• Treat AI-generated text as an object of study: rhetorical analysis of AI prose style, cliché patterns, and hallucination as a genre problem students can be taught to spot
• Teach transparent, disclosed AI use (e.g. "used for brainstorming or a grammar pass, not drafting") as an assessed skill, not a hidden violation to hunt for
• Position close reading, voice, and argument-building as the actual value proposition of an English degree in an AI-saturated workplace — this reframes literary study as more relevant, not less
• Consider a short compulsory "AI and writing" unit co-designed with the library and the Academic Integrity team, building on UNISA's compulsory Academic Integrity course launched in 2025

​

57

A Five-Level Scale for Assessing AI-Mediated Thinking

(Austin, 2026)

58

In closing….

• nobody signed up to be a forensic investigator of student writing
• extremely exhausting, frustrating, depressing
• if verifying individual authorship is structurally impossible in this format, then any staff member carrying private guilt over "letting cheating through" is holding themselves to a standard nobody could actually meet
• carry a personal/individual guilt that is actually a shared, institutional problem, not an individual one
• the same grace being asked toward students navigating an unfamiliar landscape is worth extending to colleagues navigating it too. Nobody, anywhere, has fully solved this yet — pretending otherwise helps no one

​

​

​

​

59

Any questions?

daniela.gachago@uct.ac.za

Communicating a position

Source: www.freepik.com

61

Communicating your GenAI position

• Share early and iteratively​
• Open space for conversation (anonymous?) about the document​
• Keep the document alive in class/ supervision – Beginning --> middle -> end (Shift in values)​
• Bring your position statement into a task analysis process (McCorkle, 2025 to follow)​

62

Disclosure

Source: www.freepik.com

63

Researching (with) disclosure

• Disclosure centres concerns about trust, academic integrity, and author credibility and accountability
• Type of tool(s) used
• Purpose of use
• AI Logs
• Engage with your students' logs
• Shared access (??)
• Stakeholders – external examiners, scholarly community?

​

64

When students admit AI use, but won't share their AI use with you?

The Artificial Intelligence Disclosure (AID) Framework

• Increased transparency
• Kari Weaver (Waterloo)
• Inspired by Contributor Roles Taxonomy (CRediT)
• Provide brief, targeted disclosure of GenAI use, associated with activities used for research writing
• AID Statement Example

https://crln.acrl.org/index.php/crlnews/article/view/26548/34482

https://sarua.africa/artificial-intelligence-disclosure-framework/

65

Artificial Intelligence Tool: ChatGPT v.4o and Microsoft Copilot (University of Waterloo institutional instance); Conceptualization: ChatGPT was used to revise research questions; Data Collection Methods: ChatGPT was used to create the first draft of the survey instrument; Data Analysis: Microsoft Copilot was used to verify identified themes coded from open ended survey responses; Privacy and Security: no identifiable data was shared with ChatGPT during the design of this study, only the University of Waterloo institutional instance of Microsoft Copilot was used to analyze any anonymized research data in compliance with University of Waterloo privacy and security policies; Writing—Review & Editing: ChatGPT was used in the literature review to provide sentence-level revisions and metaphor options; Project Administration: ChatGPT was used to establish a list of tasks and timelines for the study.

Resources and development

Source: www.freepik.com

66

67

Support resources available

CILT resources

​

​

UCT Libraries: AI LibGuide

Thank you

69

70

Thank you

71