About this workshop
We cover the most recent data on what students are actually doing with these tools; how software claiming to detect AI writing works, and where it doesn't; where detection sits as evidence; and workable AI use declarations to accompany assessments. We look at what the pedagogy literature says about the cognitive outsourcing that comes with AI tool use, and experiment with tools marketed as learning instruments, including Copilot's Study and Learn agent. We then turn to redesigning assessment tasks so the evidence of learning holds up, anchored by TEQSA's guidance on assessment reform and ASQA's draft principles for the responsible use of AI.
No technical background assumed. Bring your ideas and insights to share with colleagues, or an assessment task you are uneasy about. A laptop is essential.
Course sections
Each section below stands on its own and can be read in any order. Sections are added here as they are built.
What Students Are Actually Doing
The measured picture of student AI use, and how soft it is: what the Australian surveys found, why the confident staffroom numbers should be held loosely, and why a cohort, a nationality or a writing style tells you nothing about who used AI.
How Detection Technology Works
What the software claiming to detect AI writing actually does: perplexity and burstiness, trained classifiers, vendor claims against the independent evidence, and a base-rate calculator showing why the score is never the probability a student cheated.
Detection as Evidence, Not Verdict
What an integrity process can and cannot do with a percentage: the balance of probabilities and Briginshaw, the TEQSA and ASQA positions, the evidence mosaic, and what the documented Australian record shows about getting it wrong.
AI Text Watermarking and Content Provenance
The statistical mark now woven into AI text: green-list token biasing, Google's SynthID, C2PA file provenance, the EU AI Act Article 50 driver, and why a detected mark proves processing, not authorship. Pairs with an interactive walkthrough deck.
Cognitive Outsourcing
What the research really shows about handing thinking to AI: the same tool can scaffold learning or quietly replace it, why the brain-harm headlines outrun their evidence, and why even the good-news studies get retracted. The honest position holds both.
Hands On: Copilot Study and Learn
The learning agent on every CDU student account, taken seriously and then taken apart: what Microsoft built, how CDU accounts are really configured, the break-it exercise, and Gemini Notebook as the contrast case. The restraint is a prompt, not a property.
Prompt Critics
A presenter-driven game for the room: eight everyday university prompts about to be sent to Copilot, each with a problem, and the audience calls it out before the send. Full-screen in a browser, no login and no real AI, ending with a takeaway checklist. Pairs with the Copilot Study and Learn section.
Declarations That Do Something
Why most AI-use declaration forms are theatre, what a declaration can and cannot do as evidence, the range of things "used AI" can mean, a rewrite-the-form activity with a deliberately weak sample, and what to do when a declaration is plainly false.
Assessment Redesign
Redesigning tasks so the evidence of learning holds up when students can use AI: TEQSA's validity frame and secure-task options for higher education, ASQA's rules of evidence for VET, what each mode of assessment can and cannot establish, and a review lens for the task you brought.
Ancient Plagiarism
A short thought experiment: humans have always built new texts out of older ones, from the Gilgamesh flood to the Synoptic Gospels. What that history changes about the anxiety over AI writing, and the shift from judging a text's originality to judging a user's honesty. Pairs with an interactive walkthrough deck.
The CDU Generative AI Policy
A plain-language guide to the University's Generative AI Policy and the Academic Integrity Policy, and a scenario discussion activity that tests everyday teaching situations against what the policies actually say.
What We’re Not Sure About
The open questions behind the confident sections: where the evidence on detection, assessment and learning is thin, contested, dated, or missing for Australia, and what is actually settled. Distinctive to this course, and true to how it is advertised.
Resources
Recommended viewing for this workshop: documentaries, lectures, reporting and panel discussions on AI in education and assessment, each with a short annotation on why it earns its place.