Video glossary
Search plain-language definitions, filter by module, and load short video explainers only when you want them.
A fluent but false, fabricated, or unsupported AI-generated claim. An AI's smooth, confident output is a product of pattern prediction, not comprehension, so polish is no evidence of truth or learning.
The knowledge, skills, and judgment needed to understand, use, evaluate, question, and govern AI responsibly — including knowing when not to use it. AI literacy is always disciplinary because each field must decide what human judgment students still need to practice.
Assignments redesigned so that AI use is intentionally structured — neither fully banned nor fully open — allowing students to engage meaningfully while preserving the core learning goal.
The fragile goal of designing assignments AI cannot complete — undermined by capability drift, false confidence, inequitable enforcement, and brittle task design.
An assignment designed so AI use does not easily bypass the central learning goal. It assumes students may or may not use AI and makes evidence of learning robust either way through context, process, and judgment.
Systematic unfairness in AI outputs caused by biased data, design decisions, model behavior, or human assumptions. Asking AI to be unbiased is not enough; users must examine outputs, assumptions, and real-world impacts.
An assessment (Wiggins) that asks learners to apply knowledge, judgment, and skills in meaningful, discipline-specific contexts — not "real-world" decoration.
The way AI tools change over time, making yesterday's examples, policies, or limits potentially outdated. Teaching materials should date capability claims and include a realistic review cycle.
The amount of mental effort a learner must use to process information and complete a task. Too much can overwhelm learning, while too little can produce shallow processing; good design reduces unnecessary load while protecting essential thinking.
A bounded course moment where students do the essential thinking without AI substitution, so the invisible cognitive work becomes visible.
The legal and ethical questions raised when AI systems are trained on, or generate content based on, existing creative work. Useful review questions include source, consent, ownership, permitted use, and attribution.
The essential thinking, judgment, evidence, or decision-making that must remain the student's own work — the anchor of a defensible AI policy.
A broad label for machine-based systems that use data and patterns to generate predictions, recommendations, decisions, or content. AI is not magic or human thinking; the teaching question is what learning purpose it serves.
Challenges (from Bjork) that slow learners down in the short term but improve learning, retention, or transfer over time — such as retrieval, spacing, interleaving, and elaboration.
An academic-integrity approach that relies primarily on AI detectors to identify misuse. Because detectors can be unreliable, inequitable, and corrosive to trust, stronger foundations are transparent policy, sound assignment design, process evidence, and student guidance.
Unequal access to technology, connectivity, paid tools, time, language and disability support, and the confidence or help needed to benefit from digital systems. Fair AI design assumes uneven access and provides a reliable free or non-AI path.
Knowing how to use AI for a specific disciplinary task while recognizing its limits, naming the use, keeping required human review, and questioning its social effects and the power structures behind it.
The energy, water, hardware, and infrastructure costs associated with building and running AI systems. Responsible use asks whether an AI activity is educationally meaningful enough to justify those costs.
Four postures for deciding how AI should function inside a learning task: Oracle supplies answers, Tutor supports thinking, Adversary challenges claims, and Collaborator contributes ideas while the human keeps responsibility.
A component-by-component review of an assignment that identifies where AI might bypass essential thinking and where helpful support is appropriate. The goal is to preserve friction that produces learning and reduce friction that only blocks access.
AI systems that create new text, images, audio, video, code, or other outputs in response to a prompt. Their output can be useful, wrong, biased, or too polished to question, so treat it as a draft rather than a final authority.
The teaching stance of this course: refusing both AI hype and AI panic while making careful, evidence-informed design choices. It asks what learning is at stake, what evidence exists, what risks should be reduced, and what human judgment must remain central.
A design principle requiring a person to review, approve, or make consequential decisions rather than leaving them fully to AI. AI may draft, summarize, sort, or suggest; the human remains responsible for accuracy, fairness, privacy, and the final decision.
From Co-Intelligence: the uneven boundary of AI ability — a model may handle a hard task easily yet fail at something that looks simple, with no clean line between the two. Test before you trust, and never assume strength on one task transfers to the next.
The often-invisible human work behind AI systems — including data labelers and content moderators whose labor makes models usable, often for low pay and under difficult conditions.
A type of generative AI trained on massive amounts of text to predict the next likely word, or token, and produce language-like responses. There is no internal fact-checker, so fluent output can still be completely mistaken.
Difficulty that blocks access without building any learning — the counterpart to productive struggle, and often the right thing to remove with AI.
The integrity paradigm (Eaton, 2023) in which appropriate, documented AI use is normal — so the question shifts from "was AI used?" to "was the cognitive work done?"
The common design error of pairing an AI role with the wrong learner stage — most often putting a novice in front of an Oracle, or an Adversary.
Practices that prevent inappropriate sharing, collection, storage, or use of personal or sensitive information in AI tools. Prompts can carry student identities, grades, unpublished work, or sensitive context; do not enter private student data into public AI tools without institutional approval.
Artifacts that show how a student developed their work — notes, drafts, annotations, decision logs, prompt logs, or reflections — sized to reveal thinking, not to create paperwork.
The discipline that keeps evidence requirements honest: collect only what reveals a learning-relevant decision, and add at most one new element per assignment (the "rule of one").
The meaningful effort learners experience when comparing evidence, testing ideas, solving problems, or revising work in ways that build understanding. If AI removes all of that struggle, it may also remove the intended learning.
The instruction, question, context, role, example, rubric, or source material a user gives an AI system to guide its response. More context and clearer boundaries can improve the response, but no prompt guarantees accuracy.
An integrity approach (aligned with Eaton's post-plagiarism) where a suspected violation begins with a developmental conversation before any sanction. The goal is to understand what happened, repair understanding, and rebuild trust rather than making punishment the opening move.
Using AI to ask guiding questions that help learners reason step by step instead of receiving direct answers. The learner still has to explain, choose, revise, and justify their own reasoning.
A course-level statement explaining acceptable and unacceptable AI use, disclosure expectations, and the learning rationale behind those rules. Strong policies are clear, assignment-specific, connected to learning goals, and written to invite questions rather than threaten.
The scale's five contracts: Level 1 No AI; Level 2 AI Planning; Level 3 AI Collaboration (documented); Level 4 Full AI (student curates); Level 5 AI Exploration, where the AI's behavior is the object of study.
Transparency in Learning and Teaching (Winkelmes): an assignment-design framework that makes purpose, task, and criteria for success explicit.
A small unit of text — often a word, part of a word, or punctuation — that a language model reads and predicts.
A framework (CAST) for designing learning experiences that provide multiple ways to access, engage with, and demonstrate learning. AI can support alternative examples, summaries, translations, or formats, but access is not automatic; accuracy, bias, privacy, and workload still require review.
A prompt scaffold from the Pedagogical Promptbook — Instruction, Context, Role, Audience, Format, and Tone — used to structure any AI prompt students will work with. Naming each element treats a prompt as a designed object rather than a lucky guess.
