Instructional design connects learning goals, practice, feedback, and assessment. In an AI-rich environment, the key question is what students need to understand and demonstrate—not how to make every task impossible for AI.
Decide when students need independent practice, when assistance is appropriate, and what evidence will show their thinking. Make these decisions explicit in the assignment directions.
Identify the learning: What should students know or be able to do?
Choose the evidence: What would a convincing demonstration look like?
Plan practice and support: What activities and feedback prepare students to succeed?
For each step, decide whether AI supports or bypasses the intended learning. See Vanderbilt’s overview of backward design.
ADDIE organizes design into analysis, design, development, implementation, and evaluation. Revisit earlier decisions as you learn what students need; the phases need not function as a rigid one-way checklist.
Analyze: identify learners, goals, constraints, and barriers.
Design: align outcomes, assessments, practice, and feedback.
Develop: create and review materials for accuracy and accessibility.
Implement: introduce the experience and explain expectations.
Evaluate: use student evidence and feedback to improve it.
The Successive Approximation Model emphasizes iterative development. Gather essential context, create an early prototype, review it with stakeholders, and improve it through repeated cycles.
For an AI-supported activity, prototype one small task and test both useful and problematic responses before expanding. Involve students and accessibility support where appropriate.
Read Allen Interactions’ description of SAM.
The revised cognitive taxonomy includes remember, understand, apply, analyze, evaluate, and create. Use it to clarify the kind of thinking students will practice—not as a requirement to include every category or a guarantee of assignment security.
Example: Students explain a concept, apply it to a case, compare two responses, and defend a recommendation using evidence. Assess the reasoning as well as the final product.
Explore the revised taxonomy at Iowa State CELT.
Different theories offer useful perspectives on practice, understanding, feedback, and meaning-making. Combine strategies deliberately rather than assigning students a fixed “learning style.”
CAST’s Universal Design for Learning framework supports planning for learner variability through engagement, representation, and action and expression. Design access into the experience from the start. Explore CAST’s UDL Guidelines.
Learning theories explain how learning happens and help us choose what evidence to assess. The examples below are original classroom applications of the cited ideas. Theories overlap; no single activity belongs exclusively to one theory.
Brief summary: Learners actively develop meaning by connecting experience with existing understanding. Meaningful problems, multiple perspectives, and reflection can help students reconsider their interpretations.
Example assessment: Give students a fictional classroom dilemma. Ask them to propose a response using two course concepts, compare approaches with a peer, and revise their plan. Submit the first plan, the revision, and a short explanation of what changed.
Assess: How well students apply concepts, use evidence, and justify revisions—not whether everyone reaches the same answer.
Brief summary: Learning involves how people attend to, organize, store, and retrieve information. Assessment can reveal the connections and reasoning behind an answer.
Example assessment: Ask students to connect six course concepts in a map or structured outline, label the relationships, and use two connections to explain a new case. After feedback, have them correct one connection and explain why.
Assess: Accurate relationships, a clear explanation, and application to the new case. Provide a text-based alternative to a visual map.
Brief summary: Learning is viewed through changes in observable performance, shaped by environmental cues and consequences. Reinforcement and repeated practice can strengthen a desired response.
Example assessment: For a procedure such as unit conversion, provide short practice sets with immediate corrective feedback. Recognize accurate performance, offer another attempt, and finish with a new, equivalent problem completed independently.
Assess: Accuracy and consistent performance against a clear criterion. Add an explanation task if the outcome also requires conceptual understanding.
Brief summary: Metacognition is a learning process—not a separate theory in this comparison. Students monitor their understanding, choose strategies, and evaluate whether those strategies work.
Example assessment: After a quiz or project, students identify one error or weak decision, explain its cause, choose a different strategy, and demonstrate that strategy on a new task.
Assess: A specific diagnosis, a reasonable strategy, and evidence of applying it. Do not grade personal disclosure or a story of improvement unsupported by the work.
Set boundaries around the skill being assessed. For example, students can build an initial concept map independently, then critique an instructor-provided AI map. They should explain the final connections themselves. An AI-generated reflection should not substitute for a student’s own account of their decisions.
Ertmer and Newby: Behaviorism, Cognitivism, Constructivism — an author-permitted reprint of the 1993 article; republished in 2013.
Carnegie Mellon Eberly Center: Learning Principles — research-based guidance on prior knowledge, practice, feedback, and self-directed learning.
Bloom’s taxonomy, shown elsewhere on this page, classifies cognitive objectives; it is not itself a theory of how learning occurs.
Students first recall a concept, apply it to a realistic case, compare explanations with peers, and revise their reasoning after feedback. Decide at each stage whether AI support helps students practice or bypasses the skill.
Before using the activity: Can students understand the goal, access the materials, practice the required skills, get feedback, and explain their decisions? Are AI expectations clear?