From Asking to Structuring: Designing AI Interaction

Position in Module#

• Unit: 1
• Topic: From Asking to Structuring: Designing AI Interaction
• Literacy perspective: Interaction
• Collaboration level: Instruction
• Fundamentals focus: Generative AI Fundamentals

Literacy Perspective#

• Primary literacy: Interaction
• Secondary literacy, if any: None
• What this literacy explains: Information, request structure and the sequence of exchanges shape the model's response.
• What participants should manipulate, observe, and interpret: Compare a direct answer with a structured process supported by checked factual and audience information. Examine accuracy, traceability, audience fit and suitability for the intended use.

Collaboration Level#

• Level relation: Instruction, with a preview of Alignment, Co-construction and Reflection
• Collaboration move being learned: Turn an open request into an explicit task and guide it from context through review.
• What participants should be able to do differently after this Friday: Use the supplied five phases and explain their contribution. Designing an original workflow comes later.

The five phases are:

1. Organise context: retain relevant, authorised information; remove repetition and sensitive or irrelevant material.
2. Clarify: identify gaps, ambiguity and decisions; supply human judgement.
3. Agree an approach: inspect and adjust the plan before execution.
4. Execute: create the document from the accepted context and plan.
5. Review and improve: check against the task, evidence and criteria; implement and check at least one justified improvement.

Use one browser chat client without file access. All phases happen through conversation. Unit 2 will make the benefit of persistent files visible. The brief tour of collaboration levels is orientation; this Friday does not establish mastery of them all.

Generative AI Fundamentals#

Build a small mental model that participants can use during the experiment:

• Begin with y = ax + b. Parameters are a and b; training estimates them from data; inference applies the model to an input. A threshold on y can illustrate classification.
• Transfer the distinction to language models: many learned parameters, trained on large text collections, process a sequence of tokens.
• The model produces a probability distribution for the next token. Generation selects a token, appends it and repeats. This turns a prediction step into a complete response.
• A chat supplies more than the latest sentence: available history, instructions and other context also shape the continuation.
• Repeat one prompt to demonstrate possible variation. Plausibility and repetition do not establish correctness.

The practical point: participants can shape behaviour at inference time by changing the supplied context and interaction. Unresolved ambiguity may become an unexamined model choice. This is a useful warning, not a claim that every ambiguity inevitably causes an error.

Recurring Agentic Coding Demonstration#

Mention the recurring coding example only if useful. The main classroom task is a five-minute presentation on a random topic, with three to five slides and speaker notes or a script. Compare a direct request with the five phases in browser chat. Creating an executable workspace belongs to Unit 2.

Friday Activity Table#

Experiment Focus#

• Experiment focus: a direct answer versus the complete five-phase approach with checked external foundations.
• What changes: B alone receives the researched factual and audience foundations, clarification, planning, guided execution and review.
• What remains constant: document, task, audience, purpose, intended effect and four comparison criteria. Use the same model where possible.

Start A and B in separate fresh conversations. Preserve A's first complete response without corrective follow-up. This comparison combines several changes; it cannot isolate the effect of one phase.

Observation Focus#

• Observation focus: factual agreement with checked sources, traceability of important claims, audience fit, and suitability for purpose and effect.
• Evidence participants should capture: request and unchanged A; sources and audience basis; expectation, observation and decision for each phase; improved B; comparison with concrete excerpts or features.
• Interpretation participants should be prepared to make: What changed, which evidence supports that judgement, what remains uncertain, and which practice should transfer?

Challenge Kickoff Deliverables#

Participants leave with a chosen multipart document and its task, audience, purpose and intended effect; a captured direct answer or plan to capture it; a research plan for B's foundations; and a started phase record. Confirm the four criteria before comparing results.

The challenge submission is one ZIP containing Direktantwort.pdf, Fünf-Phasen-Artefakt.pdf and Nachweisdossier.pdf. Classroom presentation notes are preparation, not additional submission items. Use the challenge for the full requirements.

Teacher Setup#

Prepare the Unit 1 slides, course map, five-phase guide, presentation task, challenge and a simple evidence record. Test the browser client and have prepared outputs available if access fails. Use authorised, non-sensitive materials and demonstrate how to select useful context.

Choose the platform, work mode, examples, scaffolding and time per phase to suit the cohort. The review can finish in class or continue during the challenge. Preserve the comparison and evidence requirements either way.