Unit 4: Ideation to Visualization - Generative AI as co-creator

Design Thinking's third and fourth phases (Ideate and Prototype in the Stanford d.school model, c. 2005) move a validated problem into a spread of possible solutions and then into tangible form. When a generative model joins this work, it acts as a co-creator: a partner that expands option space rather than a tool that merely executes. This unit treats that partnership and the discipline it demands.

• Divergent–convergent rhythm: ideation first widens the field (many ideas, deferred judgement) then narrows it (selection against criteria); AI amplifies both moves.
• Co-creation, not automation: the human owns intent, framing and final judgement; the model supplies volume, variation and unexpected combination.
• Traceability: because AI output is derived from training data, every co-created idea carries a provenance question that ethics and later IP handling must answer.
• Fidelity ladder: ideas descend from abstract concept → sketch → low-fidelity mock → high-fidelity prototype, each step making the idea more testable.

II. Co-imagination Ethics

The ethics of imagining with a machine

Co-imagination ethics governs how credit, ownership and honesty are handled when a human and a generative system produce ideas jointly.

A. Plagiarism

Passing off derived or borrowed expression as one's own original work.

• Definition in the AI context: reproducing training-data expression—text, image style, code—closely enough that it substitutes for the source without attribution.
• Memorisation risk: large models can regurgitate near-verbatim passages or a distinctive artist's style; e.g. prompting "in the style of a living illustrator" can reproduce protected expression.
• Detection duty: co-creators must check output against sources (reverse-image search, plagiarism scanners) before publishing, not after.
• Attribution practice: cite AI assistance and any recognisable source; disclosure converts hidden copying into transparent derivation.

B. Creativity

The capacity to generate ideas that are both novel and useful.

• Combinatorial creativity: most machine "novelty" is recombination of existing elements—new pairings, not new primitives (Boden's combinational category).
• Human role: framing the prompt, judging usefulness and rejecting the banal is where human creativity concentrates; the model has no goal of its own.
• Fixation risk: over-relying on the first fluent AI suggestion causes design fixation, narrowing rather than widening the option field.
• Guardrail: use AI to break fixation (ask for ten opposite framings) rather than to settle it.

C. Originality

The quality of being genuinely new in source, not merely different in surface.

• Originality vs novelty: a variation can be novel (unseen) yet unoriginal (traceable wholesale to one source); originality requires a distinct creative contribution by the author.
• Legal edge (informational): many jurisdictions grant copyright only to works with human authorship, so purely machine-generated output may fall outside protection.
• Threshold test: ask whether the human added meaningful selection, arrangement or transformation—if so, originality is defensible.

III. From Ideas to Action

Converting selected concepts into committed next steps

This section addresses the gap where most design projects stall: many ideas exist, but none has an owner, a form or a deadline.

A. From Ideas to Action

Moving from a chosen idea to a concrete, testable commitment.

• Make it tangible: an idea becomes action only when expressed as an artefact—sketch, storyboard, paper prototype—that others can react to.
• Smallest testable step: define the minimum experiment that checks the riskiest assumption (a Minimum Viable Prototype), not a full build.
• Ownership and timebox: assign each action a responsible person and a short deadline; AI can draft the task breakdown but cannot carry accountability.
• Feedback loop: action produces evidence, evidence reframes the problem, closing back into ideation—the process is iterative, not linear.
  • AI role here: rapidly drafts variants, test scripts and synthetic user personas to accelerate the build–test cycle.

IV. Ideation – Frame & Reframe the Problem Space

Setting and resetting the boundary within which ideas are generated

Ideation quality is bounded by how the problem is stated; framing decides which ideas are even thinkable.

A. Frame the Problem Space

Stating the problem so that it invites solutions rather than presupposing one.

• "How Might We" (HMW) format: convert an insight into an open question, e.g. "How might we help commuters feel time is well spent?"—open enough to invite ideas, narrow enough to focus.
• Point-of-view statement: [User] needs [need] because [insight]—anchors ideation in a real person and motive.
• Right altitude: a frame too broad ("improve transport") yields vague ideas; too narrow ("add a cup-holder") forecloses them.

B. Reframe the Problem Space

Deliberately shifting the frame to expose overlooked solution territory.

• Challenge the assumption: list implicit constraints, then negate each ("what if there were no queue at all?") to unlock new directions.
• Change the actor or scale: reframe from user to bystander, or from one person to a crowd, revealing different needs.
• AI-assisted reframing: prompt the model for "five alternative problem statements from different stakeholders' viewpoints" to surface frames a single team would miss.
• Reframe is iterative: each round of testing may reveal the real problem was mis-stated, sending the team back to reframe before re-ideating.

V. Tools for Ideation & Idea Selection

Techniques for generating options and then choosing among them

A. Tools for Ideation

Structured methods that widen the field of possible solutions.

• Brainstorming: rapid group idea capture under deferred judgement; quantity first, "yes-and" builds on others' ideas.
• Brainwriting (6-3-5): 6 people write 3 ideas in 5 minutes, then pass sheets—removes the loudest-voice bias of verbal brainstorming.
• SCAMPER: prompt list—Substitute, Combine, Adapt, Modify, Put to other use, Eliminate, Reverse—forces variation on an existing idea.
• Mind mapping: radial branching from a central problem to expose associations visually.
• Worst possible idea: generate deliberately bad ideas, then invert them to reveal hidden good ones.
• Generative-AI ideation: prompt for large batches of divergent concepts, then re-prompt for combinations; treats the model as an infinite, unjudging brainstorm partner.

B. Idea Selection

Converging on the strongest ideas against explicit criteria.

• Dot voting: each participant places a fixed number of dots on preferred ideas; fast, democratic first cut.
• Now–Wow–How matrix: plot ideas on ease of implementation × originality; "Wow" ideas (original + feasible) are prioritised.
• Impact–effort matrix: two-axis grid selecting high-impact, low-effort "quick wins".
  • Worked example: given four ideas scored on impact (1–5) and effort (1–5)—A(5,2), B(4,4), C(2,1), D(3,5)—A is the clear quick win (high impact, low effort); D is a "money pit" (high effort, low impact) and is dropped.
• Weighted scoring: rate each idea against weighted criteria (desirability, feasibility, viability) and sum; makes trade-offs explicit.
• Deferring judgement rule: selection must be a separate phase from generation, or premature criticism kills fragile ideas.

VI. Visualization in Design Thinking – Importance and Methods

Making ideas visible so they can be understood, tested and improved

A. Importance of Visualization

Why turning ideas into visible form is central, not decorative.

• Shared understanding: a drawing creates a common reference that words alone leave ambiguous—"show, don't tell".
• Externalising thought: putting an idea on paper offloads working memory and exposes gaps the mind glossed over.
• Faster feedback: a visible artefact invites concrete critique early, when change is cheap.
• Bias-neutral: a sketch lets non-designers contribute, flattening expertise hierarchies.

B. Methods of Visualization

The concrete techniques for rendering ideas visible along the fidelity ladder.

1. Low-fidelity methods:
   • Sketching: quick hand drawings of concepts; deliberately rough to keep them disposable and open to change.
   • Storyboarding: a sequence of frames showing a user's journey through the solution over time, exposing interaction gaps.
   • Journey/empathy maps: diagrams charting a user's actions, thoughts and emotions across an experience.
2. High-fidelity methods:
   • Wireframes and mockups: structured layouts approaching the real interface, used once the concept is stable.
   • Prototypes: interactive models (paper, clickable or physical) that users can actually operate to generate behavioural data.

• Generative-AI visualization: text-to-image and diagramming models turn a written concept into a rendered scene or interface mock in seconds, compressing the sketch step—provided provenance and style-plagiarism checks from Section II are applied.
• Fidelity discipline: match fidelity to purpose—low fidelity to explore breadth, high fidelity to validate a near-final choice; investing in polish too early wastes effort and biases feedback toward surface, not substance.