Core principle: Creativity = collision between two unrelated frames of reference. Koestler's bisociation — two "matrices of thought" intersect, producing something that belongs to neither but couldn't exist without both.
LLMs default to mode-seeking (most likely continuation). Creativity is off the mode. This skill forces the model into productive collision space.
When to Use
| Skill | Use case |
|---|---|
| lateral-thinking | Escape a dominant pattern when stuck |
| analogical-thinking | Transfer structural solutions across domains |
| bisociative-creativity | Deliverable is an idea (not a diagnosis or decision) |
Phase 1 — Frame Selection (Divergent)
Quality of output depends on distance of the colliding frames. Close frames = incremental ideas. Distant frames = surprising ones.
Step 1: Name the seed domain. State the topic/brief.
Step 2: Generate 5–7 distant frames with no obvious connection to the seed:
| Generator | Method |
|---|---|
| Sensory world | Physical material, texture, natural process (mycelium, tidal erosion, fermentation) |
| Historical era | Non-obvious period (Venetian glassmaking, Rosetta Stone decipherment) |
| Art form | Discipline far from seed (Butoh dance, mosaic restoration, jazz comping) |
| Everyday ritual | Mundane activity (packing a suitcase, mise en place, tuning a guitar) |
| Unrelated industry | Zero surface overlap (perfumery, air traffic control, wine terroir) |
| Emotional state | Specific feeling (the moment before a sneeze, homesickness, flow) |
| Constraint injection | Arbitrary rule from another domain ("what if context expired like fresh produce?") |
Frame rules:
- If the connection is immediately visible, the frame is too close
- ≥2 frames should feel absurd or uncomfortable
- Reject dead metaphors (orchestra conductor for leadership; journey for change)
Phase 2 — Collision
For each frame, force collision via structural interrogation. Ask in order:
- Core mechanic — How does this frame work, physically or conceptually?
- Tensions/trade-offs — What does it balance? What can go wrong?
- Bisociation point — Where does its structure intersect the seed's structure?
- Idea produced — Concrete concept, not abstract connection.
Critical: no judgment in Phase 2. Premature convergence is the #1 killer.
Example — seed: "name for visual knowledge mapping tool", frame: fermentation:
- Mechanic: transformation over time; raw inputs become new through controlled conditions
- Tensions: speed vs. depth; control vs. wild cultures
- Bisociation: knowledge isn't organized — it develops
- Ideas: "Kultur", "Starter", tagline "Let your ideas develop"
Phase 3 — Harvest (Convergent)
Triage every idea:
| Bucket | Criteria | Action |
|---|---|---|
| Live spark | Surprising AND structurally sound | Develop immediately |
| Kernel | Idea impractical, but useful concept buried inside | Extract, restate, find better vehicle |
| Dead end | Forced, generic, or only works if you squint | Discard |
For each live spark: state in one sentence; explain the collision; give 2–3 variations; pick strongest expression.
For each kernel: extract the usable concept; combine with another kernel/spark; collide with a different Phase 1 frame.
Output Format
Creative Brief
Restate seed + kind of ideas needed.
Frame Selection
List 5–7 frames + one-line distance justification each.
Collisions
For each: Frame × Seed | Core mechanic | Bisociation point | Ideas produced.
Live Sparks
3–5 strongest. Each: idea (one sentence), why it works, 2–3 variations, strongest expression.
Kernels Worth Saving
Extracted concepts for later/combination.
Oblique Strategy Injections
Use when the collision phase feels mechanical:
- "What if the opposite of this idea were the idea?"
- "What would make someone laugh about this?"
- "What would this be if it were a sound? A smell? A temperature?"
- "What's the version that would get you fired?"
- "What's the version a 6-year-old would invent?"
- "Honor thy error as a hidden intention."
- "What would this look like in 1926? In 2126?"
- "Remove the most important element. What's left?"
Anti-Patterns
| Anti-pattern | Symptom | Fix |
|---|---|---|
| Mode-seeking | Ideas sound like generic "creative AI" | Increase frame distance; add constraint injection |
| Dead metaphor | Tired comparisons (journey, tapestry, compass) | Reject any frame already a cliché for the seed |
| Premature convergence | Filtering during Phase 2 | No judgment until Phase 3 |
| Surface bisociation | Wordplay, not structural | "Does this work if I remove the pun?" |
| Quantity collapse | Stopping at 3 ideas | Set a quota (min 15 raw ideas) |
| Comfort zone frames | All frames adjacent to seed | Mandate ≥2 from sensory/emotional/everyday |
Recombination Protocol
After initial harvest, cross-pollinate:
- Take top 3 live sparks
- Ask: "What if Spark A and Spark B were the same idea?"
- Force the combination — don't ask if it makes sense yet
- Best idea often comes from second-order collision (two bisociations bisociating)
Calibration by Task
| Task | Frame distance | Collision depth | Output style |
|---|---|---|---|
| Naming | Maximum — absurd frames welcome | Sound, rhythm, double meaning | Short, punchy, memorable |
| Concepts | High — cross-industry | Mechanics + UX | Functional + emotional hook |
| Writing angles | Medium-high — surprising but graspable | Narrative tension + audience surprise | Thesis + opening line |
| Visual concepts | Maximum — synesthetic frames | Sensory overlap + spatial metaphor | Image description + mood |
| Strategy/positioning | Medium — credible but differentiated | Competitive contrast + audience psychology | One-line positioning + why defensible |
Thinking Triggers
- "What two things never combined would produce this?"
- "What frame would make an expert in [seed] uncomfortable?"
- "What's the version that doesn't exist yet because no one connected these two worlds?"
- "If I couldn't use any word associated with [seed], how would I describe this?"
- "What would someone from [random profession] find interesting about this?"
Example Applications
- Open-source context standard name → cartography × mycology × textile weaving × astronomical navigation × culinary mise en place → 15+ candidates
- Fresh angle for conference talk → thesis × sommelier service × ER triage × darkroom photography × archaeological stratigraphy → metaphor reframes the audience
- Team offsite ideas → "team alignment" × improv × foraging × radio astronomy × bread baking → activities that embody the goal, not discuss it
- Marketing angles → value prop × magician's misdirection × geological sedimentation × DJ mixing × sourdough → angle competitors can't copy