---
id: wiki-2026-0508-creativity-research
title: Creativity Research
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Creativity Studies, Creative Cognition Research]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [creativity, psychology, cognition, research]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: en
  framework: research-methods
---

# Creativity Research

## 매 한 줄
> **"매 creativity 의 measurable cognitive process — 매 mystical talent 아님"**. 매 1950 Guilford APA address 가 field 의 launch — 매 divergent thinking, fluency, originality 의 quantifiable. 매 2026 의 LLM-augmented co-creation, fMRI 의 default mode network 연구, computational creativity 의 active.

## 매 핵심

### 매 4P framework (Rhodes 1961)
- **Person**: 매 traits — openness, tolerance for ambiguity, intrinsic motivation.
- **Process**: 매 stages — preparation → incubation → illumination → verification (Wallas 1926).
- **Product**: 매 novel + useful (Stein 1953 의 standard definition).
- **Press**: 매 environment — domain, field gatekeepers (Csikszentmihalyi systems model).

### 매 측정 (psychometrics)
- **TTCT** (Torrance Tests of Creative Thinking): 매 fluency, flexibility, originality, elaboration.
- **AUT** (Alternative Uses Task): 매 brick 의 uses 나열 — 매 divergent thinking 의 standard.
- **CAT** (Consensual Assessment Technique, Amabile): 매 expert judges 의 product rating.
- **RAT** (Remote Associates): 매 convergent creativity (3 cue → 1 link word).

### 매 응용
1. K-12 design thinking curriculum.
2. 매 R&D ideation workshop (IDEO 의 protocols).
3. 매 LLM prompt engineering 의 creativity scaffolding.

## 💻 패턴

### Divergent thinking score (AUT)
```python
def aut_score(responses: list[str], reference_corpus: dict[str, int]) -> dict:
    """Score divergent-thinking output: fluency, flexibility, originality."""
    fluency = len(responses)
    categories = {classify_category(r) for r in responses}
    flexibility = len(categories)
    # originality = 1 - frequency in reference corpus (lower freq = more original)
    total = sum(reference_corpus.values()) or 1
    originality = sum(
        1 - (reference_corpus.get(r.lower(), 0) / total) for r in responses
    ) / max(fluency, 1)
    return {"fluency": fluency, "flexibility": flexibility, "originality": originality}
```

### LLM-augmented divergent ideation
```python
from anthropic import Anthropic

client = Anthropic()

def co_creative_ideation(prompt: str, n: int = 20) -> list[str]:
    """Use Claude as a divergent-thinking partner — temperature high for variance."""
    msg = client.messages.create(
        model="claude-opus-4-7",
        max_tokens=2000,
        temperature=1.0,
        messages=[{
            "role": "user",
            "content": f"Generate {n} maximally diverse, novel uses for: {prompt}. "
                       f"Span categories. Avoid clichés. One per line."
        }],
    )
    return [line.strip("- ") for line in msg.content[0].text.splitlines() if line.strip()]
```

### Consensual Assessment (CAT) aggregation
```python
import numpy as np
from scipy.stats import pearsonr

def cat_reliability(ratings: np.ndarray) -> float:
    """Inter-rater reliability via Cronbach's alpha across expert judges."""
    k = ratings.shape[1]
    item_var = ratings.var(axis=0, ddof=1).sum()
    total_var = ratings.sum(axis=1).var(ddof=1)
    return (k / (k - 1)) * (1 - item_var / total_var)
```

### Incubation effect simulation
```python
def incubation_benefit(initial_attempt_score: float, incubation_minutes: int) -> float:
    """Sio & Ormerod 2009 meta-analysis: ~0.3 SD boost after incubation."""
    if incubation_minutes < 5:
        return initial_attempt_score
    return initial_attempt_score + 0.3 * min(incubation_minutes / 30, 1.0)
```

### Default Mode Network proxy (resting-state correlation)
```python
def dmn_creativity_correlation(dmn_connectivity: float, ecn_connectivity: float) -> float:
    """Beaty et al. 2018: high creativity = strong DMN ↔ ECN coupling."""
    return dmn_connectivity * ecn_connectivity  # simplified product proxy
```

### Equivalence-class feature (Mednick RAT)
```python
def remote_associates_solve(cues: tuple[str, str, str], assoc_db: dict) -> str | None:
    """Find a single word that associates with all three cues."""
    sets = [set(assoc_db.get(c, [])) for c in cues]
    common = set.intersection(*sets)
    return next(iter(common), None)
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| Quick classroom screen | TTCT short form |
| Real-world product creativity | CAT with 3+ domain experts |
| Lab divergent thinking | AUT + originality corpus |
| Insight problem solving | RAT or compound remote associates |
| LLM augmentation | high-temperature ideation + human convergent filter |

**기본값**: 매 AUT + CAT for research; 매 LLM-as-divergent-partner + human-as-convergent-filter for applied work.

## 🔗 Graph
- 부모: [[Cognitive Psychology]]
- 변형: [[Divergent Thinking]] · [[Convergent Thinking]] · [[Computational_Creativity|Computational Creativity]]
- 응용: [[Design Thinking]] · [[Brainstorming]]
- Adjacent: [[Default Mode Network]]

## 🤖 LLM 활용
**언제**: 매 divergent ideation phase — 매 broad space exploration, 매 cliché breaking, 매 cross-domain analogies.
**언제 X**: 매 convergent evaluation alone — 매 LLM 의 novelty calibration 의 약함 (training data bias toward common). 매 originality scoring 시 의 corpus-based metric 결합 필요.

## ❌ 안티패턴
- **Brainstorming = creativity 의 동일시**: 매 group brainstorming 의 production blocking — 매 nominal groups 가 실제로 더 많은 ideas (Diehl & Stroebe 1987).
- **Originality 만 추적**: 매 useful 의 손실 — 매 novel + useful 가 정의.
- **Single judge CAT**: 매 inter-rater reliability 의 unverifiable.
- **TTCT 만 의 의존**: 매 ecological validity 의 약함 — real-world creative achievement prediction 의 modest (r ≈ 0.2-0.3).

## 🧪 검증 / 중복
- Verified (Guilford 1950, Torrance 1966, Amabile 1982, Beaty et al. 2018 NeuroImage).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — 4P framework, AUT/CAT/RAT measurement, LLM co-creation patterns 추가 |