hum-historical-analogy

asgard-ai-platform ｜ 規劃與釐清 ｜ 原文 ｜ MIT

檢驗歷史類比是結構相似還是只有表面像，再拿來輔助決策

---
name: "hum-historical-analogy"
description: "透過辨識過去與現在情境之間結構上的相似與差異，用歷史類比來輔助策略決策。當使用者引用歷史前例來支持某個策略、需要評估某個歷史比較是否成立、或想從過去的事件中學習時使用，即使他們說的是「這跟網路泡沫很像」「歷史會重演」「我們能從 X 怎麼處理這件事學到什麼」也適用。"
metadata:
  category: "WP-19 文學院/人文"
  tags: ["humanities", "historical-analogy", "strategic-thinking"]
---

# 歷史類比

## 概述

歷史類比是把過去事件的教訓套用到現在的決策上。嚴謹地使用時，它能提供模式辨識與前瞻性；草率地使用時，則會因為過度擬合表面的相似、忽略結構上的差異而誤導人。

## 框架

```
鐵律：看結構上的相似，不是表面的相似

有效的類比需要共享「結構性」特徵（因果機制、權力動態、
系統性模式），不只是表面上看起來像。

「這家新創是下一個 Apple」，因為創辦人穿高領毛衣 =
表面的相似（毫無價值）。「這個市場具有與早期智慧型手機
普及時相同的需求端網路效應」 = 結構上的相似（有用）。
```

### 類比的評估步驟

1. **明確陳述類比**：「情境 A 像歷史事件 B，因為……」
2. **描繪結構上的相似**：共享了哪些因果機制、動態或模式？
3. **描繪結構上的差異**：根本上有什麼不同？
4. **評估平衡**：針對手上這個具體問題，相似之處是否勝過差異？
5. **謹慎萃取教訓**：這個類比提供了什麼具體、可行動的洞見？
6. **找出類比的極限**：類比在哪裡失效？

### 常見的類比陷阱

| 陷阱 | 說明 | 範例 |
|------|-----------|---------|
| **只挑對自己有利的** | 只選能支持你結論的歷史案例 | 「Kodak 沒能適應，所以我們必須轉型」（忽略堅持原路才是對的案例） |
| **結果偏誤** | 用歷史的結果來驗證類比 | 「Amazon 撐過了網路泡沫，所以我們也會」（倖存者偏誤） |
| **虛假的精確** | 期待歷史一模一樣地重演 | 「2008 年的危機花了 18 個月才復原，所以這次也會」 |
| **以今非古** | 用現在的知識去評斷過去的決定 | 「他們早該看出危機要來了」（他們當時沒有今天的資料） |

## 輸出格式

```markdown
# 歷史類比評估：{當前情境} ↔ {歷史事件}

## 類比
「{當前情境}像{歷史事件}，因為……」

## 結構上的相似
| 特徵 | 歷史 | 當前 | 相似程度 |
|---------|-----------|---------|-----------|
| {機制} | {當時怎麼運作} | {現在怎麼運作} | 強/中/弱 |

## 結構上的差異
| 特徵 | 歷史 | 當前 | 對類比的影響 |
|---------|-----------|---------|------------------|
| {因素} | {當時} | {現在} | 削弱/中性/強化 |

## 有效性評估
- 類比的整體強度：強 / 中 / 弱
- 適用於：{這個類比能為決策的哪些面向提供參考}
- 不適用於：{類比在哪裡失效}

## 教訓（附註保留）
1. {教訓}：保留：{這在哪裡可能不適用}
```

## 範例

### 正確的應用
**情境：**「2025 年的 AI 就像 1995 年的網際網路」

| 結構上的相似 | 1995 年網際網路 | 2025 年 AI | 強度 |
|----------------------|---------------|---------|----------|
| 能促成許多應用的通用技術 | ✓ | ✓ | 強 |
| 早期有期望膨脹的炒作週期 | ✓（網路泡沫） | ✓（AI 泡沫疑慮） | 強 |
| 基礎建設的建置階段（當時是寬頻，現在是 GPU/資料中心） | ✓ | ✓ | 強 |

| 結構上的差異 | 1995 年網際網路 | 2025 年 AI | 影響 |
|----------------------|---------------|---------|--------|
| 部署速度 | 寬頻鋪設要花好幾年 | AI 透過 API 幾分鐘就能取用 | 削弱（採用更快） |
| 既有業者的回應 | 既有業者反應遲緩（Blockbuster、報社） | 既有業者積極採用（Microsoft、Google） | 削弱（新創更難） |
| 法規環境 | 幾乎沒有管制 | 全球積極管制 AI（歐盟 AI 法案） | 削弱（限制更多） |

**結論**：中等的類比，適用於理解炒作週期模式與基礎建設投資階段，但不適用於預測新創 vs 既有業者的動態 ✓

### 不正確的應用
- 「AI 就像網際網路，所以所有 AI 公司都會成功」 → 只挑了贏家（Google、Amazon），忽略了超過 90% 的網路公司都失敗了。倖存者偏誤加上只有表面的相似。違反鐵律。

## 常見陷阱

- **存在多個類比：** 對於任何當前的情境，都可以畫出多個歷史類比，而且它們可能指向相反的結論。要考慮 2-3 個類比，不要只看最流行的那個。
- **最流行的類比往往最糟：** 「這就像網路泡沫」被到處拿來用，是因為大家熟悉，不是因為結構上的相似性很強。流行 ≠ 有效。
- **類比最適合用於模式辨識，不適合預測：** 「這個模式以前導致過 X」是有用的。「這會再次導致 X」則過度自信。
- **文化與制度脈絡會變：** 美國商業史的教訓，不一定適用於台灣的制度環境。要考慮系統性的差異。

## 參考資料

- Neustadt 與 May 的「Thinking in Time」方法論，見 `references/thinking-in-time.md`

---
name: "hum-historical-analogy"
description: "Use historical analogies to inform strategic decisions by identifying structural similarities and differences between past and present situations. Use this skill when the user draws on historical precedent to justify a strategy, needs to evaluate whether a historical comparison is valid, or wants to learn from past events — even if they say 'this is like the dotcom bubble', 'history repeats itself', or 'what can we learn from how X handled this'."
metadata:
  category: "WP-19 文學院/人文"
  tags: ["humanities", "historical-analogy", "strategic-thinking"]
---

# Historical Analogy

## Overview

Historical analogies apply lessons from past events to current decisions. When used rigorously, they provide pattern recognition and foresight. When used carelessly, they mislead by overfitting superficial similarities and ignoring structural differences.

## Framework

```
IRON LAW: Structural Similarity, Not Surface Similarity

A valid analogy requires shared STRUCTURAL features (causal mechanisms,
power dynamics, systemic patterns), not just surface resemblance.

"This startup is the next Apple" because the founder wears turtlenecks =
surface similarity (worthless). "This market has the same demand-side
network effects as early smartphone adoption" = structural similarity (useful).
```

### Analogy Evaluation Steps

1. **State the analogy explicitly**: "Situation A is like historical event B because..."
2. **Map structural similarities**: What causal mechanisms, dynamics, or patterns are shared?
3. **Map structural differences**: What is fundamentally different?
4. **Assess the balance**: Do similarities outweigh differences for the specific question at hand?
5. **Extract lessons carefully**: What specific, actionable insight does the analogy provide?
6. **Identify the analogy's limits**: Where does the analogy break down?

### Common Analogy Traps

| Trap | Description | Example |
|------|-----------|---------|
| **Cherry-picking** | Selecting only the historical case that supports your conclusion | "Kodak failed to adapt, so we must pivot" (ignoring cases where staying the course was right) |
| **Outcome bias** | Using the historical outcome to validate the analogy | "Amazon survived the dotcom bust, so we will too" (survivorship bias) |
| **False precision** | Expecting history to repeat exactly | "The 2008 crisis took 18 months to recover, so this one will too" |
| **Presentism** | Judging past decisions by present knowledge | "They should have seen the crisis coming" (they didn't have today's data) |

## Output Format

```markdown
# Historical Analogy Assessment: {Current Situation} ↔ {Historical Event}

## The Analogy
"{Current situation} is like {historical event} because..."

## Structural Similarities
| Feature | Historical | Current | Similarity |
|---------|-----------|---------|-----------|
| {mechanism} | {how it worked then} | {how it works now} | Strong/Moderate/Weak |

## Structural Differences
| Feature | Historical | Current | Impact on Analogy |
|---------|-----------|---------|------------------|
| {factor} | {then} | {now} | Weakens/Neutral/Strengthens |

## Validity Assessment
- Overall analogy strength: Strong / Moderate / Weak
- Valid for: {what aspects of the decision the analogy informs}
- Invalid for: {where the analogy breaks down}

## Lessons (with caveats)
1. {lesson} — caveat: {where this might not apply}
```

## Examples

### Correct Application
**Scenario:** "AI in 2025 is like the Internet in 1995"

| Structural Similarity | Internet 1995 | AI 2025 | Strength |
|----------------------|---------------|---------|----------|
| General-purpose technology enabling many applications | ✓ | ✓ | Strong |
| Early hype cycle with inflated expectations | ✓ (dotcom) | ✓ (AI bubble concerns) | Strong |
| Infrastructure buildout phase (broadband then, GPU/data centers now) | ✓ | ✓ | Strong |

| Structural Difference | Internet 1995 | AI 2025 | Impact |
|----------------------|---------------|---------|--------|
| Deployment speed | Years for broadband rollout | AI accessible via API in minutes | Weakens (faster adoption) |
| Incumbent response | Incumbents slow to respond (Blockbuster, newspapers) | Incumbents adopting aggressively (Microsoft, Google) | Weakens (harder for startups) |
| Regulatory environment | Minimal regulation | Active AI regulation globally (EU AI Act) | Weakens (more constraints) |

**Verdict**: Moderate analogy — valid for understanding the hype cycle pattern and infrastructure investment phase, but invalid for predicting startup vs incumbent dynamics ✓

### Incorrect Application
- "AI is like the Internet, so all AI companies will succeed" → Cherry-picks the winners (Google, Amazon) and ignores that 90%+ of dotcom companies failed. Survivorship bias + surface similarity only. Violates Iron Law.

## Gotchas

- **Multiple analogies exist**: For any current situation, multiple historical parallels can be drawn — and they may suggest opposite conclusions. Consider 2-3 analogies, not just the most popular one.
- **The most popular analogy is often the worst**: "This is like the dotcom bubble" is thrown around because it's familiar, not because the structural similarities are strong. Popularity ≠ validity.
- **Analogies work best for pattern recognition, not prediction**: "This pattern has led to X before" is useful. "This will lead to X again" is overconfident.
- **Cultural and institutional context changes**: Lessons from US business history may not apply to Taiwan's institutional environment. Account for systemic differences.

## References

- For Neustadt & May's "Thinking in Time" methodology, see `references/thinking-in-time.md`