hum-critical-thinking

asgard-ai-platform ｜ 規劃與釐清 ｜ 原文 ｜ MIT

拆解主張、證據、推理和隱藏假設，找出論證的邏輯謬誤

---
name: "hum-critical-thinking"
description: "運用結構化的批判性思考，包括辨識主張、證據、推理鏈、隱藏假設與邏輯謬誤，嚴謹地評估或建構具體的書面論證。當使用者提出一個具體的論證、主張、評論文章、研究發現或一段推理，要分析其邏輯有效性或缺陷時使用，即使他們說的是「這個論證成立嗎」「這裡面有哪些邏輯謬誤」「我這個論點做了什麼假設」也適用。不要用在隨意的計畫審視、旅遊規劃、專案風險腦力激盪或事前驗屍，「幫我挑我計畫的毛病」這類要求屬於紅隊/風險審查，不是論證分析。"
metadata:
  category: "WP-19 文學院/人文"
  tags: ["humanities", "critical-thinking", "logic", "argumentation"]
---

# 批判性思考框架

## 概述

批判性思考透過把論證拆解成主張、證據、推理與假設，系統性地評估論證。它找出論證哪裡強、哪裡弱、哪裡有謬誤，目的不是「贏」辯論，而是得出更站得住腳的結論。

## 框架

```
鐵律：把論證和說話的人分開

評估的是「論證」（主張 + 證據 + 推理），不是提出論證的人。
壞人可以提出好論證，受信任的專家也可能提出壞論證。
人身攻擊（攻擊人）和訴諸權威（信任人）都是謬誤。
```

### 論證結構

每個論證都有四個組成：
1. **主張**：在斷言什麼？（結論）
2. **證據**：有哪些事實/資料支持這項主張？
3. **推理**：證據如何支持主張？（邏輯的橋樑）
4. **假設**：推理要成立，哪些沒說出口的前提必須為真？

### 評估步驟

**步驟 1：找出主張**：到底在論證什麼？用一句話重述。

**步驟 2：檢視證據**
- 是事實，還是軼事？
- 夠不夠充分（資料點夠不夠多）？
- 相不相關（真的跟主張有關嗎）？
- 夠不夠新（沒有過時）？
- 這些證據有沒有可能支持另一個不同的主張？

**步驟 3：評估推理**
- 證據在邏輯上是否導向這項主張？
- 有沒有邏輯謬誤？（見下方目錄）
- 是不是把相關誤當成因果？
- 有沒有其他解釋？

**步驟 4：浮現假設**
- 這個論證要成立，什麼必須為真？
- 這些假設合理嗎？
- 如果某個假設是錯的，會怎樣？

### 常見的邏輯謬誤

| 謬誤 | 做了什麼 | 範例 |
|---------|-------------|---------|
| **人身攻擊（Ad hominem）** | 攻擊人，不是論證 | 「你沒資格談經濟，你又不是經濟學家」 |
| **稻草人（Straw man）** | 扭曲對方的論證，去攻擊較弱的版本 | 「你想減少軍費？所以你要我們毫無防備？」 |
| **假二分法（False dichotomy）** | 明明有更多選項，卻只給兩個 | 「你不是跟我們同一邊，就是跟我們作對」 |
| **滑坡（Slippery slope）** | 宣稱某件事必然導致極端的後果 | 「如果我們開放遠端工作，很快就再也沒人會進辦公室了」 |
| **訴諸權威（Appeal to authority）** | 用權威地位取代證據 | 「CEO 說 AI 會取代所有工作，所以一定是真的」 |
| **以偏概全（Hasty generalization）** | 從有限的案例得出廣泛的結論 | 「我兩個唸藝術的朋友都失業了，所以藝術學位沒用」 |
| **轉移焦點（Red herring）** | 引入不相關的資訊來分散注意 | 「對，我們的產品有 bug，但你看我們的公司文化多棒」 |
| **循環論證（Circular reasoning）** | 結論已經被假設在前提裡 | 「這是最好的做法，因為沒有更好的了」 |

## 輸出格式

```markdown
# 論證分析：{主題}

## 主張
{用一句話重述核心論證}

## 證據評估
| 證據 | 類型 | 充分？ | 相關？ | 夠新？ |
|----------|------|------------|-----------|----------|
| {證據 1} | {事實/軼事/專家/統計} | Y/N | Y/N | Y/N |

## 推理評估
- 邏輯有效性：{有效 / 有謬誤}
- 偵測到的謬誤：{列出並說明}
- 其他解釋：{還有什麼能解釋這些證據}

## 隱藏假設
1. {假設}：合理嗎？{Y/N，為什麼}

## 結論
- 論證強度：強 / 中 / 弱
- 主要弱點：{最大的缺陷}
- 什麼能強化它：{缺少哪些證據或推理}
```

## 範例

### 正確的應用
**情境：** 評估「遠端工作會降低生產力」這個主張
- **主張**：遠端工作會降低生產力
- **引用的證據**：「改成遠端之後，我們第三季的產出掉了 15%」
- **浮現的假設**：產出下降是遠端工作「造成」的（不是第三季的季節性、新人還在上手、產品轉向或疫情壓力）
- **謬誤**：Post hoc ergo propter hoc（在它之後，所以是因為它），把相關誤當成因果
- **結論**：弱，只有單一資料點、被多個因素混淆、沒有對照組的比較 ✓

### 不正確的應用
- 「CEO 說遠端工作不好，所以這個論證一定是錯的」 → 這是反向的人身攻擊（因為是誰說的就否定它）。違反鐵律：要評估論證，不是評估人。

## 常見陷阱

- **強的論證也可能有錯的結論：** 論證可以在邏輯上有效（推理從前提而來），但不健全（前提是假的）。兩者都要檢查。
- **「我覺得」不是證據：** 情緒作為人的體驗是有效的，但不能作為事實主張的證據。「我覺得不安全」是正當的擔憂；「我覺得這個政策沒用」不是政策失敗的證據。
- **舉證責任：** 提出主張的人要負舉證責任。「你無法證明它是錯的」不能證明它是對的（訴諸無知）。
- **鋼人 > 稻草人：** 不要去攻擊論證最弱的版本（稻草人），而是建構它「最強」的版本（鋼人），再來評估。這樣的分析比較好。
- **批判性思考不是憤世嫉俗：** 目標是讓信念更站得住腳，不是對一切都懷疑。有些論證就是很強，承認強的論證也是批判性思考的一部分。

## 參考資料

- 形式邏輯的符號與三段論，見 `references/formal-logic.md`
- 附延伸範例的謬誤目錄，見 `references/fallacy-catalog.md`

---
name: "hum-critical-thinking"
description: "Apply structured critical thinking — identifying claims, evidence, reasoning chains, hidden assumptions, and logical fallacies — to evaluate or construct specific written arguments rigorously. Use this skill when the user presents a concrete argument, claim, op-ed, research finding, or piece of reasoning to be analyzed for logical validity or flaws, even if they say 'is this argument valid', 'what logical fallacies are in this', or 'what assumptions am I making in this thesis'. Do NOT use for casual plan review, trip planning, project risk brainstorming, or pre-mortems — 'poke holes in my plan' requests are red-team / risk review, not argument analysis."
metadata:
  category: "WP-19 文學院/人文"
  tags: ["humanities", "critical-thinking", "logic", "argumentation"]
---

# Critical Thinking Framework

## Overview

Critical thinking systematically evaluates arguments by decomposing them into claims, evidence, reasoning, and assumptions. It identifies where arguments are strong, weak, or fallacious — not to "win" debates but to arrive at better-justified conclusions.

## Framework

```
IRON LAW: Separate the Argument from the Person

Evaluate the ARGUMENT (claim + evidence + reasoning), not the person
making it. A bad person can make a good argument. A trusted expert
can make a bad argument. Ad hominem (attacking the person) and appeal
to authority (trusting the person) are both fallacies.
```

### Argument Structure

Every argument has four components:
1. **Claim**: What is being asserted? (conclusion)
2. **Evidence**: What facts/data support the claim?
3. **Reasoning**: How does the evidence support the claim? (the logical bridge)
4. **Assumptions**: What unstated premises must be true for the reasoning to hold?

### Evaluation Steps

**Step 1: Identify the claim** — What exactly is being argued? Restate in one sentence.

**Step 2: Examine the evidence**
- Is it factual or anecdotal?
- Is it sufficient (enough data points)?
- Is it relevant (does it actually relate to the claim)?
- Is it current (not outdated)?
- Could the evidence support a different claim?

**Step 3: Evaluate the reasoning**
- Does the evidence logically lead to the claim?
- Are there logical fallacies? (see catalog below)
- Is correlation being mistaken for causation?
- Are there alternative explanations?

**Step 4: Surface assumptions**
- What must be true for this argument to work?
- Are these assumptions reasonable?
- What happens if an assumption is wrong?

### Common Logical Fallacies

| Fallacy | What It Does | Example |
|---------|-------------|---------|
| **Ad hominem** | Attacks the person, not the argument | "You can't talk about economics, you're not an economist" |
| **Straw man** | Distorts the opponent's argument to attack a weaker version | "You want to reduce military spending? So you want us defenseless?" |
| **False dichotomy** | Presents only two options when more exist | "You're either with us or against us" |
| **Slippery slope** | Claims one event will inevitably lead to extreme consequences | "If we allow remote work, soon no one will come to the office ever" |
| **Appeal to authority** | Uses authority status instead of evidence | "The CEO says AI will replace all jobs, so it must be true" |
| **Hasty generalization** | Draws broad conclusion from limited cases | "My two friends who studied art are unemployed, so art degrees are useless" |
| **Red herring** | Introduces irrelevant information to distract | "Yes, our product has bugs, but look at our amazing company culture" |
| **Circular reasoning** | Conclusion is assumed in the premise | "This is the best approach because there's no better one" |

## Output Format

```markdown
# Argument Analysis: {Topic}

## Claim
{One-sentence restatement of the core argument}

## Evidence Assessment
| Evidence | Type | Sufficient? | Relevant? | Current? |
|----------|------|------------|-----------|----------|
| {evidence 1} | {fact/anecdote/expert/stat} | Y/N | Y/N | Y/N |

## Reasoning Evaluation
- Logical validity: {valid / fallacious}
- Fallacies detected: {list with explanation}
- Alternative explanations: {what else could explain the evidence}

## Hidden Assumptions
1. {assumption} — reasonable? {Y/N, why}

## Verdict
- Argument strength: Strong / Moderate / Weak
- Key weakness: {the biggest flaw}
- What would strengthen it: {what evidence or reasoning is missing}
```

## Examples

### Correct Application
**Scenario:** Evaluating the claim "Remote work reduces productivity"
- **Claim**: Remote work reduces productivity
- **Evidence cited**: "Our Q3 output dropped 15% after going remote"
- **Assumption surfaced**: That the output drop was CAUSED by remote work (not by Q3 seasonality, new hires ramping up, product pivot, or pandemic stress)
- **Fallacy**: Post hoc ergo propter hoc (after it, therefore because of it) — correlation assumed to be causation
- **Verdict**: Weak — single data point, confounded by multiple factors, no controlled comparison ✓

### Incorrect Application
- "The CEO said remote work is bad, so the argument must be wrong" → Ad hominem in reverse (dismissing based on who said it). Violates Iron Law: evaluate the argument, not the person.

## Gotchas

- **Strong arguments can have wrong conclusions**: An argument can be logically valid (reasoning follows from premises) but unsound (premises are false). Check both.
- **"I feel" is not evidence**: Emotions are valid as human experiences but not as evidence for factual claims. "I feel unsafe" is a legitimate concern; "I feel this policy doesn't work" is not evidence of policy failure.
- **Burden of proof**: The person making the claim bears the burden of proof. "You can't prove it's wrong" is not evidence that it's right (argument from ignorance).
- **Steelmanning > strawmanning**: Instead of attacking the weakest version of an argument (strawman), construct the STRONGEST version (steelman) and then evaluate it. This produces better analysis.
- **Critical thinking is not cynicism**: The goal is better-justified beliefs, not skepticism of everything. Some arguments are strong. Acknowledging strong arguments is part of critical thinking.

## References

- For formal logic notation and syllogisms, see `references/formal-logic.md`
- For fallacy catalog with extended examples, see `references/fallacy-catalog.md`