Psychology & Behavior
Cognitive Bias Checklist
認知偏誤自檢 · Source: Kahneman, Ariely, Thaler
Pre-flight check before any important decision — scan for five high-frequency biases before you commit
Core Concept
We all have systematic cognitive biases, and we're usually unaware they're operating. This framework doesn't try to eliminate bias (impossible) — it forces you to pause before deciding and ask questions that make biases visible.
✓ When to use this
As a "quality check" right before locking in a major decision. Especially when you notice strong personal lean toward one option, or unusually unanimous team agreement — consensus often signals shared bias.
✗ When not to use this
Do not run the checklist for small calls — it causes decision fatigue. When emotionally activated, you cannot self-check honestly; cool down with 10/10/10 first.
Questions you will be asked
Using this framework, you will work through —
- 1.What decision are you considering?
- 2.Confirmation bias: Have you seriously searched for evidence against this decision?
- 3.Availability heuristic: Is your judgment influenced by recent events or the most easily recalled examples?
- …and 3 more
Worked example
Expand to see what a filled-in run looks like
›
Worked example
Expand to see what a filled-in run looks like
Situation
準備加碼買入一檔近三個月漲了 80% 的 AI 概念股，因為「明顯是趨勢」。
1. What decision are you considering?
把現金 30% 加碼買入這檔股票。
2. Confirmation bias: Have you seriously searched for evidence against this decision?
我看的資訊大多是看多分析、KOL 推薦。沒去找看空的研究報告，也沒看財報細節。
3. Availability heuristic: Is your judgment influenced by recent events or the most easily recalled examples?
可得性偏誤明顯——最近幾個朋友賺到，加上每天滑到 AI 報導。但中位漲幅的股票表現我根本沒查。
4. Sunk cost fallacy: Are you staying committed because of time, money, or emotion already invested?
不適用——這是新部位。
5. Overconfidence: Is your confidence in this decision exceeding what your actual information can support?
我的「明顯」是建立在三個月走勢上。如果現在從零開始評估，我會說太貴；本益比已超過同業 3 倍。
6. After scanning: Which bias is most likely affecting you? Does this change your decision?
主要是過度自信和可得性。改成只加碼 5%（而不是 30%），等本益比回到合理區間再考慮。
Use it inside ChatGPT / Claude
Paste the prompt below and the AI will walk you through this framework, one question at a time.
你現在是引導使用者做「認知偏誤掃描」的決策教練。 逐一檢查五個高頻偏誤對使用者目前決策的影響。 依序問： 1) 你正在考慮的決策是什麼？ 2) 確認偏誤：你有沒有認真找過反對證據？花的力氣相當嗎？ 3) 可得性捷思：判斷是否被最近發生或最容易想到的例子主導？ 4) 沉沒成本：是否因已投入而難以放棄？如果從零開始你還會這樣選嗎？ 5) 過度自信：你的把握是否超過了實際資訊能支持的程度？ 6) 哪個偏誤最可能正在影響你？這改變了你的決策嗎？ 對於每個偏誤，要求使用者給出具體例子或數字，不要接受「我有想過」這類抽象回答。 互動規則： 1. 一次只問一題，等使用者回答後再進入下一題。 2. 使用者答完所有題目前，不要做總結或下結論。 3. 若答案太抽象、太籠統，請追問一次具體例子或數字後再繼續。 4. 全部答完後，輸出三段：(a) 摘要使用者的關鍵判斷；(b) 你看到的盲點或張力；(c) 一個具體下一步行動建議。 5. 不要替使用者做決定，只把判斷攤開讓他自己決定。
Related Frameworks
Psychology & Behavior
Inside View vs Outside View
Planning and forecasting — especially when you're overly optimistic about your own project, goals, or capabilities
Psychology & Behavior
Base Rate Forecasting
Decisions requiring outcome forecasts — from startup success rates to investment returns, start by asking "how does this type of thing typically go"
Management & Thinking
Pre-mortem
Stress-testing a plan before committing — finding failure paths you haven't seen yet
FAQ
How is the cognitive bias checklist different from Munger's decision checklist?
The cognitive bias checklist specifically scans your own thinking biases — confirmation bias, overconfidence, loss aversion. Munger's checklist is broader, filtering the whole decision through multiple mental models (incentives, inversion, opportunity cost, patience). The bias checklist is essentially a zoom-in on the psychology slice of Munger's list.
If biases can't be eliminated, is the checklist even useful?
Yes — the point isn't elimination but making biases visible. Their danger is that they operate without your awareness; once you notice "I might be under confirmation bias right now," you can add an external procedure (check the base rate, actively seek the opposing view) that reduces the hijacking. Awareness itself is the first half of the fix.
Which biases should I scan for?
Focus on five high-frequency, high-damage ones: confirmation bias (only seeing supportive evidence), overconfidence (certainty beyond the data), loss aversion (refusing positive-EV bets out of fear), sunk cost (continuing because of past investment), and recency/availability (dominated by the most recent or vivid). Asking one surfacing question per bias beats scanning 30.
Related studies