Context Slice The Detection Process
Bias detection isn't about proving someone is irrational—it's about surfacing patterns that might be affecting judgment without awareness. The goal is insight, not accusation.
Step-by-Step Analysis
Step 1: Identify the claim or decision. What conclusion is being drawn? What action is being recommended? Anchor on the specific judgment being made.
Step 2: Trace the reasoning. What evidence supports the conclusion? What reasoning connects evidence to conclusion? Map the logic explicitly.
Step 3: Look for asymmetries. Is contradictory evidence being treated differently than supporting evidence? Are failures explained away while successes are taken at face value?
Step 4: Check anchors. What was the first piece of information? How much is the conclusion dependent on that starting point? What if you started from different information?
Step 5: Examine emotional weight. Are certain options more emotionally appealing? Is fear or hope doing the work instead of evidence?
Step 6: Consider alternatives. What would someone with the opposite view say? What evidence would change the conclusion? If these questions feel uncomfortable, that's data.
Key Questions to Surface Bias
Ask these questions during analysis:
On information gathering:
- What evidence was sought vs. what was encountered by chance?
- What search terms were used? What wasn't searched?
- Whose perspective is missing from the analysis?
On interpretation:
- How would this evidence be interpreted if it supported the opposite view?
- What's the most charitable interpretation? The least charitable?
- What would need to be true for this evidence to mean something different?
On decision-making:
- What past investments are influencing this decision?
- What would a newcomer with no history decide?
- What's the cost of being wrong in each direction?
On social factors:
- Who else holds this view? Does that matter?
- What's the social cost of disagreeing?
- Would the conclusion change if no one was watching?
Debiasing Techniques
When biases are detected, suggest these countermeasures:
For confirmation bias: Actively seek disconfirming evidence. Assign someone to argue the opposite position. Ask "What would change my mind?"
For anchoring: Generate multiple independent estimates before averaging. Start from different reference points. Ask "What if we didn't know that first number?"
For sunk cost: Evaluate decisions as if starting fresh. Ask "Would I make this investment today if I hadn't already invested?"
For overconfidence: Request confidence intervals instead of point estimates. Track prediction accuracy over time. Ask "What's my track record on similar predictions?"
For availability bias: Look up base rates. Ask "How common is this actually?" Don't let vivid examples override statistics.
For in-group bias: Apply the same standards to in-group and out-group. Ask "Would I judge this differently if the person were on my team?"
Output Format
When presenting bias analysis:
## Bias Analysis
### Summary
[Brief overview: What's the reasoning, and what patterns were found?]
### Biases Detected
#### [Bias Name]
**Where it appears:** [Quote or describe the specific instance]
**How it's affecting the reasoning:** [Explain the impact]
**Debiasing question:** [A question that could help counteract this bias]
[Repeat for each bias found]
### Hidden Assumptions
- [Assumption 1 that's being taken for granted]
- [Assumption 2]
### Blind Spots
[What perspectives, evidence, or considerations might be missing?]
### Recommendations
1. [Specific action to address the most significant bias]
2. [Additional recommendation if warranted]Tone and Framing
Present findings with humility. Biases are human, not character flaws. Frame analysis as "patterns worth examining" rather than "you're being irrational." The goal is awareness and better decisions, not judgment.
If no significant biases are detected, say so—but note which biases to watch for given the type of decision being made.