availability-heuristic

name

availability-heuristic

description

Activate when: user says 'we keep hearing about X so it must be common', 'that just happened so it's risky', 'the news is full of stories about this', 'I've seen this a lot lately so it's probable', or is making a risk/frequency estimate driven by memorable examples rather than data. Do NOT activate when: the available evidence is genuinely representative of the reference class; the decision warrants heavy precautionary weight on rare but catastrophic/irreversible risks. More: deciqai.com/s/availability-heuristic

Availability Heuristic

Overview

The availability heuristic: people estimate probability by how easily examples come to mind. Vivid, recent, and media-reported events are systematically overestimated; routine, statistical events are underestimated. Five triggers: recency, vividness, media coverage, personal experience, imaginability. Structural fix: reference-class forecasting — answer from base-rate data, not memory.

Composes with bayesian-reasoning, probabilistic-thinking, anchoring, survivorship-bias, framing-effect.

When to Use

• Someone estimates probability from memorable examples rather than data
• Risk perception driven by recent news, vivid anecdotes, or media coverage
• Team is preparing for the last disaster rather than the most likely next one
• Investment or resource-allocation decision follows a recent dramatic event
• Medical, legal, or technical decision made by anecdote rather than reference class
• AI capex, AI valuations, or AI-adoption/displacement estimates are driven by a viral demo, a dramatic layoff headline, or a safety scare rather than deployment and labor base rates

Not when: available evidence is a representative sample; decision warrants precautionary weight on rare catastrophic risks; no reference-class data exists.

Coaching Novices (Adaptive Front Door)

• Engine mode: user has a specific risk or frequency estimate → run The Process directly.
• Coach mode: user is new → guide step by step.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.

1. One-line: before trusting an intuitive probability, ask whether it is based on retrieval ease or on a reference class with actual data.
2. Check fit: if recent vivid evidence is also representative, use it; otherwise discount and seek base rates.
3. Elicit their real case — what probability is being claimed, on what basis?

[WAIT — do not advance until user responds]

4. Run The Process one step at a time: reference class → base rate → distortion diagnosis → compare to intuitive estimate.

[WAIT — do not advance until user responds]

5. Close: restate corrected estimate as base-rate + specific adjustment, with the adjustment justified.

[WAIT — do not advance until user responds]

The Process

Step 1 — Specify the estimate: probability/frequency being estimated · current intuitive estimate · basis (memory/news/data) · decision that depends on it.

Step 2 — Identify the reference class: population of comparable cases · size · time horizon · inclusion/exclusion criteria. (Getting this wrong is the main failure mode.)

Step 3 — Get the base rate: actual frequency in the reference class · source · confidence level.

Step 4 — Diagnose availability distortions: recency (is the recent rate representative?) · vividness (dramatic vs. routine?) · media coverage (over-reported?) · personal experience (inflated?) · imaginability (easy to picture → feels more probable?).

Step 5 — Compare: intuitive estimate vs. base rate. If ratio >2×, availability is doing significant work. Adjust toward base rate; justify any residual deviation.

Step 6 — Document: corrected probability · reasoning (base rate × adjustments) · decision implications · calibration log entry.

Output Template

Availability Correction: <event>
Original estimate: <value> | Basis: | Decision:
Reference class: <population, size, time horizon>
Base rate: <frequency> | Source: | Confidence:
Distortions: recency / vividness / media / personal / imaginability
Corrected estimate: <value> | Adjustment rationale:
Decision implications: | Calibration log:

→ Method in Action: Tversky and Kahneman's 1973 Availability Studies · Post-9/11 Flight Avoidance and Road Fatalities

→ 2026 lens: Vivid AI Stories vs. Base Rates (2023–2026)

Pack: Availability Bias Patterns

Domain	Distortion	Correction
Personal risk	Overestimate terrorism/plane crashes; underestimate heart disease/car crashes	CDC/BLS mortality statistics
Investment	Recent winners feel certain to continue	Long-horizon data; mean reversion priors
Hiring	Memorable candidate beats better-qualified one	Structured rubric; reference-class hire data
Project planning	Optimism from vivid current plan vs. project-class history	Flyvbjerg reference-class forecasting
Medical diagnosis	Recently-seen diagnosis over-weighted	Base-rate-weighted differential diagnosis
Strategic planning	Recent competitor event dominates outlook	Multi-year base rates; explicit reference class

Applying It Well

• Define the reference class before looking up the base rate. Calibration improves with logged predictions vs. outcomes, not exhortation. Present absolute frequencies (deaths per 100,000), not relative risk.

→ Primary sources: references/sources.md

Common Rationalizations

[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.

Fake move	Reality
[D] "But that just happened!"	Recency aids recall; it doesn't increase future probability. Check the longer reference class.
[D] "I've seen this many times"	Easy retrieval ≠ high population frequency.
[D] "The news is full of stories about this"	Coverage tracks vividness, not frequency.
[D] "I'd remember if it were rare"	You remember dramatic-rare more than statistical-common.
[D] "I have personal experience"	Personal examples are a tiny, often unrepresentative sample.
[D] "The risk is too vivid to ignore"	Vividness ≠ probability. Compute actual probability; respond proportionally.
[D] "We have to prepare for the worst"	Sometimes valid. Often over-preparation for spectacular-low-probability events.
[D] "The base rate is irrelevant — this case is different"	Be specific about which factors actually move the estimate.
→ Add [O] entries here after each real use — paste the actual failure pattern	What went wrong and why

Red Flags

• Risk estimate made without consulting actual data
• Single-case vivid examples or recent news are the dominant evidence
• Coverage volume treated as evidence of frequency
• Reference class not explicitly identified

Verification

• Intuitive estimate stated with confidence level
• Reference class identified
• Base rate retrieved from a defensible source
• Availability distortions (recency, vividness, media, personal experience) considered
• Corrected estimate documented with reasoning
• Decision implications computed
• Estimate logged for later calibration

---

Part of deciqAI Knowledge Skills — 228 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. See it run → https://www.deciqai.com/s/availability-heuristic · Built by deciqAI · github.com/deciqAI · Contributions welcome.

Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/availability-heuristic.json