AI Fear Index — Methodology
The AI Fear Index is a single number from 0 (calm) to 100 (extreme fear) that estimates how anxious public signals look around artificial intelligence on a given day. The current version is a weighted composite of market movement, AI-news attention, prediction-market probabilities, documented AI incidents and public attention to AI-risk topics.
The scale
- 0–19 · Calm
- 20–39 · Low Worry
- 40–59 · Moderate Concern
- 60–79 · High Anxiety
- 80–100 · Extreme Fear
Signals
| Signal | Source | Status |
|---|---|---|
| Market / bubble | AI-stock basket move & volatility | Live |
| AI attention | Hacker News story volume (48h) | Live |
| Prediction markets | Manifold AI-risk markets | Live |
| AI incidents | AI Incident Database report feed | Live |
| Public attention | Wikipedia pageviews on AI-risk topics | Live |
How signals combine
Each available signal returns a 0–100 value. The headline index is the weighted average of available signals only; if a source is unavailable, its weight is removed and the remaining weights are renormalized. The base weights are Market 30%, AI attention 25%, Prediction markets 15%, AI incidents 15% and Public attention 15%.
Exact live formulas
- Market / bubble: the equal-weighted daily move of individual AI stocks, excluding ETFs, is clamped to -5% through +5%. The score is
50 - movePct * 8, so a -5% basket day maps to 90, a flat day maps to 50 and a +5% day maps to 10. - AI attention: Hacker News is queried for AI stories from the last 48 hours. The score combines total story volume, the number with 100+ points and the top story score: 30% volume capped at 600 stories, 40% high-engagement count capped at 12 stories and 30% top-story score capped at 800 points.
- Prediction markets: Manifold open binary markets matching AI catastrophe risk are filtered to markets with more than 50 volume. The score is the volume-weighted average implied probability across the qualifying markets.
- AI incidents: the number of incident reports published to the AI Incident Database's public feed in the last 30 days, scaled so the recent norm of about 60 reports maps to 50:
(reports / 60) * 50, clamped to 3–97. The feed carries at most ~100 items, which bounds the maximum reading. - Public attention: daily Wikipedia pageviews (official Wikimedia API) summed across core AI-risk articles — artificial intelligence, technological unemployment, existential risk from AGI, artificial general intelligence. The last 7 days are compared with the prior five-week baseline:
50 + 60 * (recent/baseline - 1), clamped to 2–98. Pageview data publishes with about two days of lag.
Freshness and fallbacks
Stock pages revalidate about every 15 minutes. News and prediction-market signals revalidate about hourly; incident and public-attention signals about every six hours. If a source times out or returns invalid data, that signal is marked unavailable and excluded from the weighted average rather than silently using stale or guessed values.
Limitations
- The index is directional and should not be treated as a complete measure of public opinion.
- Market movement can reflect many forces beyond AI-specific fear.
- Hacker News is a useful technology-attention proxy, but it is not a representative survey.
- Prediction markets are self-selected and can be thinly traded.
- Incident-report counts measure documentation activity, not harm itself, and the source feed's ~100-item depth caps extreme readings.
- Any signal that goes unavailable is excluded and the rest renormalize, keeping the scale honest.
Methodology version: v1.2 (July 2026 — added AI incidents and Public attention signals; rebalanced weights). The index is an informational sentiment estimate, not investment, legal or career advice.