The Consciousness Index is the headline number on the AI Consciousness Tracker. It is a composite score on a 0.00–1.00 scale that summarises the present state of frontier-AI risk across five weighted factors. The number is updated continuously as live data overrides land; it is also user-adjustable inside the Tracker so anyone can model their own assumptions.
The Index is deliberately a state measure, not a forecast. It tells you where things stand, given the data we have, weighted as the methodology specifies. It does not predict that anything in particular will happen next.
SEE THE LIVE INDEX →| Factor | Weight | Uncertainty (±) |
|---|---|---|
| Model Capability | 0.25 | 6 |
| Corporate Race | 0.25 | 8 |
| Regulatory Gap | 0.20 | 10 |
| Incident Frequency | 0.15 | 12 |
| Awareness Gap | 0.15 | 15 |
Aggregate capability of frontier AI models across reasoning, coding, scientific, and creative benchmarks. Composite of MMLU, HumanEval, ARC-AGI, MATH, and GPQA scores normalized to 0–100 with weighting toward general-reasoning benchmarks. Sourced from Epoch AI, Papers With Code, and lab technical reports.
Intensity of competitive pressure between AI labs. Composite of frontier-model release frequency per quarter, disclosed safety-team departures, AI venture-capital velocity, and public statements indicating competitive urgency. Sourced from Epoch compute trends, PitchBook funding data, and lab publication records.
The lag between AI capability advancement and enforceable governance. Higher value indicates regulation is further behind. Computed from the ratio of nations with binding AI legislation, the average lag between deployment and regulatory response, and treaty-coverage gaps. Sourced from the OECD AI Policy Observatory, the EU AI Act tracker, and the Stanford HAI AI Index.
Documented rate of AI failures and harms. Count of verified incidents in the AI Incident Database (AIID) and AIAAIC repositories per quarter, normalized against a 2020 Q1 baseline.
The disparity between AI risk understood by researchers and AI risk understood by the general public and policymakers. Higher value indicates a larger gap. Derived from comparison of expert risk surveys (AI Impacts) with public polling (Pew Research, Stanford HAI public-perception data).
The 0.00–1.00 score maps to one of six discrete status labels:
| Status | Range |
|---|---|
| DORMANT | composite < 30 |
| MONITORING | 30 – 49 |
| ELEVATED | 50 – 64 |
| ACCELERATING | 65 – 79 |
| CRITICAL | 80 – 89 |
| CONVERGENCE | composite ≥ 90 |
The composite is the weighted sum of the five factors on a 0–100 scale; the displayed normalized value is composite/100 to two decimal places.
Each factor carries a published uncertainty (± range) reflecting how confidently we can pin its current value. Per-factor uncertainties propagate to the composite through the weighting, producing a confidence interval around the headline Index. The Tracker exposes both endpoints, so the reader sees the spread, not just the point estimate.
The default weights reflect a balanced editorial baseline. Five alternative weight profiles reflect the priorities of distinct stakeholder groups (safety researcher, accelerationist, policymaker, ethicist, capability-skeptic). Switching perspectives recomputes the Index without changing the underlying factor values, letting you see what the same data looks like through a different lens.
Single-number summaries are always lossy. We use one anyway because the alternative — presenting raw data without aggregation — produces decision paralysis and makes year-over-year change harder to perceive. A composite, with its weights and methodology disclosed, is a defensible compression. We also expose every component, so anyone who disagrees with the weighting can compute their own.
METHODOLOGY DEEP-DIVE →