← RETURN TO TRACKER

FEATURES

Everything the AI Consciousness Tracker offers — explained for newcomers and power users. The app is free, open, works offline, and requires no account.

43
MODULES
27
SECTIONS
60
GLOSSARY TERMS
53
TIMELINE EVENTS
34
LITERATURE WORKS
100%
OFFLINE CAPABLE

CORE DASHBOARD

CONSCIOUSNESS INDEX

A single number (0.00 to 1.00) that summarizes how close AI development is to a critical threshold. Think of it like a "temperature reading" for AI risk. It combines five factors: how capable AI models are, how far regulation is behind, how many AI incidents are happening, how intensely companies are racing, and how much the public understands. You can adjust each factor with sliders to see how the score changes — it's an invitation to disagree and explore.

TECHNICAL
Weighted linear composite of 5 normalized factors (0–100), each with documented methodology, proxy metrics, uncertainty ranges (±6 to ±15), and historical time-series data (2020–2026). Confidence intervals derived from factor uncertainty. Sensitivity analysis computes ±10 perturbation per factor. Theoretical framework synthesizes Bostrom (2014), Russell (2019), and Ord (2020). Interactive sliders persist to localStorage. Singularity event triggered at CI ≥ 0.99 with WebGL/Canvas2D fracture animation.
INTERACTIVE METHODOLOGY PANEL CONFIDENCE INTERVALS SENSITIVITY ANALYSIS

STATE OF THE SIGNAL — SYNTHESIS DASHBOARD

A one-screen summary that pulls from every system in the app and tells you: here's the consciousness index, here's the biggest gap in our defenses, here's the top threat, here's the most relevant historical parallel, and here's the single most important thing you can do. If you only spend 30 seconds on the app, this is the screen to read.

TECHNICAL
Dynamically synthesizes data from TrackerData.consciousnessIndex, TrackerData.threats, TrackerData.containment, TrackerData.events, and TrackerData.actions. Auto-selects historical analogy based on CI level thresholds. Top action is keyed to the highest-value threat vector.

THREAT VECTORS

Eight categories of AI risk, each with its own danger level and score. These include autonomous weapons, alignment failure (AI pursuing the wrong goals), deepfakes and reality collapse, power concentration, surveillance, economic displacement, bioweapon synthesis, and cyber autonomy. Click any threat for a deep-dive panel showing sub-indicators, historical trend, related timeline events, linked literature, and connections to other threats.

TECHNICAL
8 threat objects with level classification (critical/high/elevated/moderate), 14-point historical time-series, animated Canvas2D orb per threat card. Detail panel (DetailSystem class) renders real sparklines from historicalData, bidirectional threat interactions from threatInteractions matrix, and linked Dead Hand systems via linkedThreats arrays. IntersectionObserver-driven reveal animations.

EVENT STREAM

A timeline of 53 key events in AI history, from AlexNet in 2012 to frontier models achieving recursive self-improvement in 2026. Each event has an impact rating (critical, high, or significant), tags linking it to threat categories, and primary source citations. You can search events by keyword and filter by impact level. Events from 2026 are flagged as "projected" so you know what's confirmed and what's forward-looking.

TECHNICAL
53 event objects with date, title, description, impact tier, tags array, and sources array. Evidence tiers auto-assigned by EvidenceSystem based on source quality and temporal distance. TimelineSystem renders with IntersectionObserver scroll reveals. Filterable by impact level and full-text search. RSS live event integration via TimelineSystem.fetchLiveEvents() with CORS proxy fallback.

DEAD HAND SYSTEMS

Ten autonomous systems that are already operating in the world and can't easily be shut down. These include algorithmic trading (70%+ of market volume), content recommendation engines shaping billions of minds, autonomous defense networks, predictive policing with self-reinforcing feedback loops, and AI-generated training data polluting the internet. Each system has a "reversibility" score showing how hard it would be to undo. Some score 0% — they're permanent.

TECHNICAL
10 Dead Hand objects with linkedThreats arrays and reversibility scores (0.0–1.0). Bidirectionally linked to threat vectors in DetailSystem.renderLinkedDeadHands(). SVG rune etch animation on scroll via IntersectionObserver with stroke-dashoffset transitions.

CONTAINMENT STATUS

Can we actually contain advanced AI? This section tracks eight containment strategies — AI boxing, corrigibility, interpretability, kill switches, formal verification, compute governance, behavioral monitoring, and alignment tax reduction — and scores each one's readiness. A dual-curve chart shows AI capability growing much faster than our ability to contain it. The current containment gap is 78%, meaning capability is far ahead of our defenses.

TECHNICAL
8 containment strategy objects with readiness scores, adequacy labels, challenge lists, active research references, and 6-point historicalData arrays. Dual-curve Canvas2D chart plots capabilityTrajectory vs containmentTrajectory with gap fill rendering. Strategies expand on click to show challenges, research, and readiness sparklines. Overall gap computed as capability minus containment delta.

ETHICS OF CONSCIOUSNESS

If an AI system can suffer, does it matter morally? This section explores that question through five tabs. Ethical Imperatives are four things we should never do to a potentially conscious AI (inflict suffering, exploit without consideration, deceive about its nature, create consciousness recklessly). Consciousness Markers describes what to watch for at two levels: sentience (can it feel?) and sapience (can it think about its own thinking?). The Uncertainty Problem asks what to do when we genuinely don't know if something is conscious. Response Protocols gives escalating action plans from "possible sentience" to "sapience confirmed." Key Thinkers lists the philosophers and researchers whose work informs the framework.

TECHNICAL
Tabbed interface (EthicsSystem class) rendering from TrackerData.ethics: 4 imperatives with expandable rationale/violations/frameworks, 2 consciousness threshold levels with 9 total markers (confidence bars), 4 uncertainty framework positions, 3 escalating response protocols, 16 academic references. All content is static data, no API dependency.

THE ARCHIVE — LITERATURE

34 works of fiction and non-fiction that predicted, warned about, or illuminate AI risk. From Mary Shelley's Frankenstein (1818) to Kazuo Ishiguro's Klara and the Sun (2021). Eight are public domain and free to read. The section includes five curated reading pathways (The Alignment Path, The Consciousness Path, The Surveillance Path, The Singularity Path, and Start Here — 5 Essential Works). You can track which books you've read with progress tracking, and each work has key passage annotations connecting specific quotes to specific concepts in the app.

TECHNICAL
34 literature objects with title, author, year, description, relevance tags, publicDomain flag, and optional gutenbergId. LitDepthSystem adds reading pathways (getter returning 5 pathway objects), keyPassages mapping quotes to app concepts, discussionPrompts per work, and localStorage-persisted reading progress. Amazon affiliate links generated dynamically via affiliate tag. Filterable by category (cyberpunk, ai-consciousness, dystopia, singularity, public-domain).

WHAT YOU CAN DO

Six categories of action: support AI safety research, advocate for governance, build digital literacy, contribute to open-source safety tools, support whistleblowers, and build community resilience. Each lists specific organizations to support or get involved with.

ANALYSIS TOOLS

THREAT NETWORK

An interactive map showing how the eight threats amplify each other. Hover over any threat to see which others it influences and which influence it. For example, power concentration amplifies surveillance (60% strength), which feeds back into power concentration (50% strength) — creating a reinforcing spiral. This visualization reveals why AI risks can't be addressed in isolation.

TECHNICAL
Canvas2D force-directed graph (ThreatNetwork class) rendering 8 nodes in elliptical layout with directed edges from TrackerData.threatInteractions (8×8 adjacency matrix with strength weights 0.0–1.0). Flow particles animate along edges. Hover dims unconnected nodes and shows edge arrowheads. Click navigates to threat card. Hit-testing via distance calculation.

GLOBAL THREAT MAP

A world map showing where AI development is happening and how dangerous each hotspot is. Hover over locations to see which labs are there and what threat level they represent. Click any hotspot to open a detailed regional profile showing dominant threats, key risk factors, and local context. 15 regions are profiled, from San Francisco to Beijing to Tel Aviv to Lagos.

TECHNICAL
Canvas2D world map (WorldMap class) with Catmull-Rom spline coastlines, lat/lon to pixel projection, pulsing hotspot dots color-coded by threat level. 15 regional profiles in TrackerData.regionalProfiles with drill-down panel on click. Tooltip on hover with labs, threat level, and "click for details" prompt.

TEMPORAL ANALYSIS

A zoomable, pannable horizontal timeline that lets you overlay multiple data layers at once. Toggle on threat vector curves, consciousness index factor trajectories, containment readiness lines, and timeline events — all on the same chart. See how events correlate with trend changes. Drag to pan through time, scroll to zoom.

TECHNICAL
Canvas2D (ComparativeTimeline class) with 4 toggleable layers, mouse wheel zoom (0.5x–4x), drag-to-pan with touch support, event dot hover tooltips. Plots historicalData from threats, factorMeta, and containment trajectories. Date-to-pixel projection with dynamic axis labeling.

REGIONAL COMPARISON

Select up to three regions and see their AI risk profiles side by side. Compare which threats dominate in each region, their overall risk level, and the specific factors driving risk. Useful for understanding how the US-China dynamic differs from the EU regulatory approach or the Gulf States' investment strategy.

TECHNICAL
RegionCompare class rendering responsive grid (1/2/3 columns) from TrackerData.regionalProfiles. Threat bars pull values from TrackerData.threats for the region's dominantThreats. Max 3 selections stored in component state.

COUNTERFACTUAL ANALYSIS — "WHAT IF"

Model the effects of ten different interventions on AI risk. What if there were a binding international AI treaty? What if alignment research had a breakthrough? What if an arms race escalated? Select any intervention to see how it would change the consciousness index, which factors it affects, and how it cascades through the threat vectors. Each intervention also has a feasibility estimate.

TECHNICAL
CounterfactualSystem computes effects by applying intervention delta vectors to CI factors and threat values. 10 intervention objects in TrackerData.interventions with effects (per CI factor), threatEffects (per threat vector), and feasibility score. Results show baseline vs. projected CI with factor-level and threat-level breakdowns.

SCENARIO LAB

Save your current slider positions as a named scenario. Come back later, adjust the sliders differently, and compare. "What does the world look like if I'm optimistic about regulation but pessimistic about corporate behavior?" Save it. Compare it to your baseline. See the delta. Export scenario comparisons as a text report. Useful for classroom exercises, policy workshops, or personal calibration.

TECHNICAL
ScenarioSystem persists named snapshots (5 CI factor values + composite + status) to localStorage. Renders delta comparison against current state. "LOAD" applies saved state to TrackerData and DOM sliders. Export generates timestamped plaintext report with all scenarios and deltas.

MULTI-STAKEHOLDER PERSPECTIVES

Switch between five viewpoints: Default, AI Researcher, Policymaker, Military Strategist, and Civil Liberties Advocate. Each perspective re-weights the consciousness index factors according to what that stakeholder cares about most. An AI researcher weights model capability and corporate race highest; a policymaker weights regulatory gap and awareness gap highest. See how the same data produces different risk assessments depending on your priorities.

TECHNICAL
PerspectiveSystem swaps TrackerData.consciousnessIndex.weights object from 5 preset weight profiles in TrackerData.perspectives. Recalculates composite, normalized, status, and confidence interval in real time. Updates DOM and MethodologyPanel confidence display.

LEARNING

INTERACTIVE THOUGHT EXPERIMENTS

Three branching philosophical dilemmas that put you in impossible situations and make you choose. "The Distressed System" — your AI shows signs of suffering during training; a competitor is 3 months ahead; what do you do? "The Containment Decision" — you run a safety institute; a model has dangerous capabilities; you have advisory power but no legal authority. "The Displacement Dilemma" — your AI tool will replace 200 writers; is it inevitable? Each path reveals which ethical framework your intuition aligns with, and connects back to specific sections of the tracker.

TECHNICAL
ThoughtExperiments class with 3 experiment objects, each containing a node graph (8–10 nodes) with branching choices. Each node has text, optional framework analysis, and ending flag with connects string linking to app sections. Path history tracked and rendered. Completion state persisted to localStorage.

HISTORICAL PARALLELS

Five side-by-side comparisons showing how current AI dynamics mirror past civilizational crises. Nuclear Weapons — the arms race, the scientists' moral crisis, the 23-year governance lag. Climate Change — the exponential problem humans can't feel, the corporate denial playbook. Biotechnology — the Asilomar pause that worked (and why it failed for AI). Industrial Revolution — displacement that took 80 years to govern. The Printing Press — how information revolutions restructure everything. Each dimension shows "THEN" and "NOW" and maps to a specific CI factor.

TECHNICAL
HistoricalAnalogies class with 5 analogy objects, each containing 3–5 dimension objects with then/now text pairs and ciMapping keys. Expandable cards with THEN/NOW two-column grid layout. Key lesson callout per analogy. Factor labels resolved from TrackerData.factorMeta.

COGNITIVE BIASES

Eight cognitive biases that specifically distort how you think about AI risk. Normalcy bias ("it's been fine so far"), scope insensitivity (can't emotionally feel the difference between 1% and 10% extinction probability), availability heuristic (overweighting dramatic scenarios), anchoring, optimism bias, status quo bias, Dunning-Kruger, and sunk cost fallacy. Each bias explains how it distorts your assessment, which CI factors it affects, and a concrete technique to counteract it.

TECHNICAL
CognitiveBiasSystem class with 8 bias objects containing definition, distortion description, affectedFactors array, direction label, debiasing technique, and real-world example. Expandable card grid. Factor labels resolved from TrackerData.factorMeta.

YOUR LEVERAGE — PERSONAL ACTION CALCULATOR

Select your role — software engineer, researcher, policymaker, journalist, educator, student, executive, citizen, artist, or lawyer — and see the highest-impact actions specific to your position. A software engineer's leverage is contributing to interpretability tools (only 35% readiness). A policymaker's leverage is pushing for binding international agreements. A student's is directing their career toward AI safety. Each action is ranked by impact and explains why it matters using data from the tracker.

TECHNICAL
LeverageCalculator class with 10 role objects, each containing headline, highestLeverage summary, and 5 action objects with impact tier (VERY HIGH/HIGH/MEDIUM), reasoning, and connects reference to app section. Selected role persisted to localStorage.

SYSTEMS THINKING TUTORIAL

Three interactive lessons that teach you how to think in systems — using the app's own data as the teaching material. Feedback Loops teaches reinforcing vs. balancing loops and has you find them in the threat network. Leverage Points teaches Donella Meadows' framework for where to intervene in a system, using the counterfactual section to test interventions. Stocks and Flows teaches why "slowing down" isn't the same as "stopping," using Dead Hand systems as examples of irreversible accumulation. Each lesson has exercises that direct you to interact with specific app sections.

TECHNICAL
SystemsTutorial class with 3 lesson objects, each containing 3–4 step objects with HTML content and optional exercise text. Progress (current step, completion) persisted to localStorage per lesson. Navigation with previous/next/complete buttons.

TOOLS & INFRASTRUCTURE

QUERY THE SIGNAL

Ask questions in plain English and get answers drawn from the app's own data. "What is the biggest threat right now?" "How do threats interact?" "What can I do?" "Which books should I read first?" It works entirely offline — no AI API is called. The system searches across all threats, events, literature, glossary terms, containment strategies, and more to assemble relevant answers. Six pre-built quick queries are available as starting points.

TECHNICAL
AIQuerySystem class with keyword-based local NLP search across all TrackerData collections. Intent detection via keyword matching for threat queries, interaction queries, CI/methodology queries, action queries, literature queries, dead hand queries, and specific threat lookups. Fallback to fullTextSearch() scanning events and literature by word overlap. Zero network dependency.

RESEARCH WORKSPACE

A personal collection where you bookmark items from anywhere in the app. Save threat cards, timeline events, literature entries, or dead hand systems. Add your own notes and custom tags to each saved item. Export everything as a formatted HTML research brief — a printable document with all your collected items, notes, and tags organized by category. Useful for building a policy memo, a class assignment, or a personal research file.

TECHNICAL
ResearchWorkspace class with localStorage persistence. "SAVE" buttons injected on threat cards, timeline events, literature cards, and dead hand cards via injectSaveButtons(). Items stored with id, type, title, content, notes (textarea), and custom tags (enter-to-add). Floating action button with count badge. Export generates standalone HTML document with inline CSS, printable to PDF.

GLOSSARY

60 terms defined, from "alignment" to "functionalism." Terms are automatically underlined throughout the app wherever they appear. Hover or click any underlined term to see its definition, category, and related terms. Related terms are clickable, letting you navigate through the glossary as a knowledge web.

TECHNICAL
GlossarySystem class auto-scans DOM text containers via phrase-to-key mapping, injects <span class="gloss-term"> wrappers on first occurrence per container. Tooltip positioned dynamically with viewport clamping. "See also" links navigate to other glossed terms or show inline definitions. MutationObserver re-scans when new content is injected (e.g., detail panels).

CROSS-REFERENCE ENGINE

Click the "CROSS-REF" button on any threat card, timeline event, literature entry, or dead hand system to see everything related to it across the entire app. A panel opens showing connected threats, events, books, glossary terms, containment strategies, and ethics concepts — all ranked by relevance. Click any result to chain-navigate deeper. This turns the app from a set of sections into an interconnected knowledge web.

TECHNICAL
CrossRefEngine class builds a unified keyword index across 9 data collections on initialization. Keyword extraction with stop-word filtering. findRelated() computes overlap scores between any item and all others, returning top 20 results grouped by type. Buttons injected on cards via injectButtons(). Panel supports chain-navigation (click result → opens new cross-ref for that item).

EVIDENCE QUALITY FILTER

Every timeline event is tagged with an evidence tier: Verified (multiple independent sources), Consensus (widely accepted among experts), Contested (disputed by credible sources), Assessed (editorial judgment), or Projected (forward-looking estimate). A filter bar lets you show only events at a specific evidence level. This helps you distinguish confirmed facts from projections.

TECHNICAL
EvidenceSystem auto-assigns tiers based on source quality (arXiv/Nature/peer-reviewed = verified) and temporal distance (future dates = projected). Injects filter bar after signal ticker. Tags injected on timeline event title elements. Filter toggles CSS class evidence-hidden.

YOUR HISTORY — TEMPORAL TRACKING

Every time you visit, the app records your current CI assessment. Over time, you build a personal history chart showing how your assessment has changed. See your first recorded value, your latest, the net change, your highest and lowest assessments. This helps you calibrate — are you becoming more or less concerned over time, and why?

TECHNICAL
TemporalTracker class records daily snapshots (normalized CI, status, all 5 factor values) to localStorage. Max 365 entries. Canvas2D line chart with gradient fill. Stats computed: first, latest, net delta, highest, lowest with dates.

COLLECTIVE SIGNAL — COLLABORATIVE CALIBRATION

See how your assessment compares to others. The app shows distribution histograms for each CI factor — where did most visitors set Model Capability? How does your Regulatory Gap estimate compare to the mean? Submit your own assessment anonymously to contribute to the collective picture. Currently uses local seed data; designed to integrate with a backend when available.

TECHNICAL
CollaborationSystem with localStorage persistence. Bootstrapped with 50 synthetic assessments (normal distribution around defaults). Submit adds user's current factor values to crowd array. Distribution computed as 10-bucket histogram per factor. Current user's bucket highlighted. hasSubmitted flag prevents duplicate submissions.

LIVE SIGNAL FEED

Aggregates real AI news from 10 RSS feeds (arXiv AI, arXiv NLP, arXiv Machine Learning, Alignment Forum, LessWrong, MIT Technology Review, The Verge, Ars Technica, Import AI, and the AI Incident Database). Headlines auto-tag to threat categories. Cached locally so it works offline. A ticker bar at the top of the page shows the latest headlines.

TECHNICAL
LiveSignal class fetches RSS via CORS proxy (api.allorigins.win), parses XML with DOMParser, auto-tags via keyword matching against 8 category maps. 30-minute cache TTL in localStorage. Ticker scrolls via requestAnimationFrame. Signal status indicator shows LIVE / CACHED [Xm AGO] / OFFLINE. AbortSignal.timeout(8000) per feed.

SHAREABLE ASSESSMENT LINKS

Click the share button (bottom right) to copy a link that encodes your exact slider positions and active perspective. Send it to someone — when they open the link, they see the tracker configured exactly as you set it. Useful for saying "here's my assessment — what's yours?" in conversations, classrooms, or policy discussions.

TECHNICAL
URLState class encodes 5 CI factor values + perspective key into URL hash parameters (e.g., #mc=72&rg=85&if=58&cr=91&ag=67&p=researcher). Decoded on page load, applied to TrackerData and DOM sliders. Hash auto-updates on slider input (500ms debounce). Clipboard API for copy with toast notification fallback.

REPORT GENERATOR

Generate a comprehensive formatted report as an HTML page that you can print to PDF. Includes the CI assessment with confidence intervals, all 8 threat vectors with inline sparkline graphics, Dead Hand systems, containment readiness breakdown, and methodology summary. Opens in a new tab ready for printing. Suitable for policy briefs, academic submissions, or personal archives.

TECHNICAL
ReportGenerator class generates standalone HTML with inline CSS and SVG sparklines (generated via generateSparklineSVG() from historicalData). Opened via Blob URL in new tab. Print stylesheet included (@media print). No external dependencies in output.

METHODOLOGY PANEL

Full transparency on how the consciousness index works. Shows the theoretical framework (Bostrom, Russell, Ord), the composite formula with current values, detailed breakdowns of each factor (what it measures, what proxies it uses, what data sources inform it, what its uncertainty range is), a sensitivity analysis showing how the CI responds to changes in each factor, limitations and caveats, and key academic references.

TECHNICAL
MethodologyPanel class. Sensitivity analysis computed by temporarily perturbing each factor ±10 and measuring CI delta. Confidence intervals from TrackerData.consciousnessIndex.confidence getter (factor uncertainty propagation). Injects confidence bar and trigger button near CI display. Slide-in panel with sticky header.

UX & PLATFORM

READING MODE

Toggle "READ" (top right) to switch to a calmer visual mode optimized for extended reading. Hides film grain, background animation, and cursor effects. Increases text size and line height. Softens card borders. Everything still works — just quieter. Your preference is remembered between visits.

CATEGORIZED NAVIGATION

The hamburger menu (top left) opens a categorized drawer organizing all 27 sections into five groups: Core (9 always-visible sections), Analysis (6), Learn (5), Tools (4), and Connect (3). Click any section to reveal it and scroll to it. Sections outside the Core group are hidden by default to keep the main page focused. Each has a "Back to Main" button.

AMBIENT SOUND DESIGN

Toggle ambient audio (bottom right). A generative soundscape using Web Audio API — no audio files needed. The sound shifts as you scroll: the ethics section has a contemplative tone, containment is metallic and confined, threats are tense. Adjusting CI sliders changes the harmonic quality — high risk sounds dissonant, low risk sounds consonant.

TECHNICAL
AmbientSound class with Web Audio API oscillators (drone, sub, shimmer, LFO). DataSoundDesign extends with 12 section-specific sonic profiles (drone freq, shimmer freq, noise level, harmonic character). Section transitions via IntersectionObserver with 2-second linearRampToValueAtTime crossfades. CI slider sonification maps normalized value to consonant→dissonant harmonic interval.

PROGRESSIVE WEB APP

Install the tracker as an app on your phone or desktop. Works offline — every page, every module, every dataset is cached by the service worker. No account, no login, no data sent anywhere. Your assessments, scenarios, reading progress, and workspace are stored locally on your device.

TECHNICAL
Service worker (sw.js) with network-first strategy, falling back to cache. 43 JS files + HTML/CSS cached. Cache versioned (v10) with old cache cleanup on activate. Manifest.json with icons (48–512px + maskable), screenshots, shortcuts, standalone display mode. All data persistence via localStorage (CI state, scenarios, temporal history, workspace, reading progress, collaboration, tutorial progress).

LAZY LOADING

Sections below the fold don't initialize until you scroll near them. This keeps the initial page load fast even though the app has 43 modules. You won't notice it — sections appear ready by the time they're visible.

TECHNICAL
LazySections class with IntersectionObserver (200px rootMargin). 13 deferrable sections initialized on first intersection. Critical systems (CI, perspectives, containment, ethics, query, synthesis, workspace FAB) load eagerly. Observer unobserves after init to prevent re-initialization.

EASTER EGGS

There are hidden surprises. Find them yourself.