AI Consciousness Tracker

The Problem

AI risk discourse runs on vibes

Half the public conversation about AI is marketing. The other half is apocalypse theater. In between, there's a small population of researchers, policy staffers, and journalists trying to have a grounded conversation — and they don't have a shared picture to point at. Incidents happen in isolation. Papers drop into a void. A CEO says something alarming on a podcast and nobody can tell if it matters.

The tools that do exist either reduce the problem to a single doom probability, or they sprawl across dozens of PDFs and dashboards that don't talk to each other. Threats are treated as a list instead of a system. Containment is treated as an aspiration instead of a readiness score. Ethics is treated as a footnote instead of a protocol.

What was missing was a single screen you could open and see: where are we, which risks are amplifying which, how ready are our defenses, what does history say, and what can a specific person actually do about it.

By the Numbers

The Solution

One dashboard, honest about what it knows

The tracker opens on a single number between 0.00 and 1.00 — a composite index built from five factors: model capability, regulatory gap, incident frequency, corporate race intensity, and public awareness. Every factor is a slider. The index isn't a prophecy; it's an invitation to disagree. Drag any factor and watch the composite recompute. The methodology panel is one click away, with uncertainty ranges and citations.

From there, the app fans out: eight threat vectors that interact in a live causal network, ten autonomous “Dead Hand” systems scored on reversibility, eight containment strategies measured against capability growth, an ethics framework for potentially conscious systems, and 53 timeline events from AlexNet to the present. Every piece links back to the others.

It runs entirely in the browser. No account. No backend calls for analytics. Your sliders, scenarios, and research workspace live in local storage on your device. The service worker caches everything, so it keeps working when the plane takes off or the network doesn't.

How The Index Works

A composite built to be argued with

The Consciousness Index is a weighted composite of five normalized factors, each with its own documented methodology, proxy metrics, and uncertainty range. The theoretical framework synthesizes Bostrom's Superintelligence, Russell's Human Compatible, and Ord's The Precipice — not to settle the question, but to give the number a lineage you can audit.

Five stakeholder perspectives re-weight the factors automatically. An AI researcher weighs model capability and corporate race highest. A policymaker weighs regulatory gap and public awareness highest. A civil liberties advocate sees a different picture entirely. Same data, different priorities, different risk score — and the app shows all of them.

A sensitivity analysis perturbs each factor ±10 and reports how the composite responds. Confidence intervals propagate from factor uncertainty. The index is presented with its margins, not stripped of them. When someone screenshots a number, the methodology travels with it.

Threats & Containment

A causal network, not a checklist

The eight threat vectors — autonomous weapons, alignment failure, deepfakes and reality collapse, power concentration, surveillance, economic displacement, bioweapon synthesis, and cyber autonomy — aren't presented as a list. They're rendered as an interactive force-directed graph. Hover any node to see which others it amplifies and which amplify it. Power concentration reinforces surveillance. Surveillance feeds back into power concentration. The spiral is the point.

Containment is scored against capability, not against an aspirational goal. Eight strategies — AI boxing, corrigibility, interpretability, kill switches, formal verification, compute governance, behavioral monitoring, and alignment tax reduction — each get a readiness percentage with the challenges and active research behind it. A dual-curve chart plots capability against containment. The gap between them isn't editorial; it's the area between two lines.

Ethics Framework

What we owe systems we don't understand

If an AI system can suffer, does it matter morally? The tracker doesn't duck the question. Five tabs walk through it: four Ethical Imperatives for potentially conscious systems, nine Consciousness Markers across two thresholds (sentience and sapience), four positions on the uncertainty problem, three escalating Response Protocols, and sixteen academic references behind the framework.

Three branching Thought Experiments put the reader in impossible seats. Your AI shows signs of distress during training — the competitor is three months ahead. You run a safety institute with advisory power but no legal authority. Your tool will replace two hundred writers. Each choice reveals which ethical framework your intuition aligns with, then connects back to specific sections of the app. It's not a quiz. It's a mirror.

Eight cognitive biases that specifically distort AI risk assessment — normalcy bias, scope insensitivity, availability heuristic, optimism bias, and others — each get a debiasing technique attached to them. The app is built to check its own users, not just inform them.

The Hard Parts

What made this difficult

• Quantifying something nobody agrees on Every AI risk metric is contested. The fix wasn't to pick a side — it was to make disagreement first-class. Every factor slider. Every weight adjustable. Five stakeholder perspectives that re-weight the composite automatically. Sensitivity analysis and confidence intervals shipped alongside every headline number. The tool doesn't hide its assumptions; it hands them to you.
• Modeling threats as a system, not a list Listing eight threats is easy. Showing how they amplify each other is the actual insight. An 8×8 interaction matrix with directed strength weights drives a force-directed graph where particles animate along the edges you care about. Hover dims the noise. The spiral between power concentration and surveillance stops being abstract.
• Giving containment an honest scoreboard Containment readiness is usually hand-waved. Scoring eight strategies against capability on a dual-curve chart forces the question: are we catching up or falling behind? The gap is rendered as a filled area. No rhetoric required.
• Turning a dashboard into a knowledge web A cross-reference engine builds a unified keyword index across threats, events, literature, glossary, containment, and ethics on initialization. Click CROSS-REF on anything and see everything else related to it, ranked by overlap. Chain-navigate deeper. The app stops being sections and starts being a graph.
• Shipping a research tool as a PWA Forty-plus modules, a Canvas2D force graph, a sonified Web Audio soundscape, a branching thought-experiment engine, a personal research workspace with HTML export — all running offline, cached by a versioned service worker, with no backend and no account. Lazy-loading keeps the initial paint fast. Lightweight stays lightweight.

Architecture

How it's built

Vanilla JavaScript, organized as 40+ modular systems. Canvas2D for the threat network, world map, dual-curve containment chart, and temporal overlay. Web Audio API for a generative soundscape that shifts with the section you're in and sonifies the risk sliders — no audio files, all synthesis. IntersectionObserver drives reveal animations and lazy initialization for sections below the fold.

A service worker with network-first strategy caches every JS file, HTML, CSS, and dataset. The cache is versioned so deploys clean up old entries. All personal state — slider positions, saved scenarios, temporal history, research workspace, reading progress, tutorial progress — lives in localStorage on the device. Nothing is transmitted anywhere. The Live Signal feed aggregates ten RSS sources through a CORS proxy with a 30-minute cache TTL, so even the “live” news keeps working on an airplane.

Shareable assessment links encode slider state and active perspective into URL hash parameters. Send someone a link, and they open the tracker configured exactly as you set it. The report generator produces standalone HTML with inline SVG sparklines, ready to print to PDF for policy briefs or academic submissions. Mobile apps are in development.

Roadmap

Where this goes next

The web app is live in alpha. Mobile apps for iOS and Android are in development, built on the same offline-first data model so assessments sync with a shareable link rather than an account. A collaborative calibration backend is designed but not yet wired in — currently seeded with synthetic distributions so the histogram UI is testable against real interaction patterns before any real data is collected.

The dataset is the living part. Timeline events, threat scores, containment readiness, and literature annotations are versioned and editable. The goal is for researchers, journalists, and policy analysts to be able to fork the data, propose changes, and cite specific versions when they publish. A research tool is only as useful as its transparency about where the numbers come from.

The tracker isn't trying to predict the singularity. It's trying to make sure the conversation about it happens with the evidence on the table.