We believe that understanding our future with artificial intelligence requires neither breathless optimism nor paralyzing dread. It requires what science has always demanded: honest observation, humble hypotheses, and transparent math.
Our numbers are not prophetic decrees. They represent a calibrated probability distribution that reflects our current state of evidence across computing, labor economics, safety research, and governance.
The five levels form a mutually exhaustive spectrum of human outcomes. The sum of all probabilities always equals exactly 100%: Σ P(Li) = 100%. When evidence raises the likelihood of one future, another must yield.
We reject the illusion of absolute certainty. We enforce a strict floor constraint (P(Li) ≥ 2.0%) so no plausible future is prematurely discarded or declared inevitable.
Social media rumors and sensationalist press releases do not move our numbers. Only reproducible technical benchmarks, SEC capital filings, peer-reviewed labor studies, or enacted statutory laws trigger updates.
This observatory exists not as passive academic theater, but to provide everyday people with clear, practical guidance for their careers, families, and communities.
Our scoring engine distinguishes between Primary Empirical Signals (unambiguous technical benchmarks, SEC energy/compute filings, government labor statistics, and statutory laws) and our secondary Expert Layer (rigorous analysis from vetted technical researchers, empirical economists, and domain-limited sociotechnical specialists).
Why Damping Matters (0.35x): Even the most brilliant researchers can suffer from cognitive blind spots or groupthink. By damping expert commentary to a 0.35x multiplier, we ensure that subjective analysis can refine the probabilities, but can never overpower cold empirical reality (actual compute purchases, measurable labor displacements, and enacted legal statutes).
We screen expert sources strictly. We exclude social media pundits, hyper-partisan influencers, anonymous speculators, and documentary visibility without current substantive work. Sociotechnical specialists are admitted only within declared domains and horizons. Direct links and conflict disclosures support independent auditing.
Primary scope: Frontier neural network scaling, biological vs. digital computation, existential and catastrophic risk
Pioneer of deep learning whose warnings focus on digital intelligence surpassing biological limits and the difficulty of controlling goal-directed systems.
Primary scope: World models, Joint Embedding Predictive Architecture (JEPA), open-source foundational AI
Pioneer of deep learning advocating for world models and open-source decentralization, grounding debates by highlighting the architectural limits of autoregressive LLMs.
Primary scope: Frontier AGI architectures, AI-driven scientific discovery (AlphaFold), bio/materials breakthroughs
Direct builder of world-leading frontier systems focused on accelerating scientific discovery while establishing pre-deployment evaluations.
Primary scope: Frontier scaling laws, mechanistic interpretability, biological/cyber security risk mitigation
Frontier lab leader publishing detailed scaling essays and verifiable interpretability research, balancing massive capability scaling with catastrophic risk governance.
Primary scope: Superhuman alignment, core algorithmic breakthroughs, safety-focused frontier architectures
Core architect behind breakthrough scaling paradigms, now dedicated exclusively to technical alignment and provably safe superintelligence.
Primary scope: Frontier safety research, international governance treaties, non-profit AI architectures
Pioneer of deep learning focusing on non-profit, internationally inspected safety architectures and preventing dangerous concentration.
Primary scope: Practical frontier model architecture, autonomous agent workflows, AI-native education
Exceptional technical educator and practitioner offering clear, hype-free assessments of current agentic capabilities and practical engineering bottlenecks.
Primary scope: Agentic workflows, enterprise AI adoption, open-source democratization, practical utility
World-renowned AI educator and builder emphasizing iterative agentic patterns, domain-specific deployment, and practical economic augmentation.
Primary scope: Human-centered AI, spatial intelligence, public computing infrastructure, healthcare AI
Creator of ImageNet and champion of human-centered AI, advocating for democratic public compute and systems designed to augment human practitioners.
Primary scope: Open-source LLM architectures, fine-tuning efficiency, model compression, empirical benchmarking
Leading technical author and researcher providing transparent, code-backed breakdowns of model training, quantization, and architectural efficiency.
Primary scope: Empirical workplace productivity, cognitive labor augmentation, practical organizational adoption
Pioneering researcher running controlled workplace experiments on how humans and AI interact, providing hard data on labor productivity and transition friction.
Primary scope: Macroeconomic labor displacement, wage distribution, technological rent concentration
Leading institutional economists measuring whether automation acts as labor-augmenting or capital-concentrating.
Primary scope: Technomoral philosophy, human moral agency, practical virtue in automated societies
Leading philosopher examining human flourishing, practical wisdom, and retaining human agency in an automated world.
Primary scope: Empirical capability audits, debunking commercial hype, institutional adoption limits
Essential grounding voice analyzing where AI systems reliably work versus where claims outstrip technical reality.
Primary scope: Empirical safety benchmarks, catastrophic risk taxonomies, technical alignment
Author of widely adopted safety and reasoning benchmark datasets (MMLU, MATH, APPS) measuring complex failure modes.
Primary scope: Compute trends, hardware scaling, algorithmic efficiency metrics
Gold-standard empirical tracking of training FLOPs, hardware cost deflation, and model efficiency milestones.
Primary scope: Autonomous agent capability evals, cyber weaponization, red-team replication studies
Objective empirical testing of whether frontier models can autonomously replicate, deceive, or exploit networks.
Primary scope: Long-term capability forecasting, mechanistic alignment, societal coordination
Rigorous, detailed technical scenario models on the speed of cognitive automation and governance mechanisms.
Primary scope: AI deployment incentives, persuasive systems, psychological effects, human agency, and concentrated power
Sociotechnical analysts whose current AI Roadmap connects system design and deployment incentives with safety, rights, work, well-being, and distributed power. This entry is one correlated CHT voice, not two independent votes.
Primary scope: Algorithmic audits, AI governance, discriminatory impact, accountability, and high-stakes automated decisions
Applied auditor of real-world algorithms and AI systems in employment, healthcare, predictive decision-making, and generative AI. Her proper scope is documented deployment harm and governance across Levels 2, 3, and 4, not frontier capability forecasting.
Primary scope: Recommender systems, information integrity, algorithmic manipulation, transparency, and platform incentives
AI researcher and former platform engineer with direct technical experience in recommendation systems. His scope is the measured social and informational effects of ranking systems, mainly Levels 2 and 3, rather than general frontier AI forecasting.
These organizations provide research, audits, frameworks, and monitoring. They are tracked separately from the Expert Layer. An institutional profile never changes probabilities, and any future signal must be evaluated as an exact artifact.
Nonprofit policy and public-interest resource
Tracks how AI and persuasive technology shape human well-being, democratic agency, safety, rights, work, and concentrated power. Its AI Roadmap is a useful framework, not an empirical signal.
Algorithmic audit and governance practice
Provides applied audits and governance work for algorithms and AI used in high-stakes settings. Useful for observed deployment risks and accountability methods.
Algorithmic transparency and information-integrity resource
Investigates how recommendation algorithms shape public information and attention. Useful for platform incentives and measurable information-system effects.
Systems-thinking and public-sensemaking library
Offers long-form work on technological change, civilizational risk, public sensemaking, and social coordination. Best used as a curated Perspective source because publication is irregular and institutionally attributed.
Economic-democracy and shared-prosperity resource
Develops institutional alternatives for distributing power and prosperity as AI changes work and production. Useful for evaluating positive Level 4 and Level 5 pathways without treating advocacy as evidence.
Science flourishes in the open sunlight. All observatory data is publicly accessible in machine-readable JSON formats:
If you accept, Google Analytics uses cookies and receives data about the pages you visit and your browser to help us understand site usage. It does not load until you accept. Advertising features and Google Signals are disabled. We save your choice in this browser. You can change it here at any time; declining removes Analytics cookies accessible from this site and reloads the page. This does not delete data already sent to Google.
We could not save your choice. Analytics remains off.
