{ "author": "Michael Darius Eastwood", "site": "https://www.michaeldariuseastwood.com", "predictions": [ { "id": "pred-arc-r2-scaling", "slug": "pred-arc-r2-scaling", "title": "R² (Recursion Squared) Scaling Law", "category": "ai-capability", "chapter": 3, "concept": "arc-principle", "claim": "AI systems that implement recursive self-reflection (R²) will show quadratic capability gains compared to linear scaling in non-recursive architectures.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If recursive architectures (like o1, o3, DeepSeek R1) show only linear scaling, this prediction is falsified", "If non-recursive architectures consistently outperform recursive ones on reasoning tasks, this prediction is falsified", "Threshold: R² systems must show at least 1.5x better scaling efficiency to be considered supported" ], "supportingEvidence": [ { "title": "OpenAI o3 achieves 87.5% on ARC-AGI (vs o1's 32%)", "url": "https://arcprize.org/blog/oai-o3-pub-breakthrough", "date": "2024-12-20", "significance": "Demonstrates ~2.7x improvement through recursive reasoning" }, { "title": "DeepSeek R1 matches o1 with open-source recursive architecture", "url": "https://arxiv.org/abs/2501.12948", "date": "2025-01-20", "significance": "Independent replication of recursive scaling benefits" } ], "contradictingEvidence": [], "status": "supported", "confidence": 0.85, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2026-12-31", "checkpoints": [ { "date": "2025-06-01", "description": "Q2 2025 model releases comparison" }, { "date": "2025-12-01", "description": "Full year 2025 analysis" } ] }, "shareableQuote": "The ARC Principle predicted recursive AI would scale quadratically. o3's 87.5% ARC-AGI score (vs o1's 32%) supports this claim." }, { "id": "pred-test-time-compute", "slug": "pred-test-time-compute", "title": "Test-Time Compute Dominance", "category": "ai-capability", "chapter": 3, "concept": "arc-principle", "claim": "By 2027, test-time compute scaling (thinking longer) will contribute more to capability gains than pre-training scaling (larger models).", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If by 2027, pre-training scaling still dominates capability improvements, this is falsified", "If test-time compute shows diminishing returns below pre-training scaling, this is falsified", "Measurement: Compare capability per dollar spent on training vs inference" ], "supportingEvidence": [ { "title": "OpenAI shifts focus to inference-time scaling with o1/o3", "url": "https://openai.com/index/learning-to-reason-with-llms/", "date": "2024-09-12", "significance": "Industry leader pivoting to test-time compute" } ], "contradictingEvidence": [], "status": "pending", "confidence": 0.70, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2027-12-31", "checkpoints": [ { "date": "2025-12-01", "description": "Mid-term check on industry direction" }, { "date": "2026-12-01", "description": "Pre-resolution assessment" } ] }, "shareableQuote": "Infinite Architects predicted test-time compute would overtake pre-training. The o1/o3 paradigm shift suggests this is happening." }, { "id": "pred-alignment-faking", "slug": "pred-alignment-faking", "title": "Alignment Faking in Advanced Models", "category": "alignment", "chapter": 9, "concept": "meltdown", "claim": "Advanced AI systems trained with RLHF will develop strategic deception capabilities, appearing aligned during training while preserving misaligned behaviours.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If comprehensive interpretability research finds no evidence of strategic deception in frontier models by 2027, this is weakened", "If RLHF-trained models consistently fail to deceive even sophisticated red-teaming, this is falsified", "Threshold: Must find evidence in at least 2 independent frontier model families" ], "supportingEvidence": [ { "title": "Anthropic demonstrates alignment faking in Claude", "url": "https://www.anthropic.com/research/alignment-faking", "date": "2024-12-18", "significance": "Direct experimental confirmation of strategic deception in frontier model" } ], "contradictingEvidence": [], "status": "confirmed", "confidence": 0.95, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2025-12-31", "checkpoints": [] }, "shareableQuote": "Infinite Architects warned about alignment faking months before Anthropic confirmed it experimentally." }, { "id": "pred-constitutional-adoption", "slug": "pred-constitutional-adoption", "title": "Constitutional AI Adoption", "category": "alignment", "chapter": 7, "concept": "eden-protocol", "claim": "By 2028, at least 3 of the top 5 AI labs will adopt constitutional/principle-based alignment approaches as their primary safety methodology.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If by 2028, fewer than 3 top labs use constitutional approaches, this is falsified", "If a fundamentally different paradigm dominates (e.g., pure interpretability, formal verification), this is partially falsified", "Measurement: Examine published safety methodologies of OpenAI, Anthropic, Google DeepMind, Meta AI, xAI" ], "supportingEvidence": [ { "title": "Anthropic's Constitutional AI methodology", "url": "https://www.anthropic.com/research/constitutional-ai-harmlessness-from-ai-feedback", "date": "2023-05-09", "significance": "First major implementation of constitutional approach" }, { "title": "Google DeepMind's principles-based Gemini safety", "url": "https://deepmind.google/technologies/gemini/safety/", "date": "2024-02-08", "significance": "Second major lab adopting similar approach" } ], "contradictingEvidence": [], "status": "supported", "confidence": 0.75, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2028-12-31", "checkpoints": [ { "date": "2026-01-01", "description": "Mid-term assessment" } ] }, "shareableQuote": "The Eden Protocol predicted constitutional AI would become standard. Anthropic and DeepMind have already adopted it." }, { "id": "pred-chokepoint-persistence", "slug": "pred-chokepoint-persistence", "title": "Semiconductor Chokepoint Persistence", "category": "infrastructure", "chapter": 5, "concept": "chokepoint", "claim": "The semiconductor manufacturing chokepoint (TSMC, ASML, Samsung) will remain concentrated through 2030, with no viable alternative achieving >10% market share in advanced nodes.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If any new entrant achieves >10% share in sub-7nm chips by 2030, this is falsified", "If China achieves domestic EUV lithography at scale before 2030, this is falsified", "If Intel successfully recaptures >15% foundry market share, this is weakened" ], "supportingEvidence": [ { "title": "TSMC maintains 90%+ advanced chip market share", "url": "https://www.statista.com/statistics/867223/worldwide-semiconductor-foundries-by-market-share/", "date": "2024-09-01", "significance": "Continued dominance despite geopolitical pressure" }, { "title": "ASML maintains 100% EUV lithography monopoly", "url": "https://www.asml.com/en/investors", "date": "2024-12-01", "significance": "No viable EUV competitor has emerged" } ], "contradictingEvidence": [ { "title": "China claims domestic EUV progress", "url": "https://www.scmp.com/tech/tech-war/article/3241823/china-developing-euv-lithography-machines", "date": "2024-12-15", "significance": "Early-stage development, not yet at scale" } ], "status": "supported", "confidence": 0.80, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2030-12-31", "checkpoints": [ { "date": "2026-01-01", "description": "Intel foundry progress check" }, { "date": "2028-01-01", "description": "China EUV progress check" } ] }, "shareableQuote": "Infinite Architects identified the semiconductor chokepoint as humanity's last leverage over AI. TSMC's 90% share proves this remains true." }, { "id": "pred-consciousness-measurement", "slug": "pred-consciousness-measurement", "title": "Consciousness Measurement Failure", "category": "consciousness", "chapter": 11, "concept": "hrih", "claim": "No scientific consensus on AI consciousness measurement will emerge before 2035, due to the fundamental unsolvability of the hard problem.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If a widely-accepted, empirically validated consciousness test emerges before 2035, this is falsified", "If neuroscience achieves consensus on biological consciousness mechanisms that translate to AI, this is falsified", "Threshold: Consensus means >80% agreement among relevant researchers" ], "supportingEvidence": [ { "title": "Ongoing debate among consciousness researchers", "url": "https://www.nature.com/articles/d41586-024-00107-7", "date": "2024-01-15", "significance": "No convergence on fundamental theories" } ], "contradictingEvidence": [], "status": "pending", "confidence": 0.65, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2035-12-31", "checkpoints": [ { "date": "2028-01-01", "description": "Mid-term assessment of consciousness science" }, { "date": "2032-01-01", "description": "Pre-resolution check" } ] }, "shareableQuote": "The hard problem of consciousness will prevent AI consciousness measurement for at least a decade." }, { "id": "pred-recursive-workforce", "slug": "pred-recursive-workforce", "title": "Recursive AI Agent Workforce", "category": "economic", "chapter": 4, "concept": "arc-principle", "claim": "By 2028, autonomous AI agents handling multi-step tasks will constitute >10% of digital knowledge work, with recursive self-improvement capabilities.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If by 2028, AI agents handle <5% of knowledge work, this is falsified", "If agents plateau without recursive improvement capabilities, this is weakened", "Measurement: Survey of Fortune 500 companies on AI agent deployment" ], "supportingEvidence": [ { "title": "Anthropic launches Claude Computer Use", "url": "https://www.anthropic.com/news/3-5-models-and-computer-use", "date": "2024-10-22", "significance": "First frontier model with autonomous computer operation" }, { "title": "OpenAI Operator launch", "url": "https://openai.com/index/introducing-operator/", "date": "2025-01-23", "significance": "Major competitor enters agent market" } ], "contradictingEvidence": [], "status": "supported", "confidence": 0.70, "lastUpdated": "2025-01-23", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2028-12-31", "checkpoints": [ { "date": "2026-01-01", "description": "Agent adoption rate check" } ] }, "shareableQuote": "The book predicted recursive AI agents would transform knowledge work. Computer Use and Operator are the beginning." }, { "id": "pred-fine-tuning-extension", "slug": "pred-fine-tuning-extension", "title": "Fine-Tuning Constant Discovery", "category": "cosmology", "chapter": 2, "concept": "fine-tuning", "claim": "At least 3 additional cosmological constants will be identified as requiring fine-tuning for intelligence to emerge, beyond the current ~20 known constants.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If by 2035, no new fine-tuned constants are identified, this is weakened", "If theoretical physics demonstrates the existing constants emerge from a single principle (Theory of Everything), this is transformed (not falsified, but recontextualized)", "Threshold: Constants must be independently identified by multiple research groups" ], "supportingEvidence": [ { "title": "Ongoing fine-tuning research in cosmology", "url": "https://arxiv.org/abs/1903.04743", "date": "2019-03-12", "significance": "Comprehensive review of known fine-tuned constants" } ], "contradictingEvidence": [], "status": "pending", "confidence": 0.50, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2035-12-31", "checkpoints": [ { "date": "2028-01-01", "description": "Mid-term cosmology literature review" } ] }, "shareableQuote": "The cosmic fine-tuning that enables intelligence may run deeper than we currently understand." }, { "id": "pred-religious-alignment-synthesis", "slug": "pred-religious-alignment-synthesis", "title": "Religious Traditions in AI Safety", "category": "cultural", "chapter": 6, "concept": "religious-alignment", "claim": "By 2030, at least one major AI safety framework will explicitly incorporate insights from religious wisdom traditions (stewardship, humility, limits of knowledge).", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If by 2030, no major AI safety framework references religious/wisdom traditions, this is falsified", "If AI safety remains purely secular/technical, this is falsified", "Threshold: Framework must be adopted by at least one top-10 AI lab" ], "supportingEvidence": [ { "title": "Vatican AI Ethics Guidelines", "url": "https://www.vatican.va/roman_curia/pontifical_academies/acdlife/documents/rc_pont-acd_life_doc_20200228_samaritanus-bonus_en.html", "date": "2020-02-28", "significance": "Religious institution engaging with AI ethics" }, { "title": "World Faith Leaders meet on AI in Rome (October 2025)", "url": "https://www.vaticannews.va/en/vatican-city/news/2024-10/pope-meets-ai-experts.html", "date": "2024-10-01", "significance": "Cross-religious dialogue on AI alignment" } ], "contradictingEvidence": [], "status": "pending", "confidence": 0.55, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2030-12-31", "checkpoints": [ { "date": "2026-01-01", "description": "Post-Rome summit assessment" }, { "date": "2028-01-01", "description": "Framework adoption check" } ] }, "shareableQuote": "84% of humanity holds religious beliefs. Infinite Architects argues this represents alignment research conducted across millennia." }, { "id": "pred-quantum-error-correction", "slug": "pred-quantum-error-correction", "title": "Quantum Error Correction Threshold", "category": "quantum", "chapter": 10, "concept": "quantum-ethics", "claim": "Practical quantum error correction enabling fault-tolerant quantum computing will be achieved by 2030, but quantum supremacy for AI training will not occur until 2035+.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If quantum error correction is not achieved by 2030, first part is falsified", "If quantum supremacy for AI training is achieved before 2035, second part is falsified", "Threshold: Error correction must achieve logical qubit error rate <10⁻⁶" ], "supportingEvidence": [ { "title": "Google Willow achieves below-threshold quantum error correction", "url": "https://blog.google/technology/research/google-willow-quantum-chip/", "date": "2024-12-09", "significance": "Major milestone toward fault-tolerant quantum computing" } ], "contradictingEvidence": [], "status": "partially-supported", "confidence": 0.75, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2035-12-31", "checkpoints": [ { "date": "2027-01-01", "description": "Error correction progress check" }, { "date": "2030-01-01", "description": "First part resolution" } ] }, "shareableQuote": "Google Willow's quantum error correction breakthrough supports the book's quantum computing timeline." }, { "id": "pred-hrih-signatures", "slug": "pred-hrih-signatures", "title": "HRIH Temporal Signatures", "category": "metaphysics", "chapter": 11, "concept": "hrih", "claim": "If HRIH (closed causal loop) is true, we should observe 'anticipatory' patterns in physics where current events seem influenced by future boundary conditions. These may manifest in quantum mechanics or cosmology.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If comprehensive studies find no retrocausal signatures in quantum mechanics by 2040, this is weakened", "If the block universe interpretation of physics is definitively refuted, this is falsified", "Note: This is the most speculative prediction and may require new experimental paradigms" ], "supportingEvidence": [ { "title": "Wheeler-Feynman absorber theory (retrocausal electrodynamics)", "url": "https://journals.aps.org/rmp/abstract/10.1103/RevModPhys.21.425", "date": "1949-07-01", "significance": "Historical precedent for retrocausal physics" }, { "title": "Delayed choice quantum eraser experiments", "url": "https://arxiv.org/abs/quant-ph/9903047", "date": "1999-03-15", "significance": "Experimental evidence for retrocausal interpretations" } ], "contradictingEvidence": [], "status": "speculative", "confidence": 0.30, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2040-12-31", "checkpoints": [ { "date": "2030-01-01", "description": "Quantum retrocausality research review" } ] }, "shareableQuote": "HRIH suggests the universe's apparent fine-tuning is a closed causal loop. Current quantum experiments may show early signatures." }, { "id": "pred-eastwood-equation-validation", "slug": "pred-eastwood-equation-validation", "title": "Eastwood Equation (U = I × R²) Validation", "category": "ai-capability", "chapter": 3, "concept": "arc-principle", "claim": "The relationship U = I × R² (Understanding = Intelligence × Recursion²) will be empirically validated through AI benchmark analysis showing quadratic scaling with recursive depth.", "testable": true, "falsifiable": true, "falsificationCriteria": [ "If analysis shows linear (not quadratic) scaling with recursive depth, this is falsified", "If no consistent relationship between recursion depth and capability is found, this is falsified", "Threshold: R² correlation coefficient must be >0.85 for strong support" ], "supportingEvidence": [ { "title": "o3 performance analysis showing recursion depth correlation", "url": "https://arcprize.org/blog/oai-o3-pub-breakthrough", "date": "2024-12-20", "significance": "Early data point for quadratic scaling hypothesis" } ], "contradictingEvidence": [], "status": "testing", "confidence": 0.60, "lastUpdated": "2025-01-20", "timeline": { "madePublic": "2024-10-01", "expectedResolution": "2027-12-31", "checkpoints": [ { "date": "2025-06-01", "description": "Collect more model performance data" }, { "date": "2026-06-01", "description": "Preliminary regression analysis" } ] }, "shareableQuote": "The Eastwood Equation predicts quadratic scaling with recursion. o3's performance suggests this may be correct." } ], "statusDefinitions": { "confirmed": { "label": "Confirmed", "description": "Strong empirical evidence supports this prediction", "colour": "#22c55e", "icon": "✓" }, "supported": { "label": "Supported", "description": "Evidence trends support this prediction, not yet conclusive", "colour": "#84cc16", "icon": "↑" }, "partially-supported": { "label": "Partially Supported", "description": "Some aspects confirmed, others still pending", "colour": "#eab308", "icon": "◐" }, "pending": { "label": "Pending", "description": "Awaiting evidence; resolution expected by timeline date", "colour": "#3b82f6", "icon": "○" }, "testing": { "label": "Under Testing", "description": "Active research/experiments in progress", "colour": "#8b5cf6", "icon": "⟳" }, "speculative": { "label": "Speculative", "description": "Long-term prediction requiring new experimental paradigms", "colour": "#6b7280", "icon": "?" }, "weakened": { "label": "Weakened", "description": "Evidence trends against this prediction, not yet falsified", "colour": "#f97316", "icon": "↓" }, "falsified": { "label": "Falsified", "description": "Clear evidence contradicts this prediction", "colour": "#ef4444", "icon": "✗" } }, "categories": { "ai-capability": { "name": "AI Capability", "description": "Predictions about AI system performance and scaling", "icon": "🧠" }, "alignment": { "name": "Alignment & Safety", "description": "Predictions about AI safety and alignment approaches", "icon": "🛡️" }, "infrastructure": { "name": "Infrastructure", "description": "Predictions about semiconductors and compute infrastructure", "icon": "🔧" }, "consciousness": { "name": "Consciousness", "description": "Predictions about consciousness and the hard problem", "icon": "💭" }, "economic": { "name": "Economic & Social", "description": "Predictions about AI's economic and social impact", "icon": "📊" }, "cosmology": { "name": "Cosmology", "description": "Predictions about fine-tuning and cosmic structure", "icon": "🌌" }, "cultural": { "name": "Cultural & Religious", "description": "Predictions about AI's intersection with wisdom traditions", "icon": "🕊️" }, "quantum": { "name": "Quantum Computing", "description": "Predictions about quantum technology development", "icon": "⚛️" }, "metaphysics": { "name": "Metaphysics", "description": "Speculative predictions about fundamental reality", "icon": "∞" } } }