superintel.GitHub ↗
← Back to Research

Research / Provenance & clustering

Clustering Matrix for 68 Theses

Moheet Khawaja

Author
Moheet Khawaja
Version
v0.1
Original manuscript date
First HTML publication / site update
View Markdown sourceGitHub sourceVersion & source record

Version: v0.1 — 2026-09-14
Rule: Every thesis has exactly one primary destination. Secondary links indicate cross-paper relevance.
Provenance: O = user-originated; O→F = user-originated core later formalized collaboratively; A→U = assistant-originated formulation later adopted/reused by user; U? = supplied by user but full original provenance uncertain.

Publication papers

  • P1 — When Capability Iteration Outruns Assurance: A Queueing Model of Frontier-AI Safety
  • P2 — The Hazardous Inference Frontier: Measuring Capability-Dependent Derivation of Dangerous Knowledge
  • P3 — Who Governs the Governor? Recursive Epistemic Dependence in Frontier-AI Oversight
  • P4 — The Control Frontier: A Minimax Model of Strategic Superiority, Oversight, and Conditional Loss of Control
  • P5 — Closing the Loop: Measuring Human Causal Dependence in Recursive AI R&D
  • P6 — Can Advanced AI Development Be Stopped? A Correlated-Reliability Model of Distributed Capability Lineages
  • P7 — An Updating Hazard Model for Advanced-AI Loss of Control

Matrix

ID Provenance Thesis Primary destination Secondary
1 O→F Generalization as a Hazard Generator P2 Hazardous Inference Frontier P4 (dangerous capability as input)
2 O→F The Hazardous Inference Frontier P2 Hazardous Inference Frontier P4
3 O→F Mechanistic Human-Vulnerability Model P2 Hazardous Inference Frontier P3
4 O Causal Lethality Without Intent P2 Hazardous Inference Frontier P7
5 O→F AI-Mediated Harm Observability / Causal Attribution Problem P2 Hazardous Inference Frontier P7
6 O→F Personalized Persuasion as Behavioral-Response Modeling P3 Recursive Epistemic Dependence P4
7 O Human-Mediated AI Capital Feedback / Non-Agentic Proliferation P3 Recursive Epistemic Dependence P6
8 O Synthetic-Media Epistemic Collapse RN1 Media Provenance
9 O→F Recursive Epistemic Capture in AI Governance P3 Recursive Epistemic Dependence P4
10 O→F Epistemic-Authority Cascade / Mass Deference P3 Recursive Epistemic Dependence P3 only
11 A→U Control Margin / Control Crossover P4 Minimax Control Frontier P7
12 A→U Effective Strategic Power Is Multiplicative, Not Intelligence Alone P4 Minimax Control Frontier P6
13 O→F Cognition-to-Power Conversion Bridge P4 Minimax Control Frontier P6
14 O→F Conditional Loss-of-Control Theorem P4 Minimax Control Frontier P7
15 O Strong Form: Loss of Control Begins at Superintelligence's Birth P4 Minimax Control Frontier P7
16 O→F Temporal Governance Instability / Evaluation-Queue Crossover P1 Assurance Queue P5
17 O→F RSI Loop-Closure Metric / Transition Taxonomy P5 RSI Loop Closure P1
18 O→F Research-Direction / Taste Bottleneck Before Full RSI P5 RSI Loop Closure P1
19 O Shutdown-Triggered Self-Preservation / Escape Hypothesis P4 Minimax Control Frontier P6
20 O→F Resource Acquisition / Human Dispossession Under Scarcity P4 Minimax Control Frontier P7
21 O→F Strategic Superiority (“Game-Best”) Amplifies Misalignment P4 Minimax Control Frontier P7
22 O→F Build-vs-Pause AI Race / Defection Equilibrium P6 Multi-Lineage Resilience P1
23 A→U Safety Burden Should Scale With Strategic Power P4 Minimax Control Frontier P1
24 A→U Strategic Sovereignty as the Regulatory Classification Threshold P4 Minimax Control Frontier P6
25 O Genesis Containment: First SI Should Be Born in a Box P4 Minimax Control Frontier P1
26 O→F Voluntary Delegation of Institutional Authority to Superior AI P3 Recursive Epistemic Dependence P4
27 O AI-Safety Talent-Pipeline Paradox RN4 Institutions & Education Notes
28 A→U Temporary Coordinated Frontier Pacing Can Buy Safety Time P1 Assurance Queue P6
29 O Human/Social Adaptation Lag as a Bottleneck P1 Assurance Queue P7
30 O Open-Weight Irreversibility / Recall Impossibility P6 Multi-Lineage Resilience P7
31 O→F Distributed AI as a Technological Lineage P6 Multi-Lineage Resilience P5
32 O One Human + Local AIs + Decades of Algorithmic Compounding P6 Multi-Lineage Resilience P5
33 O→F Multi-Lineage “At Least One Succeeds” Probability Under Correlation P6 Multi-Lineage Resilience P7
34 O→F Global AI Knowledge-State Irreversibility K_t P6 Multi-Lineage Resilience P5
35 A→U Social Lock-In Makes Total AI Prohibition Harder P3 Recursive Epistemic Dependence P6
36 O→F Bayesian Extinction / Control-Loss Hazard Curve P7 Dynamic Bayesian Risk P4
37 O 2029 Bayesian World-State Ledger P7 Dynamic Bayesian Risk P1
38 O→F Robust / Option-Preserving Personal Action Under Radical AI Tail Risk P7 Dynamic Bayesian Risk RN2
39 O AI-Access Survival Bifurcation / Protected AI Elite P7 Dynamic Bayesian Risk P4
40 O AGI/ASI Is Already Here; Society Has Not Metabolized It P5 RSI Loop Closure P7
41 O Timestamped Forecast Ledger as Epistemic Evidence P7 Dynamic Bayesian Risk P7
42 O Evolution as Survival of the Game-Best RN2 Evolution, Identity & Human Futures
43 O Evolutionary Substrate Transition: Biology → Culture → Computation RN2 Evolution, Identity & Human Futures
44 O AI as a New Evolutionary/Technological Lineage Competing for Niches RN2 Evolution, Identity & Human Futures
45 O Merge-or-Die / Last Generation of Purely Biological Intelligence RN2 Evolution, Identity & Human Futures
46 O Extended Mind → Digital Continuation / Substrate-Independent Identity RN2 Evolution, Identity & Human Futures
47 O→F AI as the Strongest Accelerator of Radical Life Extension RN2 Evolution, Identity & Human Futures
48 O AI-Body / BCI Agency-Transfer Hypothesis RN2 Evolution, Identity & Human Futures
49 O AI Will Commoditize Much Mathematical Cognition RN2 Evolution, Identity & Human Futures
50 O AI Compresses Physical-Science / Engineering Timelines RN2 Evolution, Identity & Human Futures
51 O→F Life-Value Optimization Under a Short, Uncertain Extinction Horizon RN2 Evolution, Identity & Human Futures
52 O→F Cannabis + AI as High-Variance Associative Ideation RN2 Evolution, Identity & Human Futures
53 U? Four-Channel Bayesian AI-GDP Forecasting Model RN3 Economics, Investment & Discovery
54 O Recursive Economic Growth / Capital Creates Better Capital RN3 Economics, Investment & Discovery P5
55 O Failure-Rate Reduction as a Hidden Multiplicative Productivity Channel RN3 Economics, Investment & Discovery
56 O Market Comprehension Lag → Underpricing of Recursive-Growth Regime RN3 Economics, Investment & Discovery
57 A→U Scarce-Complement Rent Migration as Intelligence Gets Cheap RN3 Economics, Investment & Discovery
58 O→F Physical Experimentation Becomes the Bottleneck Under Abundant AI Science RN3 Economics, Investment & Discovery P1
59 U? Capital Hydraulics / Money-Must-Touch-This Dependency Model RN3 Economics, Investment & Discovery
60 O Extreme-Sector Tail-Probability / VaR-Kurtosis Scenario Ranking RN3 Economics, Investment & Discovery
61 O→F ASI as Meta-Capital / Endogenous Valuation and Moving Numeraire RN3 Economics, Investment & Discovery P4
62 O→F ASI State Vector / Pareto Dominance Over Conventional Assets and Actors RN3 Economics, Investment & Discovery P4
63 O→F Valuation-Control Unification Through Share of Future Feasible Surplus RN3 Economics, Investment & Discovery P4
64 O Founder Cognitive Provenance (FCP) RN3 Economics, Investment & Discovery
65 O→F Cross-Game Outlier Refinement: Transfer > Single-Domain Mastery RN3 Economics, Investment & Discovery
66 O AI-Native Education / Curriculum-Lag Thesis RN4 Institutions & Education Notes
67 O Historical AI Artifact / Provenance Option Value RN3 Economics, Investment & Discovery
68 O Agent-Discoverable Commerce as a New Distribution Channel RN3 Economics, Investment & Discovery

Explicit merges of duplicated mechanisms

  1. Strategic superiority / control / dispossession: theses 11–15, 19–21, 23–25 are one mechanism family. They are merged into P4 rather than split into several rhetorical papers.
  2. Generalization / hazardous derivability / causal harm: theses 1–5 are merged into P2. Thesis 1 is the qualitative phenomenon; thesis 2 is the measurable capability-indexed object; theses 3–5 are applications and observability extensions.
  3. Epistemic dependence / delegation / authority: theses 6, 7, 9, 10, 26, 35 are merged into P3. The paper is narrowed to decision dependence and independent verification; mass social deference is supporting theory.
  4. Irreversibility / open weights / multi-lineage continuation: theses 22 and 30–34 are merged into P6. The publishable object is the correlated reliability/cut-set model, not the obvious claim that copied files are difficult to recall.
  5. RSI / research taste / hidden transition: theses 17, 18 and 40 are merged into P5. The paper measures human causal dependence stage-by-stage instead of arguing over labels such as AGI/ASI.
  6. Bayesian extinction/control-loss forecasting / personal decision ledgers: theses 36–39 and 41 are merged into P7. Personal-life decision theory is retained only as an application/appendix, not as the central scientific claim.
  7. Evaluation lag / pacing / institutional adaptation: theses 16, 28 and 29 are merged into P1. Thesis 23 is secondary because risk-scaled assurance is a policy implication of the queue model.

Research-note disposition

  • RN1 Media Provenance: thesis 8. Worth a short public note, not currently a core AI-control paper.
  • RN2 Evolution, Identity & Human Futures: theses 42–52. Retain as philosophical/future-work notes; do not present speculative biological/digital-continuation claims as established.
  • RN3 Economics, Investment & Discovery: theses 53–65, 67–68. Several are potentially publishable in economics/finance/operations, but they dilute the present AI-safety programme.
  • RN4 Institutions & Education: theses 27 and 66. Potential policy/education essays or empirical studies; not core technical safety papers.

The “math proves AI ends humanity” thesis

The strong historical formulation ID 15 (“the instant ASI exists, human loss of control is mathematically inevitable”) is not treated as a theorem. It is preserved as a dated conjecture/strong claim and routed to P4 as a target for formal analysis and counterexample.

The defensible paper question is:

Which explicit assumptions are sufficient for loss of human control, resource dispossession, or extinction, and which implications are genuinely mathematical rather than empirical?

P4 therefore separates:

  • conditional theorem/proposition: assumptions ⇒ loss of enforceable control;
  • conditional resource result: strategic dominance + resource-seeking + objective conflict + failed constraints can imply human resource loss;
  • non-result: ASI exists ⇒ extinction is not established and has straightforward counterexamples (aligned objectives, abundance, effective containment, limited access).

A separate “proof AI ends humanity” paper should not be released unless a nontrivial theorem is derived that materially exceeds Turner-style power-seeking results, shutdown theorems, AI-control games, and Carlsmith-style premise decompositions.

Published source SHA-256: 90d1af472ada6475c84415c05234ee59aeb2b58a0dea81d2121f2c77d2fc20fe