Convergence Encyclopedia: The Future Pursuit Map

PART 7: THE FUTURE PURSUIT MAP 15 priority-ranked research directions for advancing the convergence thesis.

F01: Citation Audit Completion Priority: 1 Time horizon: Immediate (0–6 months) What: Verify all ~90 citations across the 25 catalogue nodes. For each: confirm author, work, year, and result match the claim. Flag discrepancies. Produce corrected entries. Why: The entire encyclopedia rests on its sources. If the sources are wrong, the structure is decoration. This blocks all downstream work — no node can be load-bearing with an unverified citation. Who’s on it: This project. Systematic verification against Semantic Scholar, CrossRef, and primary sources. What you would add: A machine-readable citation schema: each source as a typed voxel with DOI, verification status, confidence level, and correction history. The audit itself becomes an OIP receipt — a verifiable record of what was checked and what was found. Maps to: C01–C25 (all nodes), A11 (Receipt axiom) Tier if successful: T0 (the audit is a mechanical fact, not an empirical claim)

F02: Independence Graph Construction Priority: 2 Time horizon: Immediate (0–6 months) What: Trace the full influence/citation graph for every convergence edge. Who read whom? Who cited whom? When did knowledge transfer between domains occur? Produce a directed graph of intellectual inheritance. Why: The independence check is what distinguishes genuine convergence from academic incest. Without it, the Macy Conference cluster (Shannon → Wiener → von Neumann) inflates apparent convergence by shared causation. Who’s on it: History of science (Menard, Kuhn), citation network analysis (Sinatra et al. 2016), computational historiography (Lamoreaux et al.). What you would add: Automated independence scoring using citation network topology. If two “independent” discoveries share intermediate nodes within k hops, independence decays exponentially. The OIP graph structure is the natural representation — independence becomes a graph-theoretic property. Maps to: All nodes with independence_check facet, N07 (Independence Problem) Tier if successful: T1 (the graph is empirical; the independence scores are estimates)

F03: Memory Node Deep-Dive Priority: 3 Time horizon: Immediate (0–6 months) What: Fully populate the rival_frame, falsifier, critics, and independence_check facets for Node C07 (Feedback/Cybernetics/Homeostasis) — the most underspecified load-bearing node. Then template this deep-dive for all remaining nodes. Why: C07 is the linchpin connecting physical, biological, and social systems. If its convergence claims don’t survive adversarial review, the entire cross-domain bridge collapses. The deep-dive template standardizes node evaluation. Who’s on it: Cybernetics historians (Pickering, Kline), systems biologists (Kitano), control theorists (Slotine), PCT researchers (Powers, Marken). What you would add: The template: for each node, populate (a) the strongest rival explanation in the rival’s own terms, (b) the exact observation that would falsify the convergence claim, (c) three named critics and their specific objections, (d) the citation chain between domains with independence score. This becomes the adversarial standard for all future node evaluation. Maps to: C07, C08, C12, C13 (cybernetics-adjacent nodes) Tier if successful: T1 (template is methodological; individual node evaluations are T1–T2)

F04: Constructal Law Empirical Test Priority: 4 Time horizon: Medium (6–18 months) What: Design an experiment to falsify Bejan’s Constructal Law claim that “for a finite-size flow system to persist in time, it must evolve to provide greater access to its currents.” Test whether engineered systems (circuit boards, traffic networks, river models) spontaneously evolve toward configurations predicted by constructal theory, or whether alternative configurations outperform. Why: Constructal Law is either a deep principle (T1) or a tautology (T5). The convergence thesis needs to know which. Bejan claims it unifies physics, biology, and engineering — if falsified, a major convergence branch is pruned. Who’s on it: Bejan (Duke, mechanical engineering), Lorente (INSA Toulouse), Reis (Cambridge). Critics: Ghodoossi (2004), Keszthelyi (2005) dispute universality. What you would add: A competitive falsification: evolve the same flow system under (a) constructal predictions, (b) gradient descent optimization, (c) simulated annealing, (d) biological evolution. If constructal predictions systematically underperform, the law is at best a heuristic. If they match or exceed, the law is a genuine convergent principle. OIP structure captures the experiment as a typed, receipt-backed protocol. Maps to: C01 (dissipation), C10 (scale invariance), C19 (thermoeconomics), C24 (fine-tuning) Tier if successful: T1 (empirical result) or T5 (if falsified, constructal demoted)

F05: Power-Law Reanalysis Priority: 5 Time horizon: Medium (6–18 months) What: Apply Clauset-Shalizi-Newman (2009) statistical methods to all claimed power-law distributions in the catalogue (C05: SOC, C10: scale invariance, C16: branching, C11: network degree distributions). CSN methods provide rigorous goodness-of-fit testing against alternative heavy-tailed distributions (log-normal, stretched exponential, power-law with cutoff). Why: Many claimed power laws are poorly validated. Stumpf & Porter (2012) showed that “criticality is everywhere” claims are often statistical artifacts. If the power-law evidence collapses, the SOC universality claim (C05) drops from T1 to T2, and scale-invariance claims require re-evaluation. Who’s on it: Clauset (CU Boulder), Shalizi (CMU), Newman (U Michigan) — the gold-standard methodology exists. Broido & Clauset (2019) applied it to network degree distributions and found far fewer true power laws than claimed. What you would add: Systematic application across all convergence domains: neural avalanches (C05), city size distributions (C10), citation networks (C11), species extinction events (C16). Each domain gets a CSN analysis with comparison to alternative distributions. Results update the catalogue node tiers automatically. The analysis itself is an OIP-typed voxel with full methodology and data receipts. Maps to: C05, C10, C11, C16, N05 (computational irreducibility — if true power laws are rare, prediction is harder) Tier if successful: T1 (statistical finding with established methodology)

F06: Assembly Theory Evaluation Priority: 6 Time horizon: Medium (6–18 months) What: Evaluate Cronin-Walker Assembly Theory as a genuine cross-domain principle or a domain-specific heuristic. Assembly index (AI) measures molecular complexity by minimum construction steps. Does it generalize beyond chemistry? Can it distinguish biological from abiotic systems? Is it computationally tractable for large systems? Why: Assembly Theory claims to unify chemistry, biology, and physics under a single complexity measure — a T1 convergence claim if true. If it fails to generalize, it demotes to T2 or T3. The convergence thesis must assess whether this is genuine pattern or chemistry-specific. Who’s on it: Cronin (Glasgow), Walker (ASU), Marshall (ASU). Critics: the theory is new (2021–2023) and independent replication is limited. Computational complexity of exact assembly index calculation is NP-hard. What you would add: Three-pronged evaluation: (1) computational — implement exact and approximate AI algorithms, measure tractability scaling; (2) empirical — test on known abiotic vs. biological samples; (3) theoretical — derive whether AI follows from established information theory (Kolmogorov complexity, logical depth) or is independent. The independence check is crucial: if AI reduces to Kolmogorov complexity, it’s not a new convergence but a reframing. Maps to: C06 (information), C09 (selection), C21 (emergence), C24 (fine-tuning) Tier if successful: T1 (if validated) or T2 (if partially validated) or T3 (if reframing of known theory)

F07: Edge-of-Chaos Replication Priority: 7 Time horizon: Medium (6–18 months) What: Replicate the Mitchell-Crutchfield-Hraber (1993–1994) experiments on computation at the edge of chaos using modern methods: larger cellular automata, longer evolution times, better statistical sampling, and validated complexity measures. Why: The original edge-of-chaos claims were influential but the specific results were criticized (Shalizi 2001). Modern computational resources allow far more thorough exploration. If confirmed, C05 (edge of chaos) strengthens. If not, the edge-of-chaos thesis needs significant revision. Who’s on it: Mitchell (Portland State), Crutchfield (UC Davis), Hraber (historical). Modern: Shalizi (CMU, criticism), Lizier (UTS, information dynamics), Prokopenko (Sydney, information-theoretic measures). What you would add: A registered replication with pre-specified analysis plan: evolve CA rule spaces at varying λ parameters, measure three distinct complexity metrics (statistical, computational, information-theoretic), and test the specific claim that “maximal computation occurs at intermediate λ.” Include statistical power analysis and null result interpretation. The replication protocol is an OIP receipt — pre-registered, immutable, auditable. Maps to: C05 (criticality), C20 (universal computation), N05 (computational irreducibility) Tier if successful: T1 (if replicated) or T2 (if ambiguous) — either way, the uncertainty reduces

F08: OIP Convergence Scoring Priority: 8 Time horizon: Immediate (0–6 months) What: Implement the convergence strength formula G(t) as a running metric on the OIP ledger. Each encyclopedia node is a typed voxel; each citation verification updates the node’s score; each independence check updates the edge weights. Produce a live-updating convergence dashboard. Why: The catalogue’s scoring formula (Convergence_Strength = Σ tier_weight × independence × citation_depth) is currently calculated manually. Automating it makes convergence assessment continuous and transparent. The ledger structure is ideal: every verification is a receipt, every score update is an amendment, every change is auditable. Who’s on it: This project. The OIP protocol at miscsubjects.com/a/oip provides the infrastructure. What you would add: A live convergence score per node, per edge, and per pattern. The score updates when: (1) a citation is verified (+citation_depth), (2) an independence check completes (+/- independence weight), (3) a rival frame is defeated (+tier stability), (4) a no-go theorem is shown to apply (-convergence). The dashboard shows not just scores but sensitivity: which nodes are most sensitive to which types of evidence. Maps to: All 25 nodes, OIP protocol, A11 (Receipt), A12 (Recursion) Tier if successful: T1 (tool is methodological; individual scores are empirical estimates)

F09: Cross-Modal Convergence Detection Priority: 9 Time horizon: Medium (6–18 months) What: Use LLM embeddings to find convergence points (Venn points) across disciplinary boundaries. Embed ~10,000 papers from 10 disciplines, cluster by semantic similarity, and identify clusters where semantically identical patterns are described with different vocabulary. Why: Human readers miss cross-domain convergence because of vocabulary barriers. The same mathematical structure — e.g., a power-law relaxation — is described as “critical slowing down” in physics, “long-tailed distributions” in ecology, and “Zipf’s law” in linguistics. LLM embeddings can pierce the vocabulary barrier and identify semantic identity across lexical difference. Who’s on it: Bibliometrics (Sinatra et al. 2016), LLM-based science of science (Hope et al. 2023), cross-disciplinary mapping (Börner et al.). What you would add: A convergence detection pipeline: (1) embed papers from arXiv categories (physics, biology, cs, econ, math); (2) cluster in embedding space; (3) flag cross-disciplinary clusters with high semantic similarity but low citation overlap; (4) human-review flagged pairs for genuine convergence vs. false positive; (5) incorporate confirmed convergences into the catalogue. The pipeline produces OIP-typed voxels for each detected convergence. Maps to: All 25 nodes, N07 (Independence Problem — detecting hidden common causes) Tier if successful: T2 (detection is heuristic; human review required)

F10: No-Go Integration Priority: 10 Time horizon: Medium (6–18 months) What: Formalize exactly where each of the 7 no-go theorems limits which convergence claims. Produce a constraint matrix: rows = no-gos, columns = patterns, cells = the specific limitation. Why: The no-gos are not merely negative — they are boundary conditions on the convergence thesis. Knowing precisely which patterns are limited by which no-gos prevents overclaiming and identifies where the thesis must remain silent. Who’s on it: Computational complexity theory (Wolpert, Aaronson), social choice theory (Arrow, Sen), philosophy of mathematics (Penrose, Lucas). The theorems are established; their application to convergence is new. What you would add: The constraint matrix as a machine-readable artifact. For each (no-go, pattern) pair: (a) does the no-go directly limit the pattern? (b) does it limit only certain instantiations? (c) does it create a tradeoff (e.g., more generality ↔ less decidability)? (d) is the pattern entirely outside the no-go’s scope? This becomes a formal part of the encyclopedia, not an afterthought. Maps to: N01–N07 (all no-gos), all 25 patterns Tier if successful: T1 (the matrix entries are derivations from established theorems)

F11: FEP Experimental Test Priority: 11 Time horizon: Medium (6–18 months) What: Design a falsifiable prediction from the Free Energy Principle that can be tested in a real active inference system. The prediction must be: (a) derivable from the FEP formalism, (b) testable with current methods, (c) risky — the FEP would be weakened if the prediction fails. Why: The FEP is either a unifying principle (T1) or an unfalsifiable framework (T3). The critics (Biehl et al. 2021, Ramstead et al. 2023 responses) argue it explains everything and therefore nothing. A successful risky prediction would resolve the status question. Who’s on it: Friston (UCL), Da Costa (UCL), Parr (UCL), Isomura (RIKEN). Critics: Bruineberg et al. (2021) — “free energy principle does not make falsifiable predictions.” What you would add: A specific prediction: in an active inference agent with fixed generative model, the precision-weighting of sensory prediction errors will adjust to match environmental volatility following the exact form of a Kalman gain update — and this adjustment will be suboptimal compared to the Bayes-optimal update by a quantifiable margin that scales with model complexity. Test this in a standardized active inference task (e.g., the “urn task” or a reaching task). If the prediction holds within tolerance, FEP gains empirical support; if it fails, the precision-engineering claim is weakened. Maps to: C13 (active inference), C07 (feedback), C02 (least action), N03 (Gödel — if FEP claims universality) Tier if successful: T1 (if prediction confirmed) or T2 (if ambiguous)

F12: Thermoeconomic Measurement Priority: 12 Time horizon: Long (18+ months) What: Measure exergy flow in real economic systems: how much useful work potential is destroyed at each step of production, trade, and consumption? Compare measured exergy destruction with economic measures of inefficiency (deadweight loss, TFP gaps). Why: If thermoeconomics (C19) is a genuine convergence, then exergy destruction should correlate with economic inefficiency across scales and systems. If not, thermoeconomics is at best a physics metaphor for economics, not a unified framework. Who’s on it: Ayres (INSEAD), Warr (Uppsala), Hammond (Oxford), Serrenho (Cambridge). The field exists but systematic measurement at economy-wide scale is sparse. What you would add: A standardized exergy accounting framework applied to a national economy (input-output table + energy flows). Test the specific convergence claim: sectors with highest economic inefficiency (lowest TFP) should also show highest exergy destruction per unit output. The correlation, if present, supports C19; if absent, weakens it. OIP structure captures the accounting as typed, auditable voxels. Maps to: C01 (dissipation), C19 (thermoeconomics), N02 (Arrow — if exergy-optimal allocation conflicts with social choice) Tier if successful: T2 (correlational, not causal) or T1 (if causal mechanism established)

F13: Criticality in AI Priority: 13 Time horizon: Medium (6–18 months) What: Determine whether neural networks self-organize to criticality during training. Measure: (1) avalanche statistics in activation patterns, (2) power-law distributions in gradient magnitudes, (3) correlation length scaling with system size, (4) susceptibility peaks at putative critical point. Why: If neural networks self-organize to criticality, C05 applies directly to AI systems — a major convergence point. If not, the SOC-AI connection is metaphorical, not mechanistic. The question is empirically decidable with current methods. Who’s on it: Beggs (Indiana, neuronal avalanches), Munoz (IFISC, SOC in biological networks), Bialek (Princeton, statistical mechanics of learning), Papyan (NYU, spectrum of NTK). What you would add: A standardized test suite: train ResNets and Transformers on standard benchmarks while measuring (a) avalanche-size distributions in activations, (b) gradient correlation lengths, (c) spectral properties of the Fisher information matrix near convergence. Compare with known critical and non-critical systems. If criticality signatures are absent, the SOC-AI connection is metaphor; if present, quantify the critical exponents and test universality across architectures. Maps to: C05 (criticality), C01 (dissipation — training as entropy export), C20 (computation — criticality as compute-optimal) Tier if successful: T1 (empirical measurement)

F14: Compressibility Measurement Priority: 14 Time horizon: Long (18+ months) What: Quantify the Kolmogorov complexity of physical laws: how compressible are the equations that describe the universe? Compare with algorithmic complexity of alternative (unphysical) equation sets. Why: Wigner’s “unreasonable effectiveness of mathematics” (C24) is the deepest convergence claim: why does so little math describe so much world? A partial answer: physical laws are highly compressible (low Kolmogorov complexity), making them discoverable. Measuring this compressibility would quantify the “unreasonableness.” Who’s on it: Zenil (Karolinska, algorithmic information), Hutter (ANU, universal intelligence), Schmidhuber (IDSIA, low-complexity art). Direct measurement is impossible (Kolmogorov complexity is uncomputable), but approximations exist (compression-based methods, coding theorem methods). What you would add: Apply approximation methods (compression length as upper bound, coding theorem method as lower bound) to the Standard Model Lagrangian, general relativity, and thermodynamic laws. Compare with random equation sets and alternative physical theories. If physical laws cluster at low complexity, this supports the convergence thesis: the universe is not just mathematical, it is simply mathematical. The measurement itself becomes a catalogue entry for C24. Maps to: C06 (information), C20 (computation), C24 (fine-tuning), N03 (Gödel — uncomputability limits) Tier if successful: T2 (approximation methods have known limitations) or T1 (if rigorous bounds established)

F15: The Wigner Residue Defense Priority: 15 Time horizon: Long (18+ months) What: Produce the strongest possible defense of the claim that mathematics describes the universe, and assess whether any convergence pattern explains or predicts this fact. If no pattern explains it, it is the irreducible residue — the strangest fact. Why: This is the meta-claim of the entire encyclopedia. All 25 patterns assume that mathematical structures recur across domains. But why should this be so? If the encyclopedia cannot address this question, it has a founding gap. If it can, the convergence thesis achieves a deeper level of unification. Who’s on it: Wigner (1960, original essay), Hamming (1980, follow-up), Tegmark (Mathematical Universe Hypothesis), Vilenkin (probabilistic approach), Carroll (poetic naturalism), Wheeler (“it from bit”). What you would add: Not a defense of any specific answer, but a typed classification of all proposed answers: (1) Mathematical Universe (Tegmark) — T3, unfalsifiable; (2) Anthropic selection — N06 applies, deflation; (3) Cognitive convergence — we evolved to see math where it helps survival; (4) Structural necessity — low-complexity laws are the only ones that can self-consistently exist (attempt to derive from C06 + C20); (5) The residue option — no current pattern explains it; it is the founding mystery. The encyclopedia takes no position but documents all, with their no-go constraints. The Wigner Residue becomes an explicit encyclopedia entry: what survives every deflation. Maps to: All 25 nodes, all 7 no-gos, A7 (Signatures) Tier if successful: T3 (the residue is philosophical) or T2 (if a partial explanation succeeds)

Part 7 Summary: 15 directions, priority-ranked by what blocks what. Immediate (F01–F03, F08): citation audit, independence graph, memory deep-dive, OIP scoring — these unblock everything else. Medium (F04–F07, F09–F11, F13): empirical tests and computational methods — these validate or invalidate specific convergence claims. Long (F12, F14–F15): economy-scale measurement and foundational questions — these address the deepest claims. Success on F01–F03 and F08 produces a machine-readable, auditable convergence database. Success on F04–F07 and F09–F11 produces validated or corrected claims. Success on F15 produces the typed classification of the founding question.

---

Corpus map

• Previous: Convergence Encyclopedia: The AI Pattern Map
• Next: Convergence Encyclopedia: The OIP Mapping
• Encyclopedia start: The Schema
• Kin corpora: Total Structure · Signature of the Grain