cs.AIMar 15, 2026

Punctuated Equilibria in Artificial Intelligence: The Institutional Scaling Law and the Speciation of Sovereign AI

Authors: Mark Baciak, Thomas A. Cellucci, Deanna M. Falkowski

Organizations: Ekta Inc. · Georgetown University

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

The dominant narrative of artificial intelligence development assumes that progress is continuous and that capability scales monotonically with model size. We challenge both assumptions. Drawing on punctuated equilibrium theory from evolutionary biology, we show that AI development proceeds not through smooth advancement but through extended periods of stasis interrupted by rapid phase transitions that reorganize the competitive landscape. We identify five such eras since 1943 and four epochs within the current Generative AI Era, each initiated by a discontinuous event -- from the transformer architecture to the DeepSeek Moment -- that rendered the prior paradigm subordinate. To formalize the selection pressures driving these transitions, we develop the Institutional Fitness Manifold, a mathematical framework that evaluates AI systems along four dimensions: capability, institutional trust, affordability, and sovereign compliance. The central result is the Institutional Scaling Law, which proves that institutional fitness is non-monotonic in model scale. Beyond an environment-specific optimum, scaling further degrades fitness as trust erosion and cost penalties outweigh marginal capability gains. This directly contradicts classical scaling laws and carries a strong implication: orchestrated systems of smaller, domain-adapted models can mathematically outperform frontier generalists in most institutional deployment environments. We derive formal conditions under which this inversion holds and present supporting empirical evidence spanning frontier laboratory dynamics, post-training alignment evolution, and the rise of sovereign AI as a geopolitical selection pressure.

Figures & tables

Figure 1: Evolutionary Timeline of Artificial Intelligence—From Abiogenesis to the Generative AI Era.
EraPeriodDefining CharacteristicsPhase Transition
1. Abiogenesis1943–1956McCulloch-Pitts, Turing Test, Shannon info theoryDartmouth Conference (1956)
2. Paleozoic (Symbolic)1956–1986Rule-based AI, expert systems, first AI winterExpert system collapse
3. Mesozoic (Statistical)1986–2012Backprop revival, SVMs, Bayesian methods, second AI winter, shallow MLAlexNet / ImageNet (2012)
4. Cenozoic2012–2017CNNs, RNNs/LSTMs, AlphaGo, GANs (2014), VAEsTransformer (2017)
5. Generative AI2017–presentTransformers, LLMs, diffusion, RLHF, agentic AI— (current era)
Table 1: Evolutionary Taxonomy of AI Development
Figure 2: The Institutional Scaling Law. Institutional fitness F(N,ε)\mathcal{F}(N,\varepsilon) is non-monotonic: each deployment environment has a distinct optimal scale N∗(ε)N^{*}(\varepsilon) . The dashed line shows the classical capability-only view, which increases monotonically. The shaded region marks the Capability-Trust Divergence Zone where scaling up reduces institutional fitness.
Figure 3: Symbiogenetic Scaling Inversion. In a trust-weighted EU deployment environment ( wT=0.40w_{\mathcal{T}}=0.40 ), an orchestrated system of three domain-specific models (7B + 3B + 2B, total 12B parameters) with high communication density ρ(G)\rho(G) achieves institutional fitness Fagent=0.82\mathcal{F}_{\text{agent}}=0.82 —exceeding both the environment’s individual-model optimum at N∗≈45N^{*}\approx 45 B and the frontier generalist at N=400N=400 B ( F=0.46\mathcal{F}=0.46 ). The star marker denotes the system’s aggregate scale; the orchestration multiplier (Equation 10 ) elevates it above the single-model curve.
Figure 4: Allometric Growth in Generative AI—Parameter Scaling and Power-Law Performance. (a) Exponential growth of model parameters over time. (b) Kaplan scaling law: cross-entropy loss as a power law of model size.
Figure 5: Taxonomic Diversification of Generative AI Models Across Modalities (2018–2025). The cladogram shows rapid branching from the common transformer ancestor into text, code, image, video, audio, science, multi-modal, and reasoning/agent lineages across a temporal axis.
Figure 6: AI Capability Acceleration—Epoch Capabilities Index (2023–2026). Each point represents a model’s composite capability score on the Epoch Capabilities Index (ECI), which unifies benchmarks into a single scale using Item Response Theory [ 46 ] . The red line traces the frontier (record-setting models). Dashed lines show piecewise linear trends: capability progress nearly doubled in rate at the April 2024 breakpoint ( ∼\sim 8 ECI/year →\to ∼\sim 16 ECI/year), several months before the release of reasoning-augmented models (o1, September 2024). Kwa et al. [ 49 ] independently corroborate the acceleration timing via METR task-horizon data. This empirical acceleration pattern provides quantitative evidence for the punctuated equilibrium dynamic described in this paper.
Figure 7: From Symbiogenesis to Niche Construction. Left: The Symbiogenesis pattern (2024–2025)—independent components (LLMs, code interpreters, search, memory, tool-use) fuse into composite agentic systems. Right: The Niche Construction pattern (early 2026)—composite systems actively modify their deployment environments (codebases, training infrastructure, physical world), creating feedback loops where the modified environment reshapes the selection pressures on subsequent system generations. Dashed red arrows denote the feedback channel. Three early 2026 case studies instantiate this dynamic.
Figure 8: Global Frontier AI Labs—Founding Timeline, Country of Origin, and Active Period. Labs are grouped by region and sorted by founding date, with consistent color coding by geography. Vertical dashed lines mark the two defining punctuation events of the Generative AI era. The clustering of new lab formations in 2023 reflects the adaptive radiation dynamics of Epoch II.
LaboratoryCountryEntryFirst ModelKey Contributions
OpenAIUSAJun 2018GPT-1 (117M)GPT series through GPT-5, o-series reasoning, DALL ⋅\cdot E, Codex, Stargate ($500B)
Google DeepMind ∗UK/USADec 2023Gemini 1.0Predecessor labs co-originated the Transformer (2017) and BERT (2018); PaLM, Gemini series, AlphaFold
AnthropicUSAMar 2023Claude 1.0Constitutional AI (RLAIF), Claude through Opus 4.6
Meta AI (FAIR)USAFeb 2023LLaMA (65B)Llama open-weight series, PyTorch, open-source leadership
Mistral AIFranceSep 2023Mistral 7BMixtral MoE, European sovereignty, Apache 2.0
DeepSeekChinaNov 2023DeepSeek CoderR1 reasoning (Jan 2025), V3 MoE, cost-efficiency breakthroughs
Table 2: Major Frontier AI Laboratories (as of March 2026)
Figure 9: Evolution of Post-Training Alignment Methods—From RLHF to GRPO and Beyond (2022–2026). Dominant paradigms (top) are superseded at accelerating rates ( ∼\sim 18 months →\to ∼\sim 12 months →\to weeks), while variant methods (bottom) feed innovations into the next dominant paradigm. The pipeline simplifies progressively from three models to one.
Figure 10: The Rise of Sovereign AI—National Programs and Relative Investment Scale (2023–2026). Major national sovereign AI programs grouped by region and ranked by investment scale, showing the global distribution of sovereign AI initiatives across six continents.
Figure 11: Noogenesis Transition Indicators—Status Assessment (March 2026). Five necessary conditions for the Symbiogenesis–Noogenesis epoch transition are substantially met (green). The sufficient conditions—closed-loop recursive self-improvement and autonomous objective setting—remain unmet (red), with AI-assisted training optimization representing a partially met intermediate state (orange). The critical gap between necessary and sufficient conditions defines the current distance to the epoch boundary.
More in this topic

AI Scaling Laws

AI scaling laws describe how model performance changes with training compute, data, model size, architecture, and inference-time resources.

See all 27 papers

Explore similar work

CardsList
Mar 14, 2026

The Institutional Scaling Law: Non-Monotonic Fitness, Capability-Trust Divergence, and Symbiogenetic Scaling in Generative AI

Classical scaling laws model AI performance as monotonically improving with model size. We challenge this assumption by deriving the Institutional Scaling Law, showing that institutional fitness -- jointly measuring capability, trust, affordability, and sovereignty -- is non-monotonic in model scale, with an environment-dependent optimum N*(epsilon). Our framework extends the Sustainability Index of Han et al. (2025) from hardware-level to ecosystem-level analysis, proving that capability and trust formally diverge beyond critical scale (Capability-Trust Divergence). We further derive a Symbiogenetic Scaling correction demonstrating that orchestrated systems of domain-specific models can outperform frontier generalists in their native deployment environments. These results are contextualized within a formal evolutionary taxonomy of generative AI spanning five eras (1943-present), with analysis of frontier lab dynamics, sovereign AI emergence, and post-training alignment evolution from RLHF through GRPO. The Institutional Scaling Law predicts that the next phase transition will be driven not by larger models but by better-orchestrated systems of domain-specific models adapted to specific institutional niches.

Mark Baciak, Thomas A. Cellucci

Ekta Inc.

Trustworthy AILanguage Model Scaling LawsTrust in AIResponsible AI
AskPDF
Mar 9, 2018

Anchoring Distributed Agency: Institutional Metaphors for Designing Large-Scale AI and AI Techniques for Running Institutions

Artificial Intelligence (AI) started out with the ambition to reproduce the human mind, but as soon as the sheer scale of that ambition became apparent, it retreated into either studying specialized intelligent behaviours or proposing overarching architectural concepts for interfacing specialized intelligent components---conceived of as agents acting through organizational structures. The agent-based modelling paradigm proved to have interesting applications beyond the original domain, supporting the understanding, simulation, and the assessment of behaviour of socio-economic and legal structures on an aggregate level. Looking at contemporary AI, whereas text completion provides a successful example of specialized behaviour, surprisingly little of the rich foundation of multi-agent systems (MAS) and normative systems research is currently reflected in Agentic AI systems. Addressing this gap, the chapter examines cross-cutting conceptualisations, modelling problems, and design challenges shared by large-scale distributed AI and institutional systems. Three main contributions can be highlighted. First, by tracing the historical trajectory of AI, the chapter identifies an ecological perspective on intelligence, examining both the value and limits of the compression-as-intelligence hypothesis. Second, it shows how moving between internal cognitive mechanisms and external manifestations of agency supports a functional separation of concerns for both individual and collective agency. Third, it argues that epistemological, normative, and value pluralisms are necessary conditions for sustaining the ecological character of intelligence. Furthermore, we indicate how AI techniques can support institutions by modelling social roles, simulating normative interactions, detecting and diagnosing failures, and supporting procedures of review and revision.

Alexander Boer, Giovanni Sileno

KPMG, Amsterdam, the Netherlands · Complex Cyber Infrastructure, Informatics Institute, University of Amsterdam, Amsterdam, the Netherlands

AI Agent ArchitectureMulti-Agent SystemsAgent-Based Modeling
AskPDF
May 13, 2026

Position: Agentic AI System Is a Foreseeable Pathway to AGI

Is monolithic scaling the only path to AGI? This paper challenges the dogma that purely scaling a single model is sufficient to achieve Artificial General Intelligence. Instead, we identify Agentic AI as a necessary paradigm for mastering the complex, heterogeneous distribution of real-world tasks. Through rigorous theoretical derivations, we contrast the optimization constraints of monolithic learners against the efficiency of Agentic systems, progressing from simple routing mechanisms to general Directed Acyclic Graph (DAG) topologies. We demonstrate that Agentic AI achieves exponentially superior generalization and sample efficiency. Finally, we discuss the connection to Mixture-of-Experts, reinterpret the instability of current multi-agent frameworks, and call for greater research focus on Agentic AI.

Junwei Liao, Shuai Li, Muning Wen +2

1Shanghai Jiao Tong University 2Shanghai Innovation Institute · 1Shanghai Jiao Tong University · University College London.

Multi-Agent SystemsAgentic AIArtificial General Intelligence
AskPDF