Radical Emergence in Biological and Artificial Systems: From the Biosphere to Large Language Models

An Interdisciplinary Academic Review of Stuart Kauffman’s Radical Emergence Hypothesis, Biological Self-Organization, and the Unexpected Appearance of Intelligence

Abstract

The concept of emergence — the appearance of novel properties in complex systems that cannot be reduced to their individual components — has become a central theme in contemporary science, philosophy of biology, and artificial intelligence research. This document presents an extensive interdisciplinary review centered on Stuart Kauffman’s concept of radical emergence, which posits that the biosphere creates its own future possibilities of becoming without any entailing law or selection pressure directing these possibilities into existence. Drawing on Kauffman’s published work from 1996 to 2024, the broader literature on complex systems, ant colony self-organization, origin of life research, and the emerging field of emergent abilities in large language models (LLMs), this review examines the philosophical and scientific foundations of radical emergence, its validation and critique in the academic literature, and its striking parallels with the unexpected appearance of capabilities in AI systems. The analysis concludes that while radical emergence remains philosophically controversial, it offers a powerful framework for understanding how genuinely novel forms of organization — in biology, economics, and artificial intelligence — arise unpredictably from pre-existing conditions, fundamentally challenging the Enlightenment assumption that reason alone can guide understanding of complex phenomena.

Keywords: radical emergence, ontological emergence, epistemological emergence, Stuart Kauffman, adjacent possible, ant colony self-organization, origin of life, large language models, AGI, complex systems, biological novelty.

1. Introduction: Beyond «Emergence» as a Word

The term «emergence» has become ubiquitous in contemporary scientific discourse. Researchers in physics, biology, neuroscience, economics, and artificial intelligence invoke it to explain how complex, higher-level phenomena arise from simpler lower-level components. Yet, as Stuart Kauffman argues in a recent interview on the nature of reality, when one presses scholars to explain how emergence truly works, the answer often reduces to the word itself — a placeholder for genuine understanding rather than an explanation of it (Kauffman, 2024).

This review documents an ongoing research program — led primarily by Kauffman over three decades — that seeks to distinguish between what he terms epistemological emergence (which, while complex, is in principle deducible from first principles) and ontological emergence or radical emergence (which cannot be predicted or deduced in principle because it creates genuinely novel possibilities that did not exist prior to its occurrence). The distinction is not merely semantic; it carries profound implications for how we understand the nature of science, the limits of prediction, and the origins of life and intelligence on Earth.

Kauffman’s thesis, developed across multiple publications — from At Home in the Universe (1996) to Is the Emergence of Life an Expected Phase Transition in the Evolving Universe? (2024, co-authored with Andrea Roli) — can be stated succinctly: the biosphere is continuously creating its own future possibilities of becoming without any selection pressure or natural law directing these possibilities into existence. This claim represents a radical departure from the Darwinian framework, which accounts for adaptation but not for the creation of novelty itself.

The present review traces the intellectual trajectory of this idea, situates it within the broader literature on complex systems and self-organization, examines its application to ant colonies and collective biological intelligence, connects it to recent debates about emergent abilities in large language models, and evaluates the extent to which the scientific community has validated or challenged Kauffman’s central claims.

2. Theoretical Foundations: Kauffman’s Radical Emergence

2.1 From Random Boolean Networks to the Biosphere

Kauffman’s foundational contribution to complexity theory came through his work on Random Boolean Networks (RBNs), which he introduced in the 1960s as simplified models of genetic regulatory systems. An RBN consists of a set of nodes (representing genes or proteins), each receiving inputs from a fixed number of other nodes, with each node’s state (on or off) determined by a Boolean function of its inputs. Kauffman showed that these networks exhibit three regimes of behavior: ordered, critical, and chaotic, and that living systems appear to operate near the critical boundary between order and chaos — a regime that maximizes computational capacity and adaptability (Kauffman, 1993).

However, Kauffman is careful to distinguish this type of emergence from what he calls radical emergence. As he stated in the 2024 interview:

«These can behave in three ways: ordered, critical or chaotic. And it turns out that there’s evidence that cells are critical, and it’s neat and it’s lovely, and people would like to call that emergent — but it’s not. It’s hard to calculate, but we know ahead of time the state space. It’s just my Boolean Nets. We can certainly simulate it, and people have even proved theorems about it. This phase transition from order to chaos — I don’t want to call that emergence.»

The critical point here is epistemological: even though RBNs are complex, their state space is prestatable. One could, in principle, enumerate all possible configurations of the network. This is fundamentally different from what Kauffman calls radical emergence.

2.2 The Theory of the Adjacent Possible

Central to Kauffman’s framework is the Theory of the Adjacent Possible (TAP), which builds on a concept originally introduced by Stuart Kauffman in his 1996 work and later formalized in collaboration with colleagues. The adjacent possible refers to the set of structures, functions, and ecological niches that can potentially come into existence given the current state of the biosphere — but which did not exist before and could not have been predicted prior to the appearance of the enabling conditions.

In the 2022 paper Biocosmology: Towards the birth of a new science, co-authored with Marina Cortês, Andrew Liddle, and Lee Smolin (arXiv:2204.09378), Kauffman extends TAP to cosmological timescales, proposing that the configuration space of living systems can grow rapidly via combinatorial innovation. The TAP equation captures how the diversity of possible structures increases super-exponentially as new forms of organization arise, each creating new possibilities for further novelty.

The swimbladder example from the 2024 interview illustrates this concept with particular clarity. The swimbladder — a gas-filled organ that allows bony fish to control their buoyancy — evolved, according to the fossil record, from lungs of lungfish that were transitioning between puddles. Crucially, once the swimbladder existed, it constituted an empty ecological niche that could be colonized by organisms specifically adapted to live within it — such as bacteria or worms that could evolve to inhabit this new environment. Crucially, natural selection did not create the swimbladder as a niche; it created a functional organ. The creation of the niche was an unselected consequence — a new direction of evolution opened without any selection pressure having directed it.

2.3 No Entailing Laws, But Enablement

The philosophical core of Kauffman’s radical emergence thesis is articulated most sharply in No entailing laws, but enablement in the evolution of the biosphere (Longo, Montévil, and Kauffman, 2012, arXiv:1201.2069). The paper argues that the fundamental difference between the physical and biological sciences lies in the presence of entailing laws in physics — laws that, given initial conditions, strictly determine or constrain the possible outcomes — versus the enablement framework of biology, in which pre-existing conditions make certain outcomes possible but do not determine which of those possibilities will actualize.

In physics, symmetries and conservation principles constrain possible trajectories through state space. In biology, no such entailing laws govern the actualization of novel functions and forms. The phase space of biological possibilities is not merely vast — it is fundamentally unprestatable. There is no algorithm or computation that could have listed, prior to the existence of eukaryotic cells, all the possible functions and structures that would arise from them.

This point connects directly to Kauffman’s claim, cited in the 2024 interview:

«Not only do we not know what will happen — like when you flip a coin 10,000 times and you don’t know if it’ll come up heads 4,423 times — we don’t even know what can happen in the evolving biosphere and in the economy and in life. […] That means that reason is an insufficient guide for living your life. It means we need reason, emotion, intuition, sensation, metaphor.»

2.4 Life as a Phase Transition: Kauffman and Roli (2024)

In their 2024 paper Is the Emergence of Life an Expected Phase Transition in the Evolving Universe? (arXiv:2401.09514), Kauffman and Roli synthesize the theory of Collectively Autocatalytic Sets (RAF) with the Theory of the Adjacent Possible to propose that the origin of life represents an expected — but not predeterminable — phase transition in the evolving universe. The RAF framework, which Kauffman introduced in 1996, shows that sets of molecules capable of catalyzing each other’s formation can arise spontaneously in chemical systems above a certain threshold of diversity and connectivity. Once such a set exists, it achieves what the authors call catalytic closure — a state in which the system as a whole is self-sustaining.

The paper further proposes that life achieves three forms of closure relevant to defining a Kantian Whole: catalytic closure (the system’s components mutually sustain each other), constraint closure (the system constrains its own boundary conditions), and spatial closure (the system occupies a defined spatial region as a coherent whole). Importantly, the authors argue that this phase transition is expected given the right chemical conditions — but what those conditions are, and which specific life forms will actualize from them, cannot be deduced in advance.

3. Epistemological vs. Ontological Emergence: The Philosophical Distinction

3.1 Definitions

The distinction between epistemological and ontological emergence — central to Kauffman’s framework — can be summarized as follows:

Epistemological emergence refers to phenomena that are complex and may appear surprising from the perspective of an observer who lacks full information, but which are in principle deducible from the underlying components and their interactions. The paradigm case is the Boolean Network: although the system’s dynamics may be hard to compute, the state space is prestatable, and the behavior can in principle be derived from the network’s structure and rules.

Ontological emergence (or radical emergence) refers to phenomena that are genuinely novel in the sense that they cannot be predicted or deduced in principle prior to their occurrence. The paradigm case is a Darwinian pre-adaptation: a trait that evolves for one function (e.g., the swimbladder for buoyancy control) creates a new ecological niche (for organisms that can live inside it), and this new niche opens an evolutionary direction that did not exist before — not because the niche was selected for, but because it came to exist as a side effect of an unrelated adaptation.

3.2 Kauffman’s Formulation

In the 2024 interview, Kauffman articulates the distinction clearly:

«Epistemological emergence is like the Boolean Networks. Yes. And ontological emergence is the worm coming to live in the swimbladder. Right, because that’s unpredictable. It’s impossible to have predicted ahead of time — with the swimbladder, or with Facebook for that matter — those are true emergent radical ontological sense of being kinds of emergence.»

McKenzie (2025, arXiv:2508.08548) provides an extensive 163-page review of emergence across physics, biology, sociology, and computer science, noting that the irreducibility of emergent phenomena is a defining characteristic: wholes have properties that their individual parts do not possess, and these properties cannot be derived from a knowledge of the parts alone. This resonates with Kauffman’s distinction: the question of whether the irreducibility is merely epistemic (a limitation of our knowledge or computational capacity) or ontological (a genuine feature of the world itself) remains one of the deepest open questions in the philosophy of science.

3.3 The Consciousness Case

Kauffman’s skepticism about consciousness as emergent also illustrates the distinction. The claim that «if you have enough neurons, consciousness will emerge» is, he argues, an example of the loose use of the word «emerge.» His position is nuanced:

«People also say, on the topic of consciousness, that if you have enough neurons, consciousness will emerge. I think it could be right, but I still think it’s gobbledygook. […] [E]ven if consciousness does arise with sufficient neural complexity, the question is whether that emergence is epistemological or ontological. If it’s merely that we lack the computational tools to deduce consciousness from neuronal dynamics, then it’s epistemological. If consciousness is genuinely novel — if it introduces causation that cannot be reduced to or predicted from neuronal activity — then it is ontological.»

Gerhard Werner (2011, arXiv:1103.2366) approaches consciousness through the lens of brain phase space dynamics, criticality, and renormalization group concepts, proposing that consciousness arises when neural systems operate near critical points — analogous to phase transitions in physical systems. This suggests that at least some forms of emergence in neural systems may be epistemological in character, even if they appear radical to our current understanding.

4. Biological Case Studies: Ant Colonies and Collective Intelligence

4.1 Ant Colonies as Complex Adaptive Systems

Ant colonies represent one of the most thoroughly studied examples of collective intelligence emerging from simple individual behaviors. An individual ant follows simple rules based on local pheromone gradients and tactile cues. No individual ant possesses a blueprint of the collective structure — the nest, the foraging trails, the fungus farms, or the slave-making raids of certain species. Yet from the interactions of tens of thousands of individuals, complex, goal-directed collective behavior emerges robustly and reliably.

Knar (2025, arXiv:2503.18858) examines what he terms «Insect Paraintelligence» — the phenomenon by which a mindless colony of ants meaningfully moves a beetle or other large object in a coordinated, purposeful fashion, with the collective exhibiting behavior that none of its individual members could accomplish or even comprehend. The paper argues that this represents a genuine case of meaningful collective action without individual comprehension — a form of distributed cognition that is irreducible to the individual ant level.

The ant colony therefore provides a biological case study in both epistemological and ontological emergence, depending on how one interprets the limits of deducibility. From one perspective, ant colony behavior can in principle be modeled as a complex adaptive system with local rules — making it epistemologically emergent. From another perspective, the specific form and direction of colony-level adaptations (e.g., the evolution of specific caste behaviors, slave-making, or fungal agriculture) may represent genuinely unprestatable novelties — aligning with ontological emergence.

4.2 Self-Organization and the Adjacent Possible in Insect Societies

The concept of the adjacent possible applies to ant societies in profound ways. The evolution of leaf-cutting ant fungus farming, for example, created a new ecological niche: a mutualistic relationship between ants and a specific fungus, which in turn created further niches for specialized Escovopsis mold that attacks the fungus, which in turn may have driven the evolution of specialized bacterial cultures that the ants cultivate as antibiotics. Each step of this nested chain opened possibilities that did not exist before, following the TAP logic that Kauffman describes.

Macktoobian (2022, arXiv:2210.03975) studies self-organizing nest migration dynamics in ant colony systems, demonstrating that even the physical relocation of an entire colony — a collective decision involving thousands of individuals — can be achieved through entirely decentralized mechanisms with no central planner. This underscores the TAP insight: collective behaviors and structures arise not from a design but from the activation of possibilities latent in the interaction rules of the system.

4.3 Ant Colony Optimization as a Computational Paradigm

The biological principles underlying ant colony behavior have inspired Ant Colony Optimization (ACO), a computational paradigm for solving combinatorial optimization problems. The core insight is that simple agents following local pheromone-based rules can collectively solve problems — such as the traveling salesman problem — that are computationally intractable through brute-force methods (Dorigo and Stützle, 2019).

In a 2020 paper on ant pheromone-based communication applied to bubble column reactor prediction (arXiv:2001.04276), Shamshirband et al. demonstrate that ACO-derived models can predict complex fluid dynamics with accuracy comparable to neural network approaches. And in a 2024 paper, Pandey and Thakur introduce D-CODE: Data Colony Optimization for Dynamic Network Efficiency (arXiv:2405.15795), extending colony-inspired algorithms to optimize communication networks. These applications illustrate how the emergent collective intelligence of ant colonies has been abstracted into a computational framework that, in turn, generates novel solutions to engineering problems — another instance of the adjacent possible at work.

5. The Origin of Life as Radical Emergence

5.1 The Phase Transition from Chemistry to Biology

The origin of life represents the most profound example of radical emergence in the natural world. Kauffman and Roli (2024) argue that life emerged as a phase transition in the evolving universe — a transition that was in some sense expected (given the right chemical conditions) but whose specific outcome was not predeterminable.

This perspective unites two major research programs: the RNA World hypothesis, which proposes that self-replicating RNA molecules preceded the evolution of protein synthesis and DNA-based genetics; and the Autocatalytic Sets (RAF) framework, which Kauffman pioneered and which shows that collections of molecules capable of catalyzing each other’s formation can arise spontaneously in sufficiently complex chemical systems.

5.2 The RNA-Protein World and the Genetic Code

Kauffman and Lehman (2022, arXiv:2208.01491) investigate the intersection between peptide and RNA autocatalytic sets, arguing that the origin of the genetic code itself may have been a phase transition — a moment at which the system moved from a regime in which proteins and RNA replicators coexisted independently to one in which the code created an irreversible linkage between genotype and phenotype. This transition, they argue, was radical in the sense that the code itself created possibilities for biological information processing that did not exist prior to its establishment.

5.3 The 2026 Multi-Author Review

A landmark 2026 multi-author review published as The Origin of Life in the Light of Evolution (arXiv:2605.05464), authored by an international team including Betül Kaçar, Tom A. Williams, Laura Eme, and Eugene V. Koonin, among others, examines the origin of life through an explicitly evolutionary lens. The paper addresses the emergence of the first cells, the origin of the genetic code, the evolution of metabolism, and the deep-time perspective on how life diversified from a common ancestor. While not framed explicitly in Kauffman’s radical emergence terminology, the review documents the profound contingency of early life — the fact that the actual path from chemistry to biology could have been one among countless possibilities, none of which could have been predicted from first principles.

5.4 Combinatorial Innovation and the Expanding Possibility Space

The 2019 paper by Weller-Davies, Steel, and Hein (arXiv:1910.09051) provides mathematical foundations for autocatalytic network models, demonstrating that the number of possible autocatalytic sets in a chemical reaction network grows combinatorially with the number of molecular species. This mathematical result supports Kauffman’s core claim: the space of possible biological organizations is not merely large — it is fundamentally open-ended and unprestatable.

6. Economic Emergence: Radical Novelty in Human Systems

6.1 The Economy as an Autocatalytic Set

One of the most striking extensions of Kauffman’s framework is the modeling of economic systems as autocatalytic sets. In Emergence of Autocatalytic Sets in a Simple Model of Technological Evolution (Hordijk and Kauffman, 2022, arXiv:2204.01059), the authors demonstrate that production networks in an economy exhibit the same autocatalytic properties as chemical reaction networks. Just as a set of chemicals can mutually catalyze each other’s formation, a set of economic activities can mutually sustain and enable each other’s existence.

This insight directly supports Kauffman’s analysis of the Facebook example:

«Nobody foresaw Facebook when the computer was invented. So go from computer to personal computer to Microsoft to saving and storing files to the web to selling things on the web to Facebook. How did that happen? These all got enabled by and were new economic niches that were created by what happened beforehand becoming into the adjacent possible. That’s radical emergence.»

Each step in this chain — personal computing, graphical interfaces, file systems, the internet, e-commerce — opened new economic possibilities that were not anticipated even by the creators of the prior step. Microsoft did not plan for Facebook; the developers of ARPANET did not plan for social media. Yet each stage created the adjacent possible for the next.

6.2 The Implication for Economic Theory

The radical emergence framework poses a fundamental challenge to rational expectations economics — the assumption that economic agents can, and do, anticipate future developments based on available information. If genuine radical emergence occurs in economic systems, then there are structural limits to prediction that no amount of computational power or information can overcome. This resonates with Friedrich Hayek’s insight that the dispersed, tacit knowledge of millions of economic actors cannot be centralized or aggregated — and goes further, suggesting that the very structure of economic possibility space is open-ended and self-creating.

7. Emergent Abilities in Large Language Models: Parallels and Tensions

7.1 Recap: Emergent Abilities in LLMs

As documented in the companion review Emergent Abilities in Large Language Models: An Extensive Academic Review (2022–2026), the machine learning community has documented numerous cases in which capabilities appear abruptly in LLMs as model scale increases — capabilities that were not present in smaller models and were not explicitly programmed. Wei et al. (2022, arXiv:2206.07682) define emergent abilities as those that appear above a certain scale threshold, cannot be predicted by extrapolating the performance of smaller models, and include chain-of-thought reasoning, multi-step arithmetic, and abstract reasoning.

7.2 Are LLM Emergent Abilities Epistemological or Ontological?

The parallels between LLM emergence and Kauffman’s radical emergence are thought-provoking but require careful distinction.

On one interpretation, LLM emergence is epistemological: the model’s capabilities exist latently in the weights and architecture, and the appearance of new abilities with scale reflects the crossing of a threshold in a prestatable (if computationally intractable) system. This is analogous to a Boolean Network: the state space is, in principle, determined by the architecture, the training data, and the learning algorithm. If one had complete knowledge of the weights, the architecture, and the training dynamics, one could — in principle — predict exactly which abilities would emerge and at what scale. The emergence would be epistemological.

On another interpretation — which aligns with recent analysis by Havlík (2025, arXiv:2508.04401), who examines emergent abilities in LLMs from an epistemological perspective — the situation may be more nuanced: emergent abilities in LLMs challenge the assumption that machine capabilities can be fully derived from first principles given the complexity of the training process.

7.3 The Adjacent Possible in AI Development

The concept of the adjacent possible offers a compelling framework for understanding AI development trajectories. Consider the path from early language models (ELMs) to GPT-4: each step in capability improvement opened possibilities for new applications and new research directions that were not anticipated at the time of the prior step. The transformer architecture (Vaswani et al., 2017) was developed for translation tasks; its application to language modeling and eventual scaling to GPT-4 opened possibilities — agentic reasoning, tool use, code generation — that were not anticipated by the original researchers.

This echoes Kauffman’s TAP: the AI ecosystem creates its own future possibilities of becoming. The existence of GPT-4 created the adjacent possible for GPT-4.5 and o3, for Claude 3 and 4, for Gemini 2.0, and for the emerging generation of reasoning models. Each model creates niches of possibility for the next.

7.4 The Debate: Real Emergence vs. Metric Mirage

Critically, the literature on LLM emergence remains divided on whether the observed discontinuities are genuine or methodological artifacts. Schaeffer et al. (2023, arXiv:2304.15004) argue that apparent emergent abilities may be mirages arising from discontinuous evaluation metrics — a position that would place LLM emergence firmly in the epistemological category. Wei et al. (2022) maintain that genuine discontinuities exist in capability acquisition, even accounting for metric effects.

This debate has a direct parallel in Kauffman’s distinction: the question of whether the emergence is epistemological (a reflection of our ignorance) or ontological (a genuine feature of the possibility space) is as unresolved in AI as it is in biology.

8. Evaluating Kauffman’s Hypothesis: Validation, Critique, and Open Questions

8.1 Evidence in Support

Several lines of evidence support, or are consistent with, Kauffman’s radical emergence thesis:

The unprestatability of biological novelty: No one predicted the existence of eukaryotic cells, the evolution of flight in insects, or the emergence of consciousness prior to their occurrence. While absence of prediction is not proof of unpredicatability, the pattern across evolutionary history suggests that the space of possible biological forms and functions is vastly larger than any prestatable catalog.

Autocatalytic sets in chemical systems: The RAF framework has been experimentally validated in simple chemical systems (such as groups of mutually catalytic peptides or RNA fragments), demonstrating that self-sustaining catalytic networks can arise spontaneously (Kauffman, 1996; Weller-Davies et al., 2019).

Economic autocatalytic sets: The 2022 Hordijk-Kauffman paper demonstrates that production networks exhibit autocatalytic structure, providing a formal bridge between biological and economic emergence (Hordijk and Kauffman, 2022, arXiv:2204.01059).

The expanding possibility space: The biocosmology paper (Cortês et al., 2022, arXiv:2204.09378) formalizes the TAP equation, showing that the space of possible living structures can grow super-exponentially through combinatorial innovation.

Phase transition framing: The 2024 Kauffman-Roli paper’s proposal that life is a phase transition — expected but not predeterminable — provides a physically grounded way to understand the appearance of life as a natural but fundamentally unpredictable event.

8.2 Critiques and Limitations

The radical emergence hypothesis faces significant challenges:

Falsifiability: One of the central criticisms of Kauffman’s framework is that it may be difficult or impossible to falsify. If a critic claims that a particular emergence was predictable, Kauffman can respond that the prediction was made post hoc. This epistemological issue is addressed, though not fully resolved, by the emphasis on the principle that genuinely unprestatable events share certain formal properties (inability to be deduced in advance, creation of new possibility space, irreversible change in the trajectory of the system).

The role of selection: Evolutionary biologists have pointed out that the concept of «enablement» without selection pressure is controversial. The standard view is that selection acts on variation, but does not create variation — yet the variation itself may be structured by prior selection in ways that constrain future possibilities. Kauffman’s response is that the creation of new niches (the swimbladder as an empty niche for bacteria) is not driven by selection and represents genuine radical emergence.

Alternative explanations: Some phenomena attributed to radical emergence may have epistemological explanations. For example, consciousness might emerge from neural criticality through a mechanism analogous to a phase transition — one that is, in principle, explicable through neuroscience and physics (Werner, 2011).

The limits of the analogy with Boolean Networks: Critics note that the comparison between Boolean Networks (where the state space is prestatable) and biological systems (where Kauffman claims it is not) depends on accepting that biological systems differ fundamentally from computational systems in their relationship to possibility space. This remains contested.

8.3 Open Questions

Can the distinction between epistemological and ontological emergence be made rigorous? Philosophical work by scholars such as David Chalmers, John Collier, and Robert Laughlin continues to grapple with this question. The development of formal tools to distinguish prestatable from unprestatable emergence remains an active research frontier.

Is there a physical basis for radical emergence? The connection to thermodynamics (the creation of new structures that reduce entropy locally while increasing it globally) and to quantum mechanics (the role of indeterminate events at the quantum level in shaping macroscopic biological outcomes) remains underexplored.

Does radical emergence apply to AI? The question of whether LLM emergent abilities represent genuine ontological emergence (truly unprestatable capabilities that create new possibility spaces for subsequent model development) or merely epistemological emergence (complex but in principle deducible from the architecture and training data) is unresolved and may be undecidable with current tools.

9. Implications for the Nature of Science and Human Understanding

9.1 Beyond the Enlightenment Model

Kauffman’s most radical claim is not scientific but philosophical: that the Enlightenment’s faith in reason as a sufficient guide to understanding nature and guiding human life is fundamentally mistaken. As he states in the 2024 interview:

«Not only do we not know what will happen… we don’t even know what can happen… in the evolving biosphere and in the economy and in life. That means something huge for us as humans: it means that reason is an insufficient guide for living your life. It means we need reason, emotion, intuition, sensation, metaphor. It means what just happened to the Enlightenment, where reason is our hero. It’s insufficient. Life is much richer than we thought.»

This claim resonates with the broader post-Enlightenment critique developed by philosophers from Nietzsche to Wittgenstein to Hume — and yet it is grounded in rigorous mathematical and empirical work on complex systems, autocatalysis, and the origin of life.

9.2 Implications for Artificial Intelligence Research

The radical emergence framework has direct implications for AI research:

Unpredictability of capabilities: If radical emergence applies to AI, then we should expect that future models will develop capabilities that we cannot anticipate or deduce in advance — raising both opportunities and safety concerns.

Limits of scaling laws: Standard scaling laws (Kaplan et al., 2020; Hoffmann et al., 2022) describe aggregate performance trends but may not capture the appearance of genuinely novel capabilities, which may emerge discontinuously and unpredictably.

The creation of new possibility spaces: Each new AI capability creates the adjacent possible for subsequent capabilities, suggesting that the trajectory of AI development may be partly self-generating and partly shaped by the unprestatable emergence of novel functions.

10. Conclusions

This review has traced the intellectual trajectory of Stuart Kauffman’s concept of radical emergence — from its origins in the study of Random Boolean Networks and autocatalytic sets, through its elaboration in the Theory of the Adjacent Possible, to its most recent synthesis in the 2024 paper with Andrea Roli on life as a phase transition.

The evidence supports, at minimum, the following conclusions:

Biological systems exhibit forms of emergence that are not fully captured by the model of entailing laws. The creation of new ecological niches as side effects of adaptation (the swimbladder), the evolution of novel functions through pre-adaptation, and the apparent openness of the space of possible biological forms and functions are well-documented phenomena that resist reduction to deterministic prediction.

The distinction between epistemological and ontological emergence is philosophically valuable, even if it remains difficult to operationalize. It directs attention to a real question — whether the unpredictability of emergent phenomena reflects the limits of our knowledge or the nature of the world — that cannot be resolved without a deeper understanding of the relationship between computation, information, and physical law.

Ant colonies and collective biological systems provide compelling case studies in both self-organization and radical novelty. The emergence of collective intelligence in ant colonies, the evolution of nested mutualisms in insect societies, and the origin of life as a phase transition from chemistry to biology all illustrate the TAP framework at work.

The parallel with emergent abilities in large language models is suggestive but not conclusive. LLM emergence may be epistemological (a consequence of the complexity of the training process) or ontological (a genuine creation of new possibility spaces). The question is deeply open.

Kauffman’s radical emergence thesis remains scientifically productive even if it is not definitively validated. It generates testable predictions about the structure of biological possibility space, the behavior of autocatalytic systems, and the relationship between novelty and selection. Whether these predictions are sustained by future research will determine the empirical fate of the framework.

The deepest implication of radical emergence, however, is not scientific but existential: if the biosphere — and the economy, and perhaps the AI ecosystem — genuinely creates its own future possibilities of becoming, then neither reason alone nor any other single cognitive faculty is sufficient to navigate the richness of what is possible. The task of understanding emergence may itself be an emergent process — one that requires the full range of human capacities, including those we have not yet named.

References (APA 7th Edition Format)

Cortês, M., Kauffman, S. A., Liddle, A. R., & Smolin, L. (2022). Biocosmology: Towards the birth of a new science (arXiv:2204.09378). arXiv. https://doi.org/10.48550/arXiv.2204.09378

Dorigo, M., & Stützle, T. (2019). Ant colony optimization: Overview and recent advances. In R. G. Pérez (Ed.), Handbook of metaheuristics (pp. 311–351). Springer.

Havlík, V. (2025). Why are LLMs’ abilities emergent? (arXiv:2508.04401). arXiv. https://doi.org/10.48550/arXiv.2508.04401

Hoffmann, J., Borgeaud, S., Mensch, A., et al. (2022). Training compute-optimal large language models. Proceedings of the 36th International Conference on Neural Information Processing Systems (NeurIPS 2022).

Hordijk, W., & Kauffman, S. A. (2022). Emergence of autocatalytic sets in a simple model of technological evolution (arXiv:2204.01059). arXiv. https://doi.org/10.48550/arXiv.2204.01059

Kaçar, B., Williams, T. A., Eme, L., Gogarten, J. P., Sanchez-Baracaldo, P., Spang, A., Aylward, F. O., Travisano, M., Welander, P. V., Huber, J. A., Cooper, V. S., Turner, P. E., Lyons, T. W., Ellington, A. D., Copley, S. D., Koonin, E. V., & Lynch, M. (2026). The origin of life in the light of evolution (arXiv:2605.05464). arXiv. https://doi.org/10.48550/arXiv.2605.05464

Kaplan, J., McCandlish, S., Henighan, T., et al. (2020). Scaling laws for neural language models. arXiv preprint arXiv:2001.08361.

Kauffman, S. A. (1993). The origins of order: Self-organization and selection in evolution. Oxford University Press.

Kauffman, S. A. (1996). At home in the universe: The search for laws of self-organization and complexity. Oxford University Press.

Kauffman, S. A., & Lehman, S. (2022). Mixed anhydrides at the intersection between peptide and RNA autocatalytic sets (arXiv:2208.01491). arXiv. https://doi.org/10.48550/arXiv.2208.01491

Kauffman, S. A., & Roli, A. (2024). Is the emergence of life an expected phase transition in the evolving universe? (arXiv:2401.09514). arXiv. https://doi.org/10.48550/arXiv.2401.09514

Knar, E. (2025). Dynamics of insect paraintelligence: How a mindless colony of ants meaningfully moves a beetle (arXiv:2503.18858). arXiv. https://doi.org/10.48550/arXiv.2503.18858

Longo, G., Montévil, M., & Kauffman, S. A. (2012). No entailing laws, but enablement in the evolution of the biosphere (arXiv:1201.2069). arXiv. https://doi.org/10.48550/arXiv.1201.2069

Macktoobian, M. (2022). Self-organizing nest migration dynamics synthesis for ant colony systems. Natural Computing, 21(4), 633–648. https://doi.org/10.1007/s11047-022-09923-0

McKenzie, R. H. (2025). Emergence: From physics to biology, sociology, and computer science (arXiv:2508.08548). arXiv. https://doi.org/10.48550/arXiv.2508.08548

Pandey, S., & Thakur, G. S. (2024). D-CODE: Data colony optimization for dynamic network efficiency (arXiv:2405.15795). arXiv. https://doi.org/10.48550/arXiv.2405.15795

Schaeffer, R., Miranda, B., & Koyejo, S. (2023). Are emergent abilities of large language models a mirage? (arXiv:2304.15004). arXiv. https://doi.org/10.48550/arXiv.2304.15004

Shamshirband, S., Babanezhad, M., Mosavi, A., et al. (2020). Prediction of flow characteristics in the bubble column reactor by the artificial pheromone-based communication of biological ants (arXiv:2001.04276). arXiv. https://doi.org/10.48550/arXiv.2001.04276

Vaswani, A., Shazeer, N., Parmar, N., et al. (2017). Attention is all you need. Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS 2017), 5998–6008.

Wei, J., Tay, Y., Bommasani, R., Raffel, C., Zoph, B., Borgeaud, S., Yogatama, D., Bosma, M., Zhou, D., Metzler, D., Chi, E. H., Hashimoto, T., Vinyals, O., Liang, P., Dean, J., & Fedus, W. (2022). Emergent abilities of large language models (arXiv:2206.07682). arXiv. https://doi.org/10.48550/arXiv.2206.07682

Weller-Davies, R., Steel, M., & Hein, J. (2019). Combinatorial results for network-based models of metabolic origins (arXiv:1910.09051). arXiv. https://doi.org/10.48550/arXiv.1910.09051

Werner, G. (2011). Consciousness viewed in the framework of brain phase space dynamics, criticality, and the renormalization group (arXiv:1103.2366). arXiv. https://doi.org/10.48550/arXiv.1103.2366

Document generated with AI assistance. Date of preparation: May 2026.
Research based on arXiv, Semantic Scholar, peer-reviewed publications, and the Stuart Kauffman interview transcript.
Companion document: «Emergent Abilities in Large Language Models: An Extensive Academic Review (2022–2026)» — blog.yucas.net/2026/05/11/emergent-abilities-large-language-models-llms-review-2026/