What Lives? A meta-analysis of diverse opinions on the definition of life

Abstract: The question of "what is life?" has challenged scientists and philosophers for centuries, producing an array of definitions that reflect both the mystery of its emergence and the diversity of disciplinary perspectives brought to bear on the question. Despite significant progress in our understanding of biological systems, psychology, computation, and information theory, no single definition for life has yet achieved universal acceptance. This challenge becomes increasingly urgent as advances in synthetic biology, artificial intelligence, and astrobiology challenge our traditional conceptions of what it means to be alive. We undertook a methodological approach that leverages LLMs to analyze a set of definitions of life provided by a curated set of cross-disciplinary experts. We used a novel pairwise correlation analysis to map the definitions into distinct feature vectors, followed by agglomerative clustering, intra-cluster semantic analysis, and t-SNE projection to reveal underlying conceptual archetypes. This methodology revealed a continuous landscape of the themes relating to the definition of life, suggesting that what has historically been approached as a binary taxonomic problem should be instead conceived as differentiated perspectives within a unified conceptual latent space. We offer a new methodological bridge between reductionist and holistic approaches to fundamental questions in science and philosophy, demonstrating how computational semantic analysis can reveal conceptual patterns across disciplinary boundaries, and opening similar pathways for addressing other contested definitional territories across the sciences.

• The paper introduces an AI-driven meta-analysis that quantitatively maps 68 expert definitions to uncover the semantic field of life.
• It employs LLM models, hierarchical clustering, and t-SNE to reveal eight distinctive clusters, reflecting historical and conceptual debates.
• The findings challenge binary life definitions by demonstrating continuous gradients and cross-disciplinary convergence in understanding life.

Computational Meta-Analysis of "What Lives?": The Semantic Topology of Life Definitions

Introduction

The persistent challenge of defining “life” has crosscut philosophy and science for centuries, producing not only a spectrum of definitions but a landscape of irreducible conceptual tensions. The paper "What Lives? A meta-analysis of diverse opinions on the definition of life" (2505.15849) presents a rigorous, AI-driven exploration into how contemporary experts across biology, physics, computation, and philosophy demarcate and conceptualize life. The authors deploy advanced LLM-based comparative analysis, semantic clustering, and nonlinear dimensionality reduction on 68 authoritative definitions, yielding both a refined structural cartography of the space and insights into its enduring fracture lines. This work constitutes a methodological innovation, integrating manual curation with model-averaged clustering, and provides a platform for comparative definitional ontology.

Historical and Conceptual Context

The analysis situates itself against a deep intellectual backdrop, from Aristotelian teleology’s nested functional capacities to Cartesian and Hobbesian mechanist reductions, and extending through 20th-century advances in thermodynamics, molecular biology, and algorithmic self-replication. The persistent schism between substance/property ontologies (cells, molecules, or organizational states) and process/function ontologies (self-organization, information processing, agency) recapitulates in modern expert discourse. More recently, the proliferation of artificial life, synthetic biology, and cognitive-computational frameworks has rendered traditional categorical definitions—those anchored in metabolism, genetics, or cellularity—insufficient as both empirical and theoretical advances granularize and amalgamate the conceptual field.

Methodology

The methodological innovation is grounded in combining curated expert input with systematic semantic analysis using three leading LLMs (Claude 3.7 Sonnet, GPT-4o, Llama 3.3 70B Instruct). Each model independently scored all pairwise definition similarities via constrained, numerical outputs in a protocol designed to quantify the degree of semantic overlap or contradiction (-1 to 1). Multiple replicates per pair and per model were aggregated to form robust, symmetrized correlation matrices.

Agglomerative hierarchical clustering of this data generated discrete yet interrelated conceptual clusters, while t-SNE projection embedded the high-dimensional semantic distances into a tractable two-dimensional topography. Further, LLMs provided automated yet interpretable intra-cluster thematic analyses and consensus definitions reflecting the dominant conceptual schemas rather than arbitrary centroids.

Results: Semantic Clustering and Archetypal Definitions

Eight robust clusters emerged, each corresponding to a discernible epistemic frame or operational definition:

• Perceptual Categorization: Life as an emergent property of observer-relative categorization, highlighting radical constructivist and anti-realist positions.
• Self-Sustaining Dynamic Patterns: Life as dynamic, recursive self-sustaining patterns with phenomenological emphasis.
• Dynamic Relational Process: Emphasis on boundary conditions of relationality and non-isolated process—life as continuous movement and exchange.
• Pragmatic Definitional Skepticism: Definitions foregrounding the futility or context-dependence of universal criteria.
• Cognitive Autonomy: Life as autonomous, goal-directed agency, integrating cybernetic, informational, and adaptive properties, functioning as a high-density attractor for cross-disciplinary convergence.
• Dissipative Self-Organizing Systems: Thermodynamically grounded definitions stressing entropy reduction, boundary maintenance, and far-from-equilibrium organization.
• Informational Self-Replication: Strict reductionist variants focused on replicative informational substrates (e.g., von Neumann-style automata, genetic systems).
• Self-Replicating Thermodynamic Systems: Integrative classical definitions requiring both thermodynamic and reproductive capacities, bridging chemical, evolutionary, and physicalist perspectives.

Notably, 66% of all definitions congregated within the high-density interface of the “Cognitive Autonomy” and “Dissipative Self-Organizing Systems” clusters—implying an emerging, but not universal, cross-disciplinary synthesis around agency, self-organization, information-processing, and organizational homeostasis as core features.

Emergent Topological Dimensions

The t-SNE projection revealed two principal semantic axes:

1. Observer-dependent to Objective/Structural: Spanning frameworks from radical observer-relativity to materialist, operational definitions.
2. Process-based to Entity-based: Tracing the axis from processual, teleological understandings (life as “becoming” or “doing”) to substantialist, property-centric definitions (life as “being” or “having”).

Peripheral clusters, notably Perceptual Categorization and Pragmatic Definitional Skepticism, occupied maximal conceptual distances from operational materialist definitions, with bridge definitions occupying unstable positions between the major attractors. These gradients are not merely artifacts of clustering but reflect deep underlying historical, philosophical, and disciplinary tensions.

Numerical and Analytical Claims

• High cross-model clustering consistency (71–79%) and strong matrix-to-matrix correlations (>0.7) demonstrate the robustness of the clustering architecture, independent of LLM.
• The majority of expert definitions (54%) explicitly reject a binary alive/non-alive ontology, favoring continuous gradients or multidimensional “lifeness”.
• 57% of experts’ definitions were observer-independent, with the remainder adopting some observer- or context-relativity.
• Only 12% adopted a non-actionable (e.g., poetic, reflexive) stance, while the remainder specified concrete, operationalizable criteria.

Practical and Theoretical Implications

This meta-analysis has several implications:

• For Astrobiology and Synthetic Biology: The multidimensional, continuous topology challenges the utility of rigid operational definitions (e.g., the NASA definition), indicating a need for probabilistic, multi-feature assessment frameworks for non-terrestrial or synthetic biosystems.
• For Artificial Intelligence and ALife: The persistent centrality of agency, information-processing, and goal-directed self-maintenance as defining axes implies that future advances in embodied AI or cognitive robotics may push the theoretical boundaries of the definitional space, especially as systems acquire higher-order autonomy and self-modifying capabilities.
• For Philosophy of Biology: The quantification of semantic bridges and boundaries illustrates that apparent disagreements are often orthogonal projections in a high-dimensional space, suggesting that cross-disciplinary communication and integrated taxonomy require computational semantic mediation rather than philosophical reductionism.
• For Definitional Methodology: The demonstrated LLM-driven framework enables scalable, rapid semantic mapping across scientific domains, allowing for more granular metascientific analysis of “contested” or “essentially contested” concepts.

Limitations

The curated set of expert definitions, while diverse, is not statistically representative of all disciplinary vocabularies. The LLM-driven correlation may introduce subtle model-specific biases, and the computational complexity (quadratic scaling in number of definitions) currently limits large-scale expansion without self-hosting. Dimensionality reduction for visualization, while useful for interpretation, necessarily loses some high-dimensional structure.

Future Directions

Scaling this methodology to thousands of definitions across global or field-specific cohorts would enable tracking of disciplinary convergence and divergence in real time. Critically, the framework is extensible to other domains with persistent epistemic fragmentation, such as “cognition”, “consciousness”, or “intelligence”, potentially resolving or reformulating debates via quantitative topologies rather than static, monolithic definitions.

The trajectory of synthetic biology and AI strongly suggests an urgent need for flexible, continuous, and multi-criteria frameworks for defining and recognizing life-like phenomena, both for scientific inquiry and for policy/bioethical adjudication.

Conclusion

This paper delivers a computational meta-analysis that rigorously maps the semantic field of “life” definitions as a continuous latent space with discernible attractors, bridges, and gradients. Rather than enforcing consensus or dichotomy, the analysis reveals a structured topology reflecting persistent philosophical dualities, emergent interdisciplinary convergence, and robust conceptual archetypes. This framework offers a replicable paradigm for meta-analytic mapping of complex biological and cognitive domains, with tangible implications for the future of bioengineering, AI, and theoretical biology.

Reference:

“What Lives? A meta-analysis of diverse opinions on the definition of life” (2505.15849).