The TAP Equation
'The TAP Equation: Evaluating Combinatorial Innovation' is a 2025 paper by Smolin, Stuart Kauffman, and collaborators examining how novel possibilities emerge in complex systems. Published in the European Economic Review, the paper offers a mathematical framework for distinguishing between combinatorial processes that merely rearrange existing elements and combinatorial processes that create new categories of possibility. The combination of fire and metal did not rearrange existing objects; it created metallurgy — a new domain from which eventually flowed tools, weapons, machines, and, much later, the computing infrastructure on which AI depends. The paper argues that similar category-creating combinations operate throughout biological, technological, and economic evolution, and that understanding them requires mathematical tools that standard combinatorial analysis does not provide.
The Material Preconditions of Novelty — Contrarian ^ Opus
There is a parallel reading of combinatorial innovation that begins not with the mathematics of possibility space but with the physical substrate that determines which combinations can actualize at all. The fire-and-metal example is compelling precisely because it required specific geological conditions, energy regimes, and embodied human labor operating at particular scales. Metallurgy emerged not from an abstract exploration of the adjacent possible but from the conjunction of ore deposits accessible to surface mining, wood or charcoal supplies sufficient for sustained heat, and communities organized to support multi-year skill transmission. The category was not merely "created" by combination—it was extracted from the earth under conditions that required ecological surplus, social stability, and accumulated material culture. The mathematics of the TAP equation may describe the pattern of how new categories open, but it cannot predict which categories become available without accounting for energetic, material, and social constraints.
Applied to AI, this reading suggests that the relevant question is not whether current systems participate in category-creation through their combinatorial exploration, but whether the infrastructure supporting AI—data centers consuming Iceland's electrical output, rare earth mining operations displacing entire communities, the geopolitical arrangements securing semiconductor supply chains—constitutes the material precondition for new categories or the terminal consumption of the surplus that made previous category-creation possible. The adjacent possible is not a mathematical space; it is a physical territory whose boundaries are set by thermodynamics, resource availability, and power. If AI requires exponentially scaling material inputs to explore linearly expanding possibility spaces, the framework has described the shape of the trap, not the path to novelty.
In the AI Story
The paper builds on Kauffman's decades of work on the adjacent possible — the space of configurations accessible one step from any current configuration — and on Smolin's framework for understanding genuine novelty as emergence that is not predetermined by prior states. The specific contribution is the TAP (Theory of the Adjacent Possible) equation, which provides a mathematical treatment of how new possibilities open up as systems explore their current configurations.
The key insight is that the space of the possible is not fixed. In most combinatorial analyses, the elements are given and the combinations are bounded by the number and structure of the elements. If there are N elements, there are 2^N possible combinations, and that set is the universe of what can exist. The TAP framework challenges this assumption. As combinations are actualized, new elements become possible that were not available before. Metallurgy made possible the steam engine; the steam engine made possible industrial civilization; industrial civilization made possible the electronic computer; the electronic computer made possible AI. Each stage opened possibilities that were not adjacent to the prior stage.
Applied to AI, the framework offers a specific way to think about whether current systems participate in genuine category-creation or perform sophisticated exploration within a fixed space. Large language models explore vast spaces of linguistic combination. The spaces are large enough to produce outputs that are genuinely surprising to human observers. But the spaces are defined by the training data and the architecture; they do not grow as the model explores them. The model finds new arrangements within an existing space rather than creating new categories that expand the space.
Whether future AI systems could participate in category-creation is an open question. The framework does not rule it out — it simply distinguishes between the two kinds of combinatorial process and provides mathematical tools for recognizing the difference. The distinction matters for how to think about the current AI moment. If AI can contribute to category-creation (perhaps through collaboration with humans in ways that neither could accomplish alone), then it is an amplifier of genuine novelty. If AI can only explore existing spaces, then the category-creation must come from elsewhere — from conscious creatures engaged in the thick present in ways that current AI architectures do not replicate.
Origin
The paper was published in the European Economic Review in 2025, the culmination of sustained collaboration between Smolin and Kauffman on the mathematics of genuine novelty. It draws on Kauffman's decades of work on the adjacent possible and Smolin's framework for temporal naturalism and the reality of time.
Key Ideas
Category-creating combinations. Certain combinations — fire and metal producing metallurgy — create new categories of possibility rather than merely rearranging existing elements.
Expanding possibility space. As combinations are actualized, new elements become possible that were not available before.
Mathematical treatment. The TAP framework provides tools for analyzing how the space of the possible grows through exploration.
Distinction from recombination. Category-creation is qualitatively different from finding new arrangements within a fixed space.
Application to AI. Whether current AI participates in category-creation or only in sophisticated recombination is an open question the framework makes tractable.
Appears in the Orange Pill Cycle
Dual Constraints on Possibility — Arbitrator ^ Opus
The TAP equation and the material reading are answering different questions, and the topic benefits from holding both. On the question of whether new categories exist as a mathematical kind distinct from recombination within fixed spaces, the framework is correct (100%). Metallurgy is not a rearrangement of pre-existing objects; it is a new domain. The mathematical formalization captures something real about how complex systems generate genuine novelty, and applying this lens to AI usefully distinguishes between exploration of vast but bounded spaces (what current LLMs do) and participation in category-creation (an open question for future systems). The contrarian reading does not contest this—it shifts to a different question: which categories can actualize given physical constraints.
On that second question, the material view dominates (80%). The adjacent possible is shaped not only by what combinations are mathematically available but by what energetic, ecological, and social conditions permit. Fire and metal required specific ore accessibility, energy sources, and organized communities. The steam engine required coal deposits, iron production at scale, and capital structures capable of multi-decade investment. AI requires semiconductor fabs, rare earth supply chains, and electrical grids—none of which are given by the mathematics of combination. The TAP equation describes the topology of novelty but not the physics of its actualization. Ignoring material constraints leads to frameworks that explain the pattern of innovation while missing whether the conditions for continued innovation are being consumed or renewed.
The synthetic frame the topic needs: possibility space has dual constraints. Mathematical structure determines what new categories can emerge from current configurations. Physical substrate determines which of those categories can actualize and persist. Both are real. Both bind. Frameworks that address only one produce incomplete maps of how systems—biological, technological, economic—generate and sustain novelty over time.
Further reading
- Stuart Kauffman and Lee Smolin, 'The TAP Equation: Evaluating Combinatorial Innovation,' European Economic Review (2025)
- Stuart Kauffman, Investigations (Oxford University Press, 2000)
- W. Brian Arthur, The Nature of Technology (Free Press, 2009)