Un-prestatability

In the AI Story

The distinction between unpredictable and un-prestateable rests on whether the space of possible outcomes can be defined in advance. Quantum mechanics makes particle positions unpredictable but operates within a well-defined Hilbert space—the space of possibilities is known even when specific outcomes cannot be forecast. Evolutionary innovation is un-prestateable because the space itself evolves: each genuine innovation creates functional possibilities that did not exist in the prior selective environment. The swim bladder could not have been listed among possible lung-futures because 'buoyancy-regulating organ' was not a functional category until body morphology, predation patterns, and habitat structure co-evolved to create the niche that category fills.

Kauffman and Roli's 2026 paper applied this framework directly to AI, arguing that large language models produce outputs that are unpredictable—the specific text generated by a given prompt cannot be forecast—but not un-prestateable. The model recombines elements of its training corpus in novel configurations, but those configurations belong to a possibility space pre-defined by the corpus and architecture. The model explores; it does not expand. Genuine creativity, by contrast, expands the possibility space through the perception of novel affordances: seeing that a telephone network could carry data, that a search engine could train on behavioral data to become an advertising platform, that a language model could become a collaborative thinking partner. These are un-prestateable insights because they perceive relationships between existing elements and emergent needs that were not encoded in any prior description.

The practical implication cuts against every prediction-based planning framework. If future configurations are genuinely un-prestateable—not merely unknown but unknowable in advance—then strategies that enumerate future states and prepare for enumerated scenarios are structurally inadequate. The corporate three-year roadmap, the educational curriculum designed around future job categories, the regulatory framework built for anticipated AI applications—all operate on the premise that futures can be prestated. Kauffman's mathematics show this premise fails for complex evolving systems. The alternative is not abandoning strategy but shifting from prediction strategies (enumerate futures, choose the best) to enablement strategies (build capacities robust across un-prestateable configurations).

The epistemological modesty this framework demands is uncomfortable. It forecloses the expert's claim to foresee—not because experts lack knowledge but because the future they are asked to foresee does not yet exist to be known. The physicist cannot prestate the technologies that will emerge from quantum mechanics. The biologist cannot prestate the species that will evolve from current populations. The AI researcher cannot prestate the uses to which language models will be put, because those uses will emerge from the collision of technological capability with human needs that are themselves evolving. Un-prestatability is not ignorance. It is the recognition that knowledge has boundaries determined not by what we have learned but by what the universe has yet to create.

Origin

The concept crystallized in Kauffman's work on evolutionary innovation and the origins of novelty. He observed that genuine evolutionary innovations—the transition from single cells to multicellularity, from water to land, from ectothermy to endothermy—create functional possibilities that could not have been specified from the prior state. Each innovation restructures the selective environment, creating niches that did not previously exist and could not have been anticipated. The formalization came in Investigations (2000), where Kauffman distinguished the prestatable (the set of configurations deducible from current knowledge) from the un-prestatable (the set whose possibility depends on as-yet-unmade combinations). The distinction became central to his critique of reductionism and his defense of the irreducible creativity of the living world.

Key Ideas

Future Non-Existence. Un-prestateable futures do not exist to be known—they will be brought into being by the same evolutionary or innovative process that the prediction attempts to forecast.

Affordance Creation. Genuine creativity perceives novel affordances (new uses of existing elements) that expand the possibility space rather than exploring it—the lung-to-swim-bladder transition as paradigm.

Un-Prestatability Versus Unpredictability. Unpredictable outcomes belong to known possibility spaces; un-prestateable outcomes belong to spaces that do not yet exist—a categorical rather than gradual distinction.

Limits of Algorithmic Foresight. Current AI systems recombine within defined spaces but do not expand those spaces through the perception of novel affordances—they are unpredictable, not un-prestateable.

Enablement Over Prediction. In un-prestateable landscapes, the appropriate response is not forecasting specific futures but building capacities robust across many possible futures.

Debates & Critiques

Critics argue Kauffman's framework makes genuine prediction impossible and undermines scientific ambition. Defenders respond that the framework distinguishes between what can be known (dynamics, principles, patterns) and what cannot (specific future configurations), and that attempting to prestate the un-prestateable is not ambitious science but category error. The question of whether large language models can achieve genuine un-prestatability through emergent capabilities remains empirically open—some researchers point to unexpected emergent abilities as potential counterexamples, while Kauffman maintains that recombination within a training corpus, however sophisticated, remains categorically distinct from the creation of novel affordances.