Worldmaking

In the AI Story

Goodman's pluralism was not relativism—a distinction he insisted on throughout his career. Relativism holds that all versions are equally valid, that there are no rational grounds for preferring one to another. Goodman held the opposite: some versions are right and others wrong, some better and others worse. But the criteria for evaluation are internal to the worldmaking project—coherence, fit, productivity—rather than external correspondence to a reality that exists independently of all versions. A version can be wrong by its own standards (internally incoherent, unproductive, failing to fit with other versions the worldmaker accepts) without there being a version-independent fact-of-the-matter that it fails to match. The standards are rigorous. They are just not the standards of copy-fidelity.

The operations of worldmaking—composition, weighting, deletion, deformation—are not arbitrary. They are constrained by the purposes of the worldmaking project and by the conventions of the symbol system being deployed. A painter who composes a landscape from memory is not free to arrange the elements however she wishes; the compositional choices must achieve coherence with the pictorial conventions she operates within, must fit with her other accepted versions of how landscapes are structured, must serve the purposes her particular painting is designed to serve. The constraints are real, demanding, and different for each worldmaking project. What Goodman denied was not the reality of constraints but the existence of a master constraint—correspondence to the way things really are—that all versions must satisfy.

The worldmaking framework generates a precise diagnosis of what happens when AI enters creative production. A large language model trained on human-produced versions has absorbed the statistical patterns of how humans perform the operations of worldmaking—how they compose arguments, weight evidence, delete irrelevant detail, deform conventional categories. The model can reproduce these patterns with impressive fidelity, generating outputs that look like worldmade versions. But the patterns are extracted from products of worldmaking, not from the process. The model has learned what right versions look like without learning the purposes that make versions right—the experiential grounding, the evaluative criteria, the stakes in the world that determine why this configuration of symbols rather than that one.

What AI provides is rendering at scale—the production of symbols organized according to conventions the model has absorbed. What AI cannot provide is the worldmaking intention that establishes which version among the infinite possible versions is the one worth constructing. The intention must come from an agent who inhabits the world being versioned—who has reasons, grounded in lived experience, for constructing this particular configuration. The twelve-year-old asking 'What am I for?' is asking a worldmaking question: which version of a life is worth constructing? The rendering engine can produce descriptions of many possible lives. It cannot determine which life is worth living, because the determination requires purposes that only a living, finite, stake-holding being possesses.

Origin

The worldmaking thesis emerged gradually across Goodman's career. The seeds are visible in The Structure of Appearance (1951), where he argued that phenomenal reality is constructed from sense-data rather than given. Fact, Fiction, and Forecast (1955) extended the constructivism to scientific knowledge, demonstrating through the grue paradox that induction depends on entrenched predicates rather than on objective natural kinds. Languages of Art (1968) applied the framework to aesthetics, arguing that paintings construct versions of reality through symbolic reference. Ways of Worldmaking synthesized these insights into a general theory: all knowledge—scientific, artistic, perceptual—is worldmaking, and there are as many worlds as there are right versions.

The thesis was controversial from its first articulation. Realist philosophers objected that Goodman had dissolved objectivity into conventionalism. Goodman responded that rightness provides all the objectivity rational inquiry requires—that the demand for a version-independent reality is a metaphysical fantasy with no epistemological payoff. The debate has never been fully resolved, but Goodman's framework has proven generative across philosophy of science, aesthetics, and now—unexpectedly—the philosophy of AI, where the question of what makes a version right rather than merely plausible has become practically urgent for anyone evaluating machine-generated outputs.

Key Ideas

No ready-made world. There is no single reality independent of all versions—only multiple worlds constructed through different symbol systems, each organizing experience according to its own standards.

Worldmaking operations. Versions are constructed through composition and decomposition, weighting and ordering, deletion and supplementation, deformation—specific procedures that rearrange prior symbolic materials into new configurations.

Standards are internal. A version is right when it meets the standards of its own symbol system, fits with other accepted versions, and serves the purposes for which it was constructed—not when it corresponds to a version-free reality.

Purposes ground rightness. The evaluation of whether a rendering is right depends on the worldmaker's purposes, which are grounded in lived experience and evaluated by judgment—requirements AI rendering engines do not satisfy.