Ways of Worldmaking

In The You On AI Field Guide

The book emerged from Goodman's 1976 John Locke Lectures at Oxford—six lectures that compressed his mature philosophy into a form accessible to a general philosophical audience. The lectures were controversial from their delivery. The Oxford audience, steeped in realism and common-sense philosophy, resisted the claim that there are many worlds. Goodman's response was characteristically unyielding: if you insist there is one world, specify which version of it is the privileged one. The physicist's? The perceptual world of manifest experience? The world as described by a particular language at a particular time? Each answer generates problems, and the problems dissolve when one stops demanding a version-independent world and accepts that rightness is the only standard rational inquiry requires.

The framework's application to AI begins with the observation that large language models have absorbed an unprecedented range of human-constructed versions—scientific, historical, mythological, aesthetic, ordinary perceptual experience encoded in natural language. The model can generate outputs in any of these versions, deploying the conventions of physics or poetry or perception with facility that depends on the density of relevant training data. What the model cannot do is choose which version is the right one to construct for a given purpose, because the choice requires purposes—reasons, grounded in lived experience and evaluated by judgment, for constructing this world rather than that one. The model samples from the versions humanity has constructed. The human worldmaker chooses among the samples, and the choosing is the worldmaking that Goodman's framework identifies as the irreducibly human contribution.

The book's closing pages acknowledge that worldmaking is constrained—not by a single reality that all right versions must match, but by the requirement that new versions fit with other accepted versions, serve recognizable purposes, and meet the standards of their symbol systems. The constraints are real and demanding. But they are also plural, often conflicting, and resolved through judgment rather than calculation. A new scientific theory that contradicts accepted versions can still be right if it is more productive, more coherent, better supported by evidence. The judgment of whether the new version's productivity outweighs the cost of abandoning old versions is not algorithmic. It requires worldmakers—scientists, artists, philosophers—who stake their cognitive authority on the claim that this version is the one worth constructing. AI can assist the construction. It cannot make the stake, because the stake requires inhabiting the world the version constructs, and inhabiting requires being the kind of entity for whom versions matter because life is finite and choices are consequential.

Origin

Ways of Worldmaking was published by Hackett in 1978, two years after the Oxford lectures. Goodman was seventy-two, with five major books and dozens of articles behind him, and the book reads as a summation—the worldmaking framework that had been implicit in his earlier work made explicit, radicalized, and defended against the most powerful objections. The book is short (142 pages), dense, and unapologetic about its philosophical commitments. It generated immediate controversy and has sustained attention for nearly five decades, becoming one of the most cited works in contemporary philosophy of science and aesthetics.

Key Ideas

Many worlds, not one. There is no version-free reality—only multiple worlds constructed by different symbol systems, each organizing experience according to its own conventions and standards.

Worldmaking operations are specific. Versions are constructed through composition, decomposition, weighting, ordering, deletion, supplementation, deformation—rearranging prior materials into new configurations.

Rightness is plural. Multiple incompatible versions can all be right—rightness is internal to worldmaking projects, not grounded in correspondence to a single privileged reality.

Choice requires worldmakers. Selecting which version to construct requires purposes grounded in lived experience—a requirement AI rendering engines do not satisfy, making them dependent on human worldmakers for direction.