Nelson Goodman

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI describes the AI moment as an amplification event—a technology that collapses the distance between imagination and artifact. Goodman supplies the conceptual instrument that tells us what an artifact made by an amplifier actually is, and whether it qualifies as the thing it resembles. His answer is functional, not genetic: what makes something a symbol system is not where it came from but how it operates—whether it denotes, exemplifies, and expresses through structured conventions in ways that organize experience and yield understanding.

The diagnosis this makes available is precise. When a diffusion model generates an image or a language model writes a paragraph, the output may possess every surface property of a version of the world—the right vocabulary, the right conventions, the right appearance of coherence—without the scheme-content relation that gives versions their significance. The scheme-content relation requires a worldmaker: an agent who intends the symbols to refer in particular ways, selects among alternatives, and constructs a version rather than sampling from a probability distribution. The distinction between selecting and sampling, between worldmaking and rendering, is Goodman’s most bracing contribution to the present.

He is, in the cycle’s gallery, the thinker who explains why the question “Did a human make this?” is the wrong question, and why the right question—“Does this function as a symbol system that organizes experience through structured reference?”—is harder to answer and more demanding in its consequences. Those consequences fall unevenly. Work produced through genuine human-AI collaboration, where the human maintains the worldmaking project, may fully qualify. Work produced through casual prompting, where no worldmaking project exists, fails Goodman’s test regardless of its surface quality. The autographic crisis in literature he did not live to see is, at bottom, the crisis his framework predicted: the sudden relevance of productive history in a medium that had agreed to treat it as irrelevant.

Where Judea Pearl provides the instrument for measuring what machines can reason and what they cannot, Goodman provides the instrument for measuring what they can make and what they only simulate. Both instruments return the same discomfiting verdict: that fluency without the right underlying structure is not intelligence, and not art, and not a version of the world. It is a sophisticated appearance of all three, and the sophistication is precisely what makes the distinction urgent.

Origin

Born in Somerville, Massachusetts in 1906, Goodman spent his early career as an art dealer in Boston before completing his doctorate in philosophy at Harvard under C. I. Lewis—a combination of vocations that was not accidental. The dealer’s world was a world of forgeries, authentication, and the question of whether a perfect perceptual copy of a Vermeer was a Vermeer; the philosopher’s world gave him the tools to answer the question with a rigor that neither the dealer nor the aesthetician had previously achieved. The answer appeared in Languages of Art in 1968, and it divided the arts into autographic ones, in which history of production is constitutive of identity, and allographic ones, in which a notation system determines identity across instances. Painting is autographic; music is allographic; and the van Meegeren forgery case—the Dutch painter who sold fake Vermeers to museums and was exposed only when he confessed—became Goodman’s paradigm case for why origin matters in autographic arts even when perception cannot detect the difference.

The second major work, Ways of Worldmaking (1978), extended the framework from aesthetics to epistemology. Goodman argued that there is no version-free world behind our various descriptions of it—no table-in-itself that the physicist, the carpenter, the painter, and the economist are all describing from different angles. There are only the versions, each constructed by its symbol system, each answering to its own standards of rightness. This was not relativism: Goodman was emphatic that some versions are better than others, that rightness is real and demanding. It was the claim that rightness is internal to symbol systems, not a matter of correspondence to a pre-given reality, and that the evaluation of any version—scientific, artistic, everyday—must proceed through examination of how the symbols function rather than through comparison with a world that exists prior to all symbolization.

Goodman retired from Harvard in 1977 and died in 1998, one year before the first transformer architecture appeared in the literature. He did not anticipate AI, but his framework anticipates the AI question with a precision that only becomes visible now—which is the mark of philosophy that has found its moment.

Key Ideas

Symbol systems and worldmaking. The central claim of Goodman’s aesthetics is that art operates through structured systems of symbols—each possessing syntactic and semantic rules, each constructing a version of the world rather than copying one. Worldmaking proceeds through composition, decomposition, weighting, deletion, and deformation of prior versions; there is no version-free starting point. The plurality of right versions is not a license for anything-goes relativism but a demand for the most rigorous kind of evaluation: assessing whether this version, built by these means, achieves the coherence, fit, and productivity that make it worth having.

Autographic versus allographic arts. Goodman’s autographic/allographic distinction turns on whether a work’s identity is determined by its history of production (autographic—painting, sculpture) or by a notation system that can be correctly instantiated by anyone (allographic—music from a score, literature from a text). The distinction, built for a world of single human authorship, becomes unstable when AI enters the productive process—generating the autographic crisis in literature in which texts produced through human-AI collaboration cannot be evaluated by the allographic criterion alone.

Exemplification and expression. Goodman’s analysis of reference beyond denotation. Exemplification occurs when a symbol refers to a property it actually possesses by displaying it—a tailor’s swatch exemplifies the color and texture of the fabric it is cut from. Expression is metaphorical exemplification: a painting “expresses” sadness by metaphorically possessing sadness as a property and directing attention to it. Both modes of reference are cognitively substantive—they are ways of organizing experience that yield understanding unavailable through denotation alone.

Density and differentiation. The formal ground of Goodman’s claim that aesthetic symbol systems make unique cognitive demands. A dense symbol system, like painting, provides for infinitely many characters ordered so that between any two there is always a third—every minute variation in color or line is potentially meaningful. A differentiated system, like digital type, permits exact replication because the characters are separated by gaps. AI output that looks like a painting is syntactically differentiated at the pixel level—it is a grid of discrete values—while a painting is syntactically dense. The difference is not merely technical; it determines what kind of attention the work demands and what kind of understanding it can provide.

Rightness of rendering. Goodman’s replacement for truth-as-correspondence. A version is right when it achieves internal coherence, fits with other accepted versions, serves the purposes for which it was constructed, and meets the standards of the symbol system within which it operates. Rightness applies to scientific theories, artworks, and everyday descriptions alike—and to AI outputs, which must be evaluated by whether they achieve rightness rather than by whether they were generated by human hands.