A Definition Is Not Its Image: Computed Pictures and the Limits of Persuasion

A Definition is Not Its Image: Computed Pictures and the Limits of Persuasion

By Andreas Ehstand

This essay examines a series of 101 computed images, each derived by a fixed drawing rule from the definition of a proposed concept about human experience with artificial intelligence. The images are not produced by an image generator from a prompt; they are rendered from the text of the definition under a rule that, once fixed, always yields the same picture for the same definition. Written by the maker of the series, the essay asks a narrow question: what is gained aesthetically when a definition becomes an image, and why does that gain not amount to validation of the definition? The question is placed in the lineage of conceptual art, where a definition has stood beside an object and an instruction has served as the generator of a work. Two works from the series serve as cases. The first renders a concept for the moment one lacks the right words for an experience with AI; the second renders a concept for the private satisfaction of carrying a meaning intact between two very different sides. The analysis distinguishes three layers that viewers tend to merge: the definition, which carries a claim; the rule of transformation, which carries none; and the interpretation of the viewer, which adds tone, value and coherence. It argues that the aesthetic gain, simultaneity and comparability, is real, but that the persuasive force of a well-formed image attaches to the image rather than to the concept. The essay proposes that series of this kind are best understood as instruments of attention: they return the viewer to the definition with sharper questions rather than supplying an answer. It closes by connecting this boundary to a wider condition of visual culture, in which automatically formatted summaries and charts borrow the authority of clarity.

Introduction

There is a particular pleasure in seeing an idea take shape. A sentence that one has turned over many times suddenly appears as a form, with edges and weight and colour, and the idea seems to have arrived. The pleasure is genuine. It is also, in a precise sense, misleading, and this essay attempts to say exactly where the pleasure ends and the misleading begins. The occasion is a public series of 101 images, each belonging to a proposed concept about how people experience their dealings with artificial intelligence.[1] Every concept has a German name, an English name and a definition in plain language. Every image is computed from that definition. The rule that performs the computation was developed with the assistance of AI tools and then fixed; from that point on, the same definition always yields the same image.[2] Nothing in the series is drawn by hand case by case, and nothing is produced by typing a description into an image generator and selecting the most attractive result. I write about this work as its maker. That position has obvious limits: I cannot report on how the images strike a viewer who meets them without context, and I have an interest in their being taken seriously. It also has one advantage. I know what the rule does and does not know about the definitions it renders, and that knowledge is the subject of this essay.
Body

A definition beside an image

The relation between a definition and a picture is not a new problem for art. Joseph Kosuth’s One and Three Chairs (1965) places a wooden folding chair, a photograph of that chair and a photographic enlargement of the dictionary definition of “chair” side by side.[3] The work does not ask the viewer to believe the definition because the photograph is convincing; it holds the three forms apart so that their differences become visible. Two years later, Sol LeWitt wrote that in conceptual art “the idea becomes a machine that makes the art.”[4] In LeWitt’s wall drawings, the instructions were carried out by other people, and the instruction, not the hand, defined the work. A series of computed definition images sits between these two positions. Like LeWitt’s instructions, a fixed rule generates the work. Unlike them, what the rule reads is the text of a definition, so that the wording of a proposed concept, not a separately formulated idea, determines the image. Like Kosuth’s chair, the series confronts a definition with an image. Unlike Kosuth, it risks the opposite effect: instead of holding the definition and the image apart, it can let the image stand in for the definition. That risk is the subject of what follows.

Two cases

The first concept is called Begriffskante in German and Edge of Language in English. In its plain-language form it names “the moment you simply lack the right words for an experience with the AI – ‘it understood me’ says too much, ‘it just matched patterns’ says too little, and there’s no word yet in between.”[5] The everyday example extends this beyond AI: “You want to describe a strong feeling and realize ‘nice’ is too weak and ‘overwhelming’ too big – and you stand there speechless, because the right word simply doesn’t exist yet.”[6]
The image shows two tilted panels facing each other across a dark ground, like the pages of an open book seen at an angle. The left panel is ruled with many horizontal lines; the right panel is almost empty, marked only by a few scattered points. In the gap between them, small jagged marks point towards a single ring at the centre (figure 1).

The second concept is Brückenstolz, or Bridge Pride: “The quiet, private joy when you’ve carried a message intact from one very different part to the next – a small feat of translation nobody sees, that still feels like a little victory.”[7] Its everyday example: “You explain something complicated about the phone to your grandma so she really gets it – nobody claps, but inside you’re quietly proud the thing came across cleanly.”[8] Here the image shows two shores. The left is rounded and irregular; the right is angular. A span crosses between them. Fine vertical lines drop from it, and small markers sit along the deck. A faint arc rises above the span, and a soft glow sits beneath its centre (figure 2).

What the transformation gains

Two kinds of gain are visible in these cases, and both are aesthetic in the strict sense: they concern what can be perceived. The first is simultaneity. A definition is read in sequence. One meets “the moment,” then “lack the right words,” then the two failed descriptions, and the relation among these parts is assembled over time. An image presents its relations at once. In Edge of Language, the viewer grasps immediately that the experience lies between two inadequate options, one too full and one too empty, before registering what either option is. The structure of the concept precedes its content. In this case the image follows a structure that the definition itself states: “too much” and “too little.” The second gain is comparability. Because the rule is fixed and applied identically across the series, the 101 images share a visual grammar. One can set Edge of Language beside Bridge Pride and see that both are organized around a gap between two sides: in the first, a gap that is not crossed; in the second, one that has been. This comparison is not available in an alphabetical list of definitions, where the two concepts sit far apart. The series becomes a map, and maps allow kinds of seeing that inventories do not.
Neither gain is trivial. Concepts about subtle experience are hard to hold in mind; giving them a perceivable structure is a way of making them available for thought.

What the transformation chooses and adds

Every translation also selects. The case of Bridge Pride shows this precisely. Its plain-language line speaks of carrying a message “from one very different part to the next” without naming the parts. Its everyday example involves two people, a grandchild and a grandmother. The more exact formulation on the same page names a different pair as an illustration: a meaning passed “between two very different sides – say, between a human and an AI.”[9] The wording therefore holds at least two readings open. The image, by contrasting a rounded shore with an angular one, leans towards one of them, the meeting of an organic and a technical side. That reading is supported by the text; it is not the only one the text supports. The image does not misrepresent the definition, but it decides something that the definition leaves undecided.
The image also adds tone. The soft glow beneath the span reads as quiet satisfaction, which is faithful to the definition’s “quiet, private joy.” Yet the dark ground and golden lines that run through the series lend every concept a certain preciousness. A concept for an unpleasant experience, rendered in the same palette, acquires some of the same dignity. The rule does not know whether a concept describes something welcome or something troubling. It applies its grammar regardless, and the grammar has a mood.
Finally, the image adds coherence. A well-composed form looks resolved. Edge of Language is, as a concept, about irresolution: about standing speechless at the boundary of one’s vocabulary. As an image it is balanced in placement, even though its two panels remain unequal, one ruled and one nearly empty. This is not a fault of the image; it is the nature of composition. But it means that the viewer receives, alongside the concept of being stuck, an experience of something settled.

Three layers

These observations point to a distinction that is easy to state and hard to keep in view while looking. Any computed image of this kind involves three layers.
The first is the definition. It is a proposal: a claim that there is an experience worth naming and that this wording delimits it. Such a claim can be examined, refined or rejected. It is the only layer that asserts anything. The second is the rule of transformation. It is a procedure that turns text into form. In a different part of the same project, a public workshop uses a separate, simpler rule in which, as the workshop puts it, “each word becomes a golden vein, each sentence a ring, each initial a point of light.”[10] Whatever its details, a rule of this kind operates on the text as text. It carries no claim about whether the concept is sound. The third is the interpretation of the viewer: everything the viewer brings, from the association of an angular shore with technology to the sense that a balanced image describes a resolved state. This layer is free, and that freedom is part of the value of looking. The confusion this essay is concerned with arises when the qualities of the second and third layers are credited to the first. The image is coherent, so the concept seems coherent. The image is beautiful, so the concept seems true. The image belongs to a consistent series, so the concept seems to belong to an established system.

Why persuasion is not validation

A definition is validated, if at all, by showing that the experience it names occurs, that people can recognize it, that it can be distinguished from neighbouring experiences and that naming it is useful. None of these is a visual property. An image can be harmonious whether the concept it renders is precise or vague, common or rare, well bounded or overlapping with a dozen others. The rule would render a poorly formed definition as confidently as a well-formed one. The persuasive force of the image is therefore real but misattributed. It belongs to the image’s success as an image. When it migrates to the concept, it does so without passing through any of the tests that would justify it. The risk is not that viewers are deceived by a false claim, but that a proposal comes to feel settled before anyone has argued for it. This matters for a series of proposed concepts in a way it might not for abstract painting. The concepts are offered as candidates for a vocabulary, and vocabularies shape what people notice. If an image makes a concept feel inevitable, it may speed its adoption ahead of its examination. Kosuth’s arrangement suggests one way of keeping the layers distinguishable: the definition shown beside its image, the rule named as a rule, and the viewer invited to ask from which layer a perceived quality comes.

Conclusion

What, then, does the series offer, if not evidence? I would describe it as an instrument of attention. After looking at Edge of Language, one returns to its definition and notices that it offers no third description: the experience lies in standing between two familiar ones, “it understood me” and “it just matched patterns,” both of which ring false. After looking at Bridge Pride, one notices “intact”: the joy is not in having explained, but in having lost nothing on the way. The images did not establish these readings. They sent the viewer back to the sentence with a question the sentence alone had not provoked. This boundary extends beyond a single series. Contemporary visual culture is full of automatically formatted clarity: generated summaries, one-click charts, dashboards that turn uncertainty into clean lines. Each borrows the authority of form, and each can make a claim feel established because it looks resolved. Art can openly declare its transformation, and a gallery of computed definitions can therefore serve as an exercise in a discipline that is increasingly needed elsewhere: to take full pleasure in a form while knowing precisely what it does not guarantee. A definition is a claim. An image is an encounter. The value of bringing them together lies in keeping them apart.

Author Biography

Andreas Ehstand is an independent AI researcher based in Starnberg, Germany. He studies how people interpret and check the output of AI systems, and how defined terms can be kept stable when they pass between people, organisations and machines. He is the author of an individual Internet-Draft submitted to the Internet Engineering Task Force on human oversight acts. As the maker of a public series of 101 computed definition images, he explores the relation between vocabulary and visual form. Before turning to AI research, he worked for 25 years as a coach in competitive sport, including in the German Tennis Bundesliga. Email: andreas@augmanitai.com Website: augmanitai.com/autor · ORCID: 0009-0006-3773-7796

Notes
[1] “The Work Gallery: 101 Definition-Based Artworks,” AUGMANITAI, accessed 16 October 2026, https://augmanitai.com/en/werk-galerie.
[2] The gallery states that each work is “computed from its definition, not made with an image generator” and that “the same thought always yields the same picture.” “The Work Gallery.”
[3] Joseph Kosuth, One and Three Chairs, 1965, wood folding chair, mounted photograph of a chair and mounted photographic enlargement of the dictionary definition of “chair,” Museum of Modern Art, New York, accessed 16 October 2026, https://www.moma.org/collection/works/81435.
[4] Sol LeWitt, “Paragraphs on Conceptual Art,” Artforum 5, no. 10 (June 1967): 79–83.
[5] “Edge of Language,” AUGMANITAI Lexicon, accessed 16 October 2026, https://augmanitai.com/en/lexicon/terms/edge-of-language.
[6] “Edge of Language.”
[7] “Bridge Pride,” AUGMANITAI Lexicon, accessed 16 October 2026, https://augmanitai.com/en/lexicon/terms/bridge-pride.
[8] “Bridge Pride.”
[9] “Bridge Pride.”
[10] “Turn a Defined Thought into an Image,” AUGMANITAI, accessed 16 October 2026, https://augmanitai.com/en/denk-bild.

Bibliography

AUGMANITAI. “Bridge Pride.” AUGMANITAI Lexicon. Accessed 16 October 2026. https://augmanitai.com/en/lexicon/terms/bridge-pride.
AUGMANITAI. “Edge of Language.” AUGMANITAI Lexicon. Accessed 16 October 2026. https://augmanitai.com/en/lexicon/terms/edge-of-language.
AUGMANITAI. “The Work Gallery: 101 Definition-Based Artworks.” Accessed 16 October 2026. https://augmanitai.com/en/werk-galerie.
AUGMANITAI. “Turn a Defined Thought into an Image.” Accessed 16 October 2026. https://augmanitai.com/en/denk-bild.
Kosuth, Joseph. One and Three Chairs. 1965. Museum of Modern Art, New York. https://www.moma.org/collection/works/81435.
LeWitt, Sol. “Paragraphs on Conceptual Art.” Artforum 5, no. 10 (June 1967): 79–83.

_____

Publishing in Art Style Magazine is free of charge for anyone. There are no article processing charges or other publication fees. Art Style Magazine is independent and supports the Open Access Movement. The editors of Art Style Magazine cannot be held responsible for errors or any consequences arising from the use of information contained in essays and articles published on the Art Style Magazine’s website and editions. Authors agree to the terms and conditions and assure that their submissions are free of third parties’ rights. The views and opinions expressed in the essays and articles are those of the author and do not reflect the views of Art Style Magazine. The authors of Art Style Magazine’s essays and articles are responsible for its content. The Art Style Magazine‘s website provides links to third-party websites. However, the magazine is not responsible for the contents of those linked sites, nor for any link contained in the linked site content of external Internet sites (see Terms & Conditions).

Art Style Magazine’s Key Media & Audience Highlights

Art Style Magazine appreciates your support.

For further information on how to offer support, kindly contact Art Style Magazine or donate at your convenience.