ArtGraph, the book
a summary.

Abstract painting ↔ language ↔ person.

The book describes a language built to sit between those two ends. It is not a language about abstraction, which is criticism and already exists. It is a language between two parties, doing what a language does, which is to let one side reach the other.

This book is built on a working system, not a proposal for one. The catalog is scored, the notation is specified, the concierge runs. Some parts work better than others. Some are still being sorted out, and where that is true it is said so on the page rather than left for a reader to discover.

Each part of the system is one level of abstraction above the part beneath it. The painting is abstract. The reading is an abstraction of the painting. The mark is an abstraction of the reading. The shape is an abstraction of the mark. Nothing in the system points at a thing in the world, and that is what makes it possible for it to point at a painting.

The argument. Every available way of sorting abstraction describes the object and calls it finished: medium, period, movement, dimensions, color, price. None of them records what happened in the room.

The taxonomy separates two things that had been collapsed. What the painting is and what the encounter is. Both are recorded. Neither is treated as the other.

The language. Nine in total, in two families.

Six read the painting. Texture, Form, Color Range, Palette, Contrast, Abstraction Quotient. Five of them are measured against matter — against what is physically on the canvas, whether or not anyone is standing there. The sixth, the Abstraction Quotient, is measured against language: how far the work has travelled from anything nameable. It runs from −5 to +5 and its true centre is zero.

Three read the encounter. Space, Transformation, Feeling. Each has two ends, because an encounter has two sides.

Contrast is computed from the measured pigments. Palette is categorical rather than scalar — a list, not a position. The remaining seven are scored by hand.

The notation. One mark carrying a whole reading, identical at any size, on any surface.

Five of the six dimensions position the corners of a pentagon. The sixth, Palette, fills the shape — one opaque band per pigment, at the share it occupies on the canvas. Nothing is written on the mark that a reader has to translate back into numbers.

This is the second abstraction. A logograph like 木 points at a tree, and the tree was in the world before the mark was. The ArtGram points at a painting that is itself a likeness of nothing. The mark has no ground beneath it of the kind words normally stand on, and that is the condition it was designed for: describe an abstract painting well enough and you have made it representational again, of your description.

Across 246 paintings the marks produce 195 distinct silhouettes, 165 of them belonging to a single work, out of the 3,125 arrangements the geometry allows.

Shapes

Marks fall into a limited set of recurring forms. Eight of them, named after they recurred rather than before: Stone, Field, Blade, Keel, Fan, Shield, Spire, Rudder.

A shape is a meta mark. It names not the painting but the form its reading takes. The set holds on work from outside the catalog, scored under other assumptions. What a shape does to the person standing in front of the painting is not yet known.

What the language is for. A person says what they are reaching for instead of scanning a wall of thumbnails.

It begins with space, because space is the shallowest of the three encounter dimensions and the only one people can describe. Nobody can say what they want to feel. Everyone can say their living room faces east. From there it offers, reads the response to the object rather than the response to a question, and makes the next offer a hypothesis about why the last one almost worked.

The public instance runs on the catalog as Find Your Painting.

What leaves with the person. At the end of a session they are given a letter, the works that stopped them, and each one's ArtGram. No price, and nothing asked of them.

It is the only object in the system that travels. A private encounter becomes something that can be kept, sent to somebody who was not standing there, or taken to a gallery.

The infrastructure. A catalog goes in; one that can be read comes back.

The engine is the transformation, separate from any product built on it. The Concierge is one such product. Find Your Painting is one instance of that product, running on one catalog.

What it reads today is a catalog scored by hand, one work at a time. What it is being built to read is a catalog it has never seen, and to keep reading it as people move through it.

Scale, and the language models

Both limits are addressable now in a way they were not when the framework was begun.

Four of the six painting dimensions are physical and measurable from an image; Contrast is already computed rather than authored. Form is a judgment. The Abstraction Quotient is the interesting case: it is the only dimension measured against language rather than against matter, and estimating how far a work has travelled from anything nameable is the operation a language model performs natively. The dimension hardest for a person to score consistently may be the easiest to hand over.

That changes what the catalog can be. A framework scored by one person tops out at what one person can see. A framework a model can score reaches ten thousand works, then a hundred thousand, at which point the question stops being filtering and becomes matching — and matching requires a model of the person, built from what they respond to rather than from what they say they like.

It also changes what can be learned from an encounter. What somebody writes about a painting that moved them is unstructured and there will be a great deal of it. Turning that into something a system can use is where the language stops being one person's vocabulary and starts acquiring words from the people using it.

The audit is part of the design rather than a check bolted on afterward. Sampling, disagreement detection, and publication wherever machine and hand diverge — so that a reading made at scale can be examined the same way a reading made by hand can.

With three questions it does not answer.

Is there a relationship, at scale, between the abstractions and what people want? The system now holds several hundred paintings, each with a reading, each reading resolved into a mark, each mark falling into a shape. Whether any of that corresponds to desire — whether particular people are drawn to particular shapes, whether a reading predicts a response — is unknown. The data to find out does not yet exist in enough quantity.

Can the second abstraction help abstract painting be shared? A painting sent to someone arrives on a phone slightly larger than a postage stamp. The mark does not degrade at that size. Whether a mark travelling beside an image actually carries anything a person can use is a claim the book makes and cannot yet demonstrate.

Will any of this help the person standing in front of the painting? That is the whole test. Not whether the notation is consistent, not whether the catalog is complete, but whether someone who has spent their life saying I just liked it can stand in a room and say I got it — and mean something by it that they could not have meant before.

None of the three is settled. The instrument exists. What it is worth depends on answers that will take years and a great many more encounters than one person can supply.

Ritu Raj is a contemporary abstract painter based in Phoenix, Arizona. The manuscript of ArtGraph — A Language for Abstract Art, and a Way to Find Yours is in production.