How a Reading Works

Artmetria uses AI to help users move from impression to structure.

A user may begin with a visual preference, an artwork, an artist, a market signal or a question. Where a model genuinely helps, it organises this material into clearer language, possible directions, contextual cues and next steps.

Today a language or vision model is asked to do a few precise things: the reading of a work in the Reading Room, from a photograph or from a work's page in Eye; the answer to a question you ask about a work; recognising a work or reading its label from a photograph; the reading of the day; the editorial context of the Artist of the Day, marked as written with AI assistance; and the reading of an object against ten explicit criteria, computed inside the Artmetria Index and not published as a figure. A third use is quieter: images and texts are represented by models so that neighbours can be found — in Eye's search, beside a lot in Market. That representation is a relation, never an assertion, and it enters no value and no indication. Elsewhere the intelligence is in the material and its structure — the corpus, the records with their sources, the evidence register — and no model is involved.

AI is one component of a larger layer, Artmetria Intelligence, not the layer itself. It is designed to assist interpretation, not to produce definitive truth.

Depending on the feature, Artmetria’s AI layer may help with:

• describing visual patterns in a work;
• suggesting artistic filiations or affinities;
• structuring an artwork reading;
• setting a work beside its visual and textual neighbours;
• making complex information easier to understand;
• connecting educational content with artists, segments or questions;
• helping users ask better questions before going further.

These outputs are meant to be starting points for reflection, comparison and research.

Artmetria’s AI does not provide:

• authentication opinions;
• certificates of authenticity;
• condition reports;
• legal advice;
• tax advice;
• financial advice;
• investment recommendations;
• buy, sell or hold instructions;
• guarantees of value;
• definitive expert opinions.

Where a decision has legal, financial, conservation or acquisition consequences, users should consult qualified professionals.

The Aesthetic Profile is built from your own choices. A short series of visual contrasts, and later what you look at, follow, save and set aside, are turned into a sense of your looking by explicit, inspectable rules. No model reads you, and no model decides what you like.

That sense describes tendencies in visual preference: light, composition, rhythm, materiality, figuration, abstraction, historical affinity, emotional tone or other perceptual signals.

An aesthetic profile is not a psychological diagnosis, a fixed identity or a closed category. It is a temporary map of how someone looks, refined through interaction and context, and yours to correct.

The purpose is to open discovery, not to lock users into a taste label.

Eye's recommendations suggest artists, works, or directions that appear related to the sense Eye has formed of your looking. They are drawn from your own choices and the corpus by explicit rules — not by a model, and never from market data, which Eye keeps out of what it suggests.

These recommendations are exploratory.

They should be understood as paths to investigate, not instructions to follow. A recommendation does not mean that an artwork is suitable for acquisition, undervalued, safe, liquid or likely to appreciate.

Artmetria favours discovery and structured curiosity over automated decision-making.

When AI is used to help read an artwork, it may organise visual, historical, stylistic or contextual observations.

Such analysis may help users notice details, formulate hypotheses, compare influences or understand how an artwork may be situated.

It should not be treated as attribution, authentication, valuation, condition assessment or expert certification.

Artwork analysis should remain connected to direct viewing, documentation, provenance research, specialist review and, where needed, conservation expertise.

It reads paintings and drawings, and sculpture in a first, provisional coverage.

Artmetria may use AI to explain market context, but market context is not certainty.

Auction records, visibility, liquidity, institutional attention, geography and category trends can help structure a reading, but they do not determine future value.

Where data is incomplete, uneven or difficult to compare, Artmetria aims to make uncertainty visible rather than hide it.

AI outputs should be read together with source quality, coverage level, confidence indicators and methodological notes when available.

Artmetria trains no model of its own, and does not present itself as doing so. Where a task calls for a language or vision model, Artmetria uses external models as components inside its own pipelines, under its own constraints: a fixed vocabulary of what a reading may say and may never assert, checks on the output before anything is shown, and a rule-based fallback when a model is unavailable.

Model use can be paused, tool by tool or altogether, and the pause fails closed: if the system cannot tell whether it may call a model, it does not. A surface will say that a model-assisted tool is paused rather than produce a weaker answer in its place.

What is disclosed is what the system relies on — the corpus, the record, the reading framework, the stated coverage. Which model, from which provider, with which weights and thresholds, is not disclosed: naming it would not make a reading more inspectable, and the reading does not depend on any one component staying the same.

Similarity found by a model is a relation between images or texts. It is never an attribution, and model output is never treated as historical fact.

Art is not reducible to data, and expertise is not reducible to automation.

AI can help structure information, but it cannot fully replace looking, experience, specialist knowledge, provenance research, material examination, market relationships or human judgement.

Artmetria’s position is simple: AI should help users see more clearly, not decide for them.

Artmetria’s AI features are intended for educational, analytical and research-oriented use.

Users should treat AI outputs as one layer among others. They should verify important information, compare sources, seek expert advice when appropriate and avoid relying on a single automated output for consequential decisions.

The AI layer helps ask better questions. It does not remove the need to ask them.