AI can describe what is visible in a photograph — palette, composition, apparent medium, the presence of a signature. Attribution is a different discipline entirely. Knowing the difference protects both owners and buyers.
What AI analysis does well
Given good photographs, an AI model can suggest an object type, describe the visual character of a work, notice a visible signature, and propose search-friendly tags. It does this quickly, consistently and without fatigue. On Maison Levit, this makes first-pass cataloguing dramatically faster.
What it cannot do
- It cannot authenticate. Authenticity depends on physical examination — pigments, grounds, canvas weave, tool marks.
- It cannot attribute. Attribution rests on connoisseurship, comparative material and documentation.
- It cannot value. Market value depends on condition, provenance, demand and sale context.
- It cannot verify provenance. Documents must be examined, not inferred.
Why every result carries a confidence score
A suggestion with 0.4 confidence is a hypothesis to check, not a fact to publish. Maison Levit shows these scores openly and requires the owner to review every field before it enters the artwork's passport.
The right division of labor
Let AI do the tedious first draft. Let the owner correct it with what only the owner knows. And when a work seems to warrant it, bring in the professionals — conservators, catalogers and specialists — whose judgment no model replaces.
An AI analysis is the beginning of a conversation about an artwork, never the final word.