How AI is used to evaluate the authenticity of paintings, as conservators express concerns over whether the tech can account for wear, damage, and other factors
Context & Ripple Effects
AI’s role in art was already expanding beyond image generation: artists were adopting tools such as DALL-E and Midjourney while weighing their effect on livelihoods in the early use of generative tools by artists. Applying AI to authentication moves the technology into a higher-stakes judgment where an artwork’s physical condition can complicate a model’s assessment.
The story matters because conservators’ concerns make clear that technical pattern recognition does not automatically substitute for condition-specific expertise. Its value depends on how well AI results can be interpreted alongside evidence of wear, damage, and restoration.
First-order effects
- Authentication professionals and conservators must assess AI outputs against the physical state of each painting rather than treating a model’s conclusion as self-sufficient.
- AI vendors and adopters face an immediate credibility test: systems used in authentication need to demonstrate how condition-related variables affect their assessments.
Second-order effects
- Buyers, sellers, insurers, and institutions that rely on authenticity judgments may demand clearer human review and documentation around AI-assisted evaluations.
- Conservators’ expertise becomes more central to the workflow, creating pressure for tools that incorporate condition records and restoration history instead of relying only on visual comparison.
Third-order effects
- If AI becomes a regular input to art authentication, the market may shift toward governed human-machine review rather than fully automated provenance decisions; acceptance will hinge on accountability when assessments are disputed.
- The case is an early example of a broader divide between AI’s ability to detect patterns and the domain knowledge needed to interpret imperfect real-world objects.
The trend: AI is moving from creative experimentation into expert decision workflows, where adoption depends on demonstrable limits, human oversight, and trust in the resulting judgment.