The rapid development of artificial intelligence is beginning to transform one of the art world's most complex and contested fields: artwork authentication. While AI is unlikely to replace the judgement of experienced specialists, it is emerging as a powerful new tool that can reveal visual patterns beyond the limits of human perception and strengthen traditional methods of expertise.
Text by Angie Afifi
Can artificial intelligence determine the authenticity of a painting more accurately than an experienced art historian? Only a few years ago such a question would have seemed like science fiction. Today, however, AI is becoming an increasingly important component of the authentication process, providing specialists with new ways of analysing works of art.
Traditional methods of authentication face a number of inherent limitations.
When relying on visual analysis alone, even the most experienced art historians inevitably draw upon professional experience, memory, intuition and comparisons with previously attributed works. Although experts examine a wide range of characteristics—including composition, style, technique, brushwork and colour palette—such assessments inevitably contain an element of subjectivity, particularly in the absence of universally accepted standards for art historical expertise.
Even the application of numerous analytical criteria cannot guarantee complete objectivity. Human perception is inevitably influenced by experience, expectations, institutional pressures and prevailing scholarly opinion.
Scientific laboratory methods also have their limitations. Modern forgeries are increasingly produced using materials identical to historical ones, meaning that laboratory analysis cannot always distinguish an original from a highly sophisticated imitation. Moreover, such examinations are often expensive, may require invasive procedures, are not always readily available, and cannot be applied to every artistic medium.
In recent years, art historians have gained access to an important new tool for visual analysis: machine learning. Rather than replacing traditional expertise, AI is designed to complement it by providing an additional source of evidence and identifying visual patterns that remain invisible to the human eye. In this way, machine learning becomes another element within the broader evidential framework supporting expert decision-making.
These technologies are now being developed both in universities and research laboratories and by specialised commercial companies. Some focus primarily on AI-based analysis of works of art, while others incorporate artificial intelligence into a broader investigative framework that combines art historical expertise, scientific examination and provenance research. Today, such technologies are already being employed by museums, auction houses, collectors and private owners of works of art.
AI is believed to analyse a vast number of micro-level visual characteristics that cannot realistically be assessed comprehensively by humans. The precise nature of these characteristics, however, remains largely unknown because neural networks operate as models with limited interpretability.
One of AI's greatest strengths is that it is largely unaffected by the cognitive biases that inevitably influence human judgement. The algorithm has no knowledge of a painting's market value, the identity of its owner, the opinions of prominent experts or its museum status. Consequently, its assessment is not shaped by authority, market expectations or reputational considerations. Nor does it possess intuitive preferences. Instead, it reaches its conclusions solely through the analysis of visual patterns.
To train these systems, researchers typically assemble datasets containing works with securely established authorship alongside stylistically similar paintings by other artists. This enables the algorithm to identify the distinctive characteristics of an individual artist's manner rather than simply recognising a particular artistic movement or period. Instead of analysing an image as a complete composition, the model usually examines thousands of small fragments from which it independently identifies statistically significant patterns. The success of these systems, however, depends heavily on close collaboration between art historians and machine learning specialists. Art historians contribute to the creation of high-quality datasets and play a crucial role in interpreting the algorithm's conclusions.
Like any analytical method, AI also has its limitations. Its effectiveness depends heavily on the quality and breadth of the training dataset, while results may be influenced by restoration, the quality of digital images or atypical works within an artist's oeuvre. Nevertheless, artificial intelligence is already regarded as one of the most promising technologies for significantly strengthening traditional approaches to artwork authentication.
Importantly, AI can do far more than estimate the probability that a work is authentic. It can identify areas that are most characteristic of an artist's hand, distinguish passages that may have been executed by assistants, detect stylistically inconsistent sections and even indicate varying levels of confidence across different parts of a painting.
In this way, AI becomes not merely a system that produces a simple "yes" or "no" answer, but a sophisticated tool for analysing the internal structure and execution of an artwork.
Ultimately, artificial intelligence should be viewed as a means of supporting expert authentication rather than replacing it. The most reliable assessments will continue to emerge from the integration of multiple forms of evidence—including art historical expertise, scientific examination, provenance research and AI-assisted visual analysis. It is precisely this comprehensive approach that offers the greatest potential for improving the objectivity and reliability of artwork authentication.
