Here, it is not the fad of a contemporary transhumanist artist. The work "Portrait of Edmond de Belamy," which will sell for the modest sum of $432,500 on that autumn evening, was created by the Generative Adversarial Network (GAN), an artificial intelligence algorithm developed by the French art collective Obvious, trained on a dataset consisting of thousands of historical portraits. The signature, to understand it, is a mathematical formula that represents and underscores the non-human origin of the work, opening up intriguing-for some, perhaps, even disturbing-questions about the conceptual path we are invited to take in attempting to "attribute" the painting correctly.

Almost six years have passed since that auction. And the world, in the meantime, has changed. The exponential spread of generative algorithms (Chat Gpt, Gemini) and text-to-image (TTI) technologies such as DALL-E or Midjourney foment collective debate on a daily basis. AIs are deeply revolutionizing business and industrial innovation processes, sketching a dystopian-tinged future that, in pop imagery, often evokes Matrix-like scenarios.

Circumscribing the discussion to art, it must be said that the uproar generated by the sale of "Portrait of Edmond de Belamy" did not turn into the apocalypse of the art market that some, at the time, readily prophesied. The 2018 transaction had been masterfully packaged, precisely, to generate scandal. A short time later, Sotheby's tried to follow Christie's example by auctioning Mario Klingemann's "Memories of Passersby." The work, still generative, however, went for only 40,000 pounds.

The market value of AI-generated art, on balance, has gradually diminished: as the novelty effect has worn off, generative algorithms have by no means replaced - exquisitely - "human" artistic production. On the contrary, they have become an innovative tool for exploring the creative possibilities of technology, inserting themselves with increasing speed and rapid spread into the world of graphic design, video editing, and commercial photo production. Viewing AI with concern today are primarily professionals who work with images for purely applied purposes, who feel (rightly, in my view) threatened by the generative power of a technology that is evolving and refining exponentially. Artists, for the time being, seem to view AI with curiosity, bordering on suspicion, aware that the aura of sacredness and uniqueness that historically surrounds their position is hardly affected by TTI technologies. Nevertheless, the impact of AI in the contemporary landscape has the merit of igniting an extremely interesting debate, one that forces us to think deeply about the boundaries of human creativity.

Let's go back to "Portrait of Edmond de Belamy." How does the technology that created the "painting" work? Simplifying brutally, we can say that artificial intelligence, through the use of neural networks, has developed the ability to analyze and replicate complex patterns found in works of art. Algorithms are trained on thousands of images, learning the stylistic characteristics of various artists and historical periods to generate new works.

Neural networks are computational models inspired by the structure of the human brain: an artificial neural network is composed of layers of nodes (neurons, to draw an anthropomorphic parallel) connected together. Each node processes an input and transmits it to subsequent layers, allowing the network to "learn" to recognize and reproduce complex patterns. GANs, in particular, are composed of two neural networks working together: a generative network, which creates new images, is associated with a discriminative network, which evaluates how realistic these images are. This process of "competitive learning" leads to the creation of increasingly sophisticated works.

To understand how AI can create these works, however, it is essential to explore how the human mind works. Our brains use labels and categorization to process information quickly. This cognitive schema allows us to navigate the complex world around us, but it also exposes us to extremely limiting biases and categorizations. In 1951, a very famous experiment conducted by Solomon Asch showed that people tend to conform to the opinions of others, even when those opinions are demonstrably wrong. The phenomenon is known as "social conformity." Similarly, AI algorithms, which are trained on large data sets, tend to reproduce patterns that reflect the training data. If these data contain biases, as is often the case, creations from AI can reflect those biases. The MIT Media Lab, to give a concrete example, has shown that facial recognition systems are more accurate for light-skinned people than dark-skinned people, reflecting the biases inherent in training data.

Now, art is universally recognized as one of the purest forms of human expression, a territory where creativity reigns supreme, where the "labels" we use to categorize reality are fatally challenged. But can such absolute creativity really be replicated by an algorithm? Those who argue that AI cannot be truly creative take refuge in a simplistic assumption: the algorithm, they say, merely recombines existing elements in statistically efficient ways. But isn't that exactly what artists have always done? Beyond the computational aspect, humans also rework ideas, styles and influences by fashioning "chimeras" (to quote Eco) that could not exist without the pre-existing creative fabric.