Who is the author?¶

The advent of generative artificial intelligences has triggered a profound transformation in the creative landscape, bringing with it a series of fundamental questions. Among these, the question of who can be considered the author of a work generated using AI-based tools emerges as crucial. This question is not only theoretical but has significant practical implications in the artistic, legal, and economic fields. Traditionally, the concept of author has been closely linked to the idea of human creation. The author was the one who through their own ingenuity and sensitivity gave life to unique and original works. This has worked well for centuries, adapting to the evolution of artistic techniques and means of expression. However, the introduction of AI into the creative process has called this consolidated view into question.

Currently, the debate on authorship in the context of AI focuses mainly on legal issues, particularly on copyright protection and ownership of author's rights.

The predominant approach tends to seek a clear separation between the human contribution and that of the machine. This view, however, could prove limiting and inadequate for fully understanding the new creative dynamic that is developing.

The attempt to clearly separate the human contribution from that of the AI risks not grasping the profoundly interconnected nature of the creative process that takes place with these new tools.

AI systems are not simple passive tools in the artist's hands, but active partners in the creative process, capable of introducing elements of novelty and unpredictability into the final result. It therefore becomes necessary to rethink the very concept of authorship, considering the emergence of a new figure that I would call the hybrid author. This new type of creator is neither completely human nor completely artificial, but a unique synthesis of both.

The hybrid author exploits the potential of AI not as a simple tool, but as a collaborator capable of expanding the boundaries of human creativity.

This evolution of the concept of author raises some undoubtedly complex questions. How can creative contribution be defined in a work generated with the assistance of AI? What are the criteria for attributing authorship in this new context? How are the rights of the human author balanced with the role of artificial intelligence in the creative process?

In the following paragraphs, we will explore this new conception of hybrid author in more detail, analyzing how it manifests itself in contemporary artistic practice and what challenges it poses to the current legal and cultural system.

The hybrid author: a new synergy between human and machine¶

The concept of hybrid author represents a new frontier in understanding creativity in the digital age. This figure arises from a unique synergy between human ingenuity and the computational capabilities of artificial intelligences, creating a creative process that goes beyond the simple sum of the parts.

The hybrid author manifests as a dynamic collaboration in which both the human and artificial elements play crucial and complementary roles.

Artificial intelligences are distinguished by three fundamental characteristics in their approach to creativity. First, they have extreme computing power that allows them to process and combine enormous amounts of data in very short times. In addition, this capacity integrates with their ability to generate and combine ideas in unexpected ways, drawing from a vast knowledge of information and styles. Finally, they develop a peculiar form of "computational creativity," which allows them to produce variations and iterations on a theme with a speed and in a quantity that exceed human possibilities.

In the creative process involving artificial intelligence, the role of the human being remains central and manifests through multiple aspects. They are in fact responsible for the conception of the initial idea and the overall creative vision of the project, elements that constitute the foundation of every work. During development, they assume the role of direction, controlling the process and guiding the AI toward the realization of their idea. Their critical function emerges in the selection and interpretation of results produced by the AI, through which they confer meaning and context to the generated works. The creative enriches the process with the distinctive elements of human experience, such as emotionality, intuition, and understanding of cultural context. Furthermore, they develop the ability to identify potential value even in apparently incoherent or unexpected outputs, transforming possible anomalies into creative opportunities.

In this new reality, the human author is no longer the direct executor of the work but assumes the role of director and curator of the creative process. The user sets the parameters, provides the initial inputs, and then guides the AI through a series of iterations and refinements. The AI is not a simple passive tool, but an active collaborator capable of introducing elements of surprise and innovation into the creative process. Continuing with the theatrical metaphor, if the human creative is the director, the artificial intelligence behaves like a talented but egocentric actor, ready to dominate the scene in the absence of rigorous direction.

This collaboration manifests in different ways depending on the type of AI used and the degree of control exercised by the person. In some cases, the AI might be used as a tool for generating ideas or initial drafts, which are then elaborated and refined manually. In other cases, the final work might be the direct result of the AI's output, with the creative acting primarily as curator and selector.

The hybrid nature of this new type of author raises interesting philosophical questions about the nature of creativity itself. Can a work generated with an AI be considered truly creative? How is creative contribution evaluated when part of this process takes place inside a computational "black box"? These questions have no simple answers but reflect the complexity and richness of this new creative paradigm.

The hybrid author, therefore, does not stop at using new technologies but pushes to explore still unexplored expressive territories. This figure challenges traditional conceptions of creativity, authorship, and originality, opening new expressive possibilities while at the same time posing unprecedented challenges to the world of art and law.

In this context, the hybrid author represents precisely the push toward the discovery of new artistic frontiers. The collaboration between human and machine is not a new concept; consider the comparison between the marathon and Formula 1. The marathon is a purely human sport, while the car race represents a perfect fusion between human skill and advanced technology. Despite this difference, today no one would question that a Formula 1 driver is a true athlete.

Case study: "Unsupervised" by Refik Anadol¶

To better understand how the concept of hybrid author manifests in contemporary artistic practice, let's analyze a work by Refik Anadol, an artist of Turkish origin currently residing in the United States, a leading figure in the panorama of contemporary digital art. His installations, which fuse art and technology in innovative ways, have been exhibited in numerous cities around the world, including Milan.

Unsupervised is an installation visible at the MoMA in New York; the work presents itself with an imposing screen showing a sequence of abstract images in continuous evolution. These images are generated by an artificial intelligence trained on a vast dataset of works selected from the museum's collections, comprising both images and videos. Unsupervised is part of a larger project, Machine Hallucination, which on Anadol's site is described as follows:

Unsupervised is part of Machine Hallucinations, Refik Anadol studio's ongoing project exploring the aesthetics of data based on collective visual memory. Since the beginning of the project in 2016, Anadol has used artificial intelligence as a collaborator of human consciousness, in particular DCGAN, PGAN and StyleGAN algorithms trained on vast data sets to unveil unrecognized layers of our external reality. Anadol and his team collect data from digital archives and from publicly available resources and process them with machine learning classification models. As a masterfully curated multi-channel experience, Machine Hallucinations brings to the public a self-regenerating element of surprise and offers a new form of sensory autonomy through cybernetic serendipity.

Returning to the work exhibited at MoMA, the creative process behind it is complex and multi-phase. Anadol personally directed every aspect of this process.

At first glance, one might be tempted to consider the AI as the primary author of the work, given that the images displayed seem autonomously generated by the system. But Anadol's crucial role in the entire creative process leads to fully recognizing him as the author of Unsupervised.

Anadol's creative process is articulated through various interconnected phases, each fundamental to the realization of the final work. The initial conception of the project reflects his personal artistic vision, in this case applied as an exploration of the museum's collective memory through artificial intelligence. The process develops with dataset management, where the selection of works for AI training represents in itself a creative act that decisively influences the final output. In the system design phase, Anadol selected and configured the artificial intelligence, defining specific parameters and algorithms to govern its functioning. Even the final element, the large square screen of over 7 meters on each side, is to be considered as an element of the work along with decisions related to the exhibition environment, as they transcend mere image generation becoming an integral part of the artistic experience. Finally, through his artistic interpretation, Anadol confers meaning and context to the images generated by the AI, integrating them into the overall narrative of the work.

Wanting to try to describe the technical steps and also the origin of the project name, once the dataset was created with the 138,151 elements selected from the MoMA collections, the AI training phase begins. This takes place using a sophisticated algorithm developed by NVIDIA, StyleGAN2 ADA. This phase leads to the construction of a latent space that mathematically represents the original dataset.

The latent space constitutes the core of the artificial intelligence built for the project, enclosing everything the system has learned in a multidimensional structure and therefore not directly observable. The learning process is unsupervised, a characteristic that gives the project its name, as it leaves the AI the task of autonomously identifying patterns in the dataset material. This methodological choice responds to the objective of searching for unrecognized layers of our external reality, discovering new connections, meanings, and relationships that might escape human perception.

Anadol exploits this peculiarity of unsupervised learning algorithms for the creation of original and innovative works of art. The AI, free from constraints and preconceptions, explores the immense visual space of the MoMA collection generating new interpretations, which manifest as artistic "hallucinations" capable of expanding our perception of art and reality itself.

The exploration of this latent space takes place through the Latent Space Browser, software developed specifically by Anadol's studio. This tool is not limited to a simple "observation" of the elements in the latent space but operates dynamically. The browser is distinguished by its active nature: moving through the latent space through the dynamic variation of its parameters, it "stimulates" the present elements, generating complex animations. The Latent Space Browser is the connection between the latent space and the large screen seen by the public, actively acting as an image generation engine.

Even an artisan or technician could technically execute the same workflow, but it is the artist's creative thought, present from conception and in every phase of the process, that transforms the latter into an integral part of the work of art. The difference does not lie in the actions performed but in the artistic vision that precedes and guides them, conferring on the process itself an expressive value that transcends mere technical execution. The artist guides the AI, interprets its outputs, and contextualizes them within their broader artistic vision.

The example of Unsupervised perfectly illustrates the concept of hybrid author. Anadol does not directly create every single image, but by orchestrating the entire process—from the selection of provided data (input) to the final presentation—he establishes himself as the author of the work as a whole.

The mechanism of the kaleidoscope comes to mind: the images visible inside it are unpredictable but are strictly derived from the construction technique of the instrument and the type of internal elements inserted by the builder. If the builder puts elements of only one color inside, the images visible when the eye is placed at the hole will have only that color, since the instrument cannot invent others. In the case of the kaleidoscope, is the work the instrument itself or the fleeting image that the user sees while manipulating it?

This parallel with the kaleidoscope seems to me to illustrate well the relationship between the artist and the AI as found in Unsupervised. On one hand, the kaleidoscope builder determines the possibilities of the images that will form; on the other, Anadol, through the selection of the dataset and the design of the AI system, defines the field of creative possibilities of the work. The specific images generated may be unpredictable, but they are the direct result of Anadol's artistic and technical choices. In my view, therefore, even the images generated by Unsupervised are to be considered works by Anadol, of which he is therefore the author, having decided the entire upstream process.

The work, what is it really?¶

The time has come to ask ourselves what the work is. It is an apparently simple question but the answer is complex; my idea is that the entire creative process is an integral part of the work itself.

In Anadol's case, as described in the previous paragraph, this includes the selection of works from the museum collections, the construction of the dataset, the training of the artificial intelligence, the definition of algorithms, and the choice of presentation mode. In this context, the work of art is no longer confined to the visible final product but extends to include multiple dimensions, all explicitly defined by the artist with the purpose of constructing the image that can be enjoyed by the public.

The starting point includes Anadol's artistic vision and his interpretation of the museum's collective memory through his curatorial work of selecting works to include in the dataset. A true creative act that profoundly influences the final output. Technological development, which includes the design and training of the AI, including the definition of parameters and algorithms, in turn becomes an essential component of the work.

The artist, who has long since surpassed the role of mere creator of objects to become an ideator of artistic projects, now becomes the architect of a complex creative ecosystem where technology, data, physical space, and human interaction merge into a single artistic experience. It should be remembered that we are talking about a particular work of art, extremely structured; however, as we will see later, the hybrid author can also exist with simpler systems.

Resuming the parallel with the kaleidoscope, we could affirm that the work is not only the fleeting image seen on the large screen but includes the design of the instrument, the selection of materials, and even extending to the act of observation by visitors.

To answer the question in the title of the paragraph, in the case of Refik Anadol's work, every decision in this process contributes to the meaning and definition of the work.

It can be considered that the visible element of the work—the image on the large screen in the case of Unsupervised as well as that which forms inside the cylinder of the kaleidoscope—represents the point of convergence of two aspects of the work itself, not immediately perceivable as such: on one hand, the meticulous preparation work that leads to the final vision, and on the other, the experience of enjoyment by the public. These two elements, apparently distinct, actually prove to be integral parts of the work as a whole.

In this regard, Sol LeWitt can be quoted, according to whom "the idea becomes a machine that makes the art" [See Sol LeWitt, Paragraphs on conceptual arts in Artforum vol. 5, summer 1967].

This new dimension of the work of art raises complex questions on the legal and philosophical level. How can a work that exists partly as code, partly as dataset, partly as ephemeral experience, and partly as idea be defined and protected? Where does the artist's creative contribution begin and end in such an interconnected system?

The challenge for the future will be to develop new conceptual and legal frameworks that can embrace this complexity while preserving the unique value of the human creative contribution in an increasingly technology-mediated world.

Legal implications and future of authorship in the AI era¶

The emergence of the hybrid author in the contemporary artistic panorama not only challenges our traditional conceptions of creativity but also poses significant legal questions, particularly in the field of copyright. The current legal system, based on an anthropocentric vision of creativity, now finds itself having to face scenarios for which it was not originally conceived.

An emblematic case that can help us understand the complexity of these questions is the ruling in the Cattelan-Druet case. In this case, French justice recognized Maurizio Cattelan as the sole author of his statues, despite not having materially created them. The court established that authorship resided in the construction of the project and in the direction of the work, rather than in its material execution.

However, the case of generative AI presents further complications. Unlike the artisan who faithfully executes the artist's instructions, the AI has a wide margin of maneuver in interpreting prompts and completing unspecified parts. This raises crucial questions: to what extent can the machine's output be considered a direct expression of the artist's vision? How can the boundary between the artist's creative contribution and that of the AI be determined?

The traditional copyright approach, I believe, is inadequate to answer these questions as it is based on legal principles born in an era when creativity was considered exclusively human, and therefore struggles to properly frame the role of AI in the creative process. This inadequacy becomes increasingly evident with the growing diffusion of these tools and their (pressing) technical progress.

Furthermore, the very structures of the new services offering AI use, with their common environments and specific terms of service (also known by the English acronym TOS), further challenge the current formulation of copyright.

These services not only allow but often actively encourage the sharing and reuse of prompts and images among users through the mechanism of remixing, a term that in this context could be translated as "substantial modification," in practice the possibility of being able to use others' works but only after having applied modifications such that the resulting work can be considered in turn original and capable of expressing the personality of the modifying author.

An emblematic example is Midjourney, where sharing, mutual inspiration, and reuse are an integral part of the creative experience, so much so that in the TOS we read:

Note: Midjourney is an open community that allows others to use and remix User's images and prompts whenever they are posted in a public setting. By default, User's images are publicly viewable and remixable. As described below, User grants Midjourney a license to allow this.

It is important to emphasize that this practice should not be seen as a simple appropriation of others' content. On the contrary, it wants to be a new creative method that encourages the innovative and transformative use of material produced by others to generate one's own original content through substantial modifications. This approach challenges the traditional conception of originality and authorship, promoting a more collaborative and interconnected model of creativity, so much so as to become part of a document that has the value of a contract.

This dynamic raises complex questions: how can originality be defined in a context where inspiration and reworking are so intrinsically linked to the creative process, in such an explicit way compared to what happens in the pre-AI world? How are the rights of original authors balanced with those who create new works based on shared prompts or ideas? These questions further highlight the need to rethink the foundations of copyright in the era of artificial intelligence.

A phrase that continues to circulate and that in some way gives the measure of the polemics is that of "it's not art!" (but it also applies to creativity) in relation to works generated with some AI, as if the tool used were a determining element in defining the quality of the work. It is obvious that this is a senseless polemic since the artistic level is certainly not determined a priori and based on the tools used. In this regard, I quote a phrase from the book L'autore artificiale. Creatività e proprietà intellettuale nell'era dell'AI (Ledizioni, 2023) by Simone Aliprandi, a text that I consider to be an important point of reference on the legal aspects of this world.

It makes no sense to say a priori that these creations are not art just because they are made with these systems; because in any case the possible recognition of an artistic value would occur at a later time, by people with particular sensitivity and experience in the sector (and in the case of legal dispute, also by a judge).

With his book L'autore artificiale. Creatività e proprietà intellettuale nell'era dell'AI (Ledizioni, 2023), Aliprandi offers an in-depth analysis of the question and allows us to understand the state of the art in this rapidly evolving field, with a professional eye attentive to the various implications.

Looking to the future, I think that the concept of authorship will continue to evolve alongside technological development. This could involve not only adjustments to the legal framework but also a broader reconsideration of what it means to be an author or creator in the digital age.