The Death of the AI Author.

• Author: Craig, Carys
• Position: Artificial intelligence

MUCH OF THE recent literature on AI and authorship asks whether an increasing sophistication and independence of generative code should cause us to rethink embedded assumptions about the meaning of authorship. It is often suggested that recognizing the authored--and so copyrightable--nature of AI-generated works may require a less profound doctrinal leap than has historically been assumed. In this essay, we argue that the threshold for authorship does not depend on the evolution or state of the art in AI or robotics. Rather, the very notion of Al-authorship rests on a category mistake: it is an error about the ontology of authorship.

Building on the established critique of the romantic author, we contend that the death of the romantic author also and equally entails the death of the AI author. Claims of AI authorship depend on a romanticized conception of both authorship and AI, and simply do not make sense in terms of the realities of the world in which the problem exists. Those realities should push us past bare doctrinal or utilitarian considerations about what an author must do. Instead, they demand an ontological consideration of what an author must be. Drawing on insights from literary and political theory, we offer an account of authorship that is fundamentally relational: authorship is a dialogic and communicative act that is inherently social, with the cultivation of selfhood and social relations being the entire point of the practice. This discussion reorientates debates about copyright's subsistence in Al-generated works; but it also transcends copyright law, going to the normative core of how law should--and should not--think about robots and AI, and their role in human relations.

LA MAJORITE DES publications recentes sur l'intelligence artificielle et la question de l'identite d'un auteur ou d'une auteure vise a determiner si la sophistication et l'independance grandissante du code generatif ne devraient pas nous amener a repenser tout ce qui est devenu une partie integrale de notre notion de la question du droit d'auteur. Il est souvent propose que le fait de reconnaitre l'auteur ou l'auteure d'une ceuvre redigee au moyen de l'intelligence artificielle, et donc de reconnaitre la susceptibilite d'etre protege par le droit d'auteur, ne necessiterait pas un aussi grand saut d'un point de vue doctrinal que nous le pensions historiquement. Dans cet essai, nous soutenons que le seuil de determination de l'identite de l'auteur ou l'auteure ne depend pas de l'evolution ou de l'etat de la technologie de l'intelligence artificielle et de la robotique. Au contraire, la notion meme de l'identification de l'auteur ou de l'auteure d'une ouvre redigee artificiellement repose sur une erreur de categorisation : il s'agit d'une erreur ontologique de la notion du droit d'auteur. En nous basant sur la critique etablie de l'auteur ou l'auteure romantique, nous soutenons que la mort de l'auteur romantique ou l'auteure romantique entraine egalement la mort de l'auteur artificiel. La revendication de la propriete d'une cuvre par un auteur ou auteure repose sur le concept romance de l'intelligence artificielle et du droit d'auteur, et n'a aucun sens face aux realites du monde dans lequel le probleme existe. Ces realites devraient nous pousser a aller au-dela des simples definitions doctrinales ou utilitaires de ce que fait un auteur ou une auteure. Elles exigent plutot une consideration ontologique de ce qu'est un auteur ou une auteure. En nous appuyant sur les idees de la theorie litteraire et politique, nous proposons une revendication de la propriete d'une ceuvre qui est fondamentalement relationnelle: l'identification d'un auteur ou d'une auteure est un acte dialogique et communicatif qui est intrinsequement social, avec comme point central la preservation de l'individualite et des relations sociales. Cette discussion reoriente les debats sur la survie du droit d'auteur dans les cuvres generees par intelligence artificielle; mais elle transcende egalement le droit d'auteur, et va au ccur de la question, a savoir si le droit devrait--ou ne devrait pas--penser aux robots et a l'intelligence artificielle, et a leurs roles dans les relations humaines.

CONTENTS

The Death of the AI Author

Carys Craig & Ian Kerr

1. Introduction 35

2. Romanticism and The Death of the Author 45

3. Romanticizing AI 57

4. De-romanticizing Authorship 72

5. An Ontology of Authorship 81

6. Conclusion: The Death of the AI Author 86

   We should have expected this: we call them a network, we talk about devices locating and connecting to one another..., we give them the language of act and emotion: find, connect. We project ourselves--our idea of ourselves--onto them. --"Night Shift": An Essay by Orit Gat (1) I. INTRODUCTION

   The fantasy of creating artifacts that can themselves create is both old and new. In ancient times, Aristotle envisaged new instruments of production that would, of their own accord, compose and perform music and weave new textiles. (2) Towards the end of the previous millennium, science fiction writers imagined machines that would replace the proverbial million monkeys at typewriters--artificial intelligence (AI) that could not only reproduce the complete works of Shakespeare and all the books in the British Museum, but could also author the uncollected works of the future (3)--in one variation, with a view to cornering the market on fiction. (4)

   Today's AI often seems stranger than fiction (if not, perhaps, stranger than the fiction that AI has recently generated). By way of example, Sunspring, (5) a 2016 science fiction film written entirely by an AI, tells the tale of three people caught in a love triangle on a space station. The Long Short-Term Memory (LSTM) recurrent neural network that generated the screenplay (subsequently naming itself "Benjamin") was trained on a dataset of dozens of online sci-fi screenplays from the 1980s and 1990s to re-assemble sci-fi type plots and language. Over time, the AI became capable of mimicking the structure of a screenplay, including stage directions and lines of dialogue. Although Sunspring has a surprisingly sound plotline and includes some terrific one-liners ("Well, I have to go to the skull," (6) declares one character, whereupon the actor is directed to shine green lasers into his own eyes), the stage directions were sometimes a little perplexing ("He is standing in the stars and sitting on the floor" (7)), while the dialogue ranges from stilted and surreal to nonsensical. Still, as Annalee Newitz describes it, "[s]omehow, a slightly garbled series of sentences became a tale of romance and murder, set in a dark future world. It even has its own musical interlude... with a pop song Benjamin composed after learning from a corpus of 30,000 other pop songs." (8) Sunspring placed top ten in Sci-Fi London's annual film festival, beating out hundreds of other entries composed by humans.

   In addition to word assemblage, today's AIs are generating stunning abstract images that similarly raise fascinating questions about the nature of art and authorship. Consider Canadian artist and experimental composer Adam Basanta's All We'd Ever Need Is One Another--"a mixed-media installation which creates images autonomously through self-generating techniques: a continuously running 'art-factory' operating independently of human input." (9) Once produced, its outputs are "validated as art" by a machine-learning algorithm trained to spot patterns that replicate existing images found in a database of contemporary abstract art. Controversially, when one of its randomly generated images bears at least an 83 percent likeness to a known artwork, that image is automatically uploaded to a dedicated website where it is displayed as an art-factory output and titled with an auto-generated cross-reference to the similar--human-made--art.

   As Basanta describes it, "the installation acts as a golem-like assemblage, continuously and mindlessly self-producing without regard for human spectators." (10) Of course, this does not mean that human spectators have no regard for productive processes of the automaton. Amel Chamandy, Montreal artist and owner of Galerie NuEdge, alleged that Basanta's art-factory infringed copyright in her photographic work, Your World Without Paper. The claim settled, but not before attracting a slew of media attention and plenty of speculation about how the law might apply to this novel scenario. (11) The output was declared by the algorithm to be substantially similar to Chamandy's work, after all, which is the threshold for establishing copyright infringement--but only if the similarity is the result of copying. (12) An infringement claim in respect of the matching machine-generated work would be untenable here because the images were created by two desktop scanners randomly capturing the shifting light on the other's glass surface. It was not the captured images but their identification as a worthy 'artistic' output that depended on the prior existence of a similar, human-made artwork. (13) But that does not resolve the matter of the digital copy of Chamandy's work that must be in the database from which the match was identified, or any transitory digital copies made during the validation process, and whether these constitute infringing reproductions. (14) Nor does it resolve the question of potential copyright liability in future cases where a generative AI produces outputs similar to protected images within its training dataset--cases, that is, where such similarities cannot be dismissed as purely coincidental. (15) The obscurity of Basanta's algorithmic search-and-match function (with many matches, including Chamandy's 85.81 percent match, not appearing particularly similar to the human eye) should also cause us to question copyright owners' increasing reliance on AI to automatically identify infringing copies in digital environments. (16) But most importantly, as Basanta intended...