Table of Contents Introduction I. Status Quo: Technology and Law A. Be Careful What You Ask For: The Rise of Artificial Creativity B. Hic Sunt Dracones: AI Creativity and IP Protection 1. The Romantic Myth: No Human, No Copyright 2. Holey Gap-Filler: Neighboring Rights and Design Protection 3. The Path Untrodden: Misappropriation Doctrine II. Critique and Analysis A. Philosophical Shadow-Boxing, or Why Artificial Creativity Is Protectable B. Economic Analysis and Its Shallows 1. Lawyers ' Stepchild: The Economic Model of Copyright Protection 2. Wrong Incentive: It Was Me--Not the Robot! C. Lest We Forget: Labor Markets, Global Competition, and the Public Infosphere III. RECONCEPTUALIZATION: A ROADMAP A. Starting Point: Rights Characterization and Transferability B. Turing Reloaded: Minimum Creativity C. Another Economic Lesson (So Far Untaught): Personal Allocation of Rights 1. At First Sight: AI Programmer and Manufacturer 2. Clarification: "User" and "Investor" 3. What Really Matters: Power of Disposition Regarding AI D. Term of Protection E. Merger of Human and Artificial Creativity Conclusion "Don't mourn for me, friends, don't weep for me never, For I'm going to do nothing for ever and ever. With psalms and sweet music the heavens 'Il be ringing, But I shall have nothing to do with the singing. "
--J.M. Keynes, Economic Possibilities for our Grandchildren, in: Essays in Persuasion, 358, 367 (1930)
Artificial intelligence (AI) is on everyone's lips and seems to be in everyday use. (1) Yet the debate on its implications for our present and future has only just begun. This is true for the discussion about the changes that AI will bring to our life in general, but especially for the revolutions that it may bring to the legal sphere. To start, there is no authoritative definition of AI. The initiators of the 1956 Dartmouth Conference (the birth of AI research as an academic discipline) did not provide a definition. (2) In more recent years, AI scientists and scholars of other disciplines have attempted to describe the essence of AI, but the results are largely divergent, if not contradictory. (3) The only point of general agreement seems to be that AI systems are supposed to and do perform tasks that would normally require human intellect, particularly image recognition, language comprehension, and decision-making. (4)
Moreover, the changes that AI's evolution and proliferation will bring remain unforeseen. While some topics are in the spotlight--such as search algorithms, nursing robots, and self-driving cars--many aspects are still being discussed superficially, if at all. This particularly concerns the relevance of AI for innovative and creative activities and production, a field that is regulated by intellectual property (IP) law--but only insofar as human beings act as innovators, authors, or creators. Legal theory has only hesitantly accepted the challenges that the advent of AI will bring to IP. With respect to the protection of artificial creativity, the regulatory void is indeed dramatic. Copyright law is virtually devoid of rules and doctrines on AI. To mention but a few of the most pertinent questions: Can AI be "creative" in the sense that humans are? In other words, can a robot "create" a work of music, art, or literature? If yes, should works created by AI enjoy protection that is comparable or even equivalent to human-made works? How long should such protection last? And last but not least, who should own rights in AI-generated works?
These questions are far from fictional. Examples of so-called generative art or algorithmic art--i.e., works of art that are autonomously created by AI apparatuses--abound. (5) A very early example is AARON, a computer program from the 1970s that creates high-quality artistic paintings, many of which are indistinguishable from human-made art. (6) AARON's modern counterparts are no less impressive. Recently, an artificial neural network became famous for creating an algorithmic painting in the style of eighteenth-century portraits. The auction of an effigy of one "Edmond de Belamy" at Christie's brought $432,500--more than forty times the initially expected price. (7) In the field of music, the completion of unfinished symphonies of Franz Schubert (symphony no. 8) and Ludwig van Beethoven (symphony no. 10) by computer algorithms caught public attention. (8) Finally, ever more kinds of literature are being written with the help of so-called natural language generation. Such AI is able to directly transform raw data into readable texts, such as business and stock market reports, weather forecasts, and newspaper articles. (9) There is no doubt that had humans created any of these emanations of AI activity, copyright protection would exist.
As these examples illustrate, we are at the dawn of a third technological cataclysm. Unlike in the first two technological revolutions, the issue at hand is not the accelerated duplication of creative works or their ubiquitous dissemination via the web. Rather, we are on the threshold of an age of substitution of human creativity by artificial creativity. Hence, the task that faces us is not about optimizing human output, but rather about determining whether humankind will retain its role as creator--that is, whether the human monopoly on artistic and informative contents will persist. As a consequence, not only are labor markets at the brink of disruption, but, in the long run, cultural contents and the public infosphere are facing fundamental changes. At the same time, we must not neglect the opportunities of artificial creativity. After all, AI innovation and the concomitant proliferation of creative production by AI may sooner rather than later--and more rather than less--contribute to our economic welfare.
Against this background, the importance of more legal analysis in the field is evident. Questions of whether and how to protect artificial creativity will soon become pressing. In 2016, Japan was the first (and so far only) country to commence legislative reform aimed at offering extended protection to works produced by AI. (10) Around the same time, the European Union (EU) Parliament called on the European Commission to define criteria under which "works produced by computers or robots" could be considered protectable by copyright law. (11) But not much has happened since then, whether in the EU or anywhere else. And not surprisingly, the academic debate still focuses on the lex lata--the law as it exists--if at all. But inaction is the wrong approach. For one, problems are sure to emerge from the uncertainty of practitioners and courts who, confronted with new and untested legal issues, are left to their own devices. Second, without at least a rudimentary statutory framework for the protection and regulation of artificial creativity, the development of large parts of the industry may be at risk of withering or of following a path that could ultimately prove irreversible. (12) Both prospects are alarming.
In this Article, I will first provide an overview of the technical realities of artificial creativity and explain the legal status quo with regard to the protection of AI-generated works in different jurisdictions. My focus will be on copyright and neighboring areas of intellectual property protection, namely the protectability of AI-generated products. (13) I will then critically analyze this doctrinal status with an eye toward various legal-philosophical, microeconomic, and macroeconomic considerations of artificial creativity regulation. Lastly, using this backdrop, I will outline a reconceptualization of structures for protecting AI creativity. This may serve as a blueprint for statutory reform in copyright law and, pending reform of the statutory framework, as a guideline for the handling of AI creativity under unfair competition law and the misappropriation doctrine.
Before starting, a few clarifying remarks are in order. First, my analysis concerns cases of "autonomous AI" only, and not the use of computers or other apparatuses as tools or instruments of human creativity. (14) Production that is merely supported by the use of a computer or software--e.g., desktop publishing, video editing, and the composition of digital music--is still due to the creative direction of a human actor. In this way, computer-aided production is no different from the use of pens for writing or cameras for photography. Accordingly, its results fall into the domain of human creativity and will seldom raise new copyright issues.
By contrast, autonomous AI output is produced independently and without direct human influence. This means that humans have no immediate bearing on the process of production. Technically, such a scenario requires that an AI apparatus--as the combination of hardware and software--has evolved beyond the confines of its initial conception, typically as a consequence of so-called processes of machine learning. Machine learning is a subfield of AI research. In essence, it deals with the conception and training of algorithms to identify structure and patterns in data and to apply this knowledge to new data. (15) As a starting point, it requires human input. This input programs the AI algorithm, namely by setting its software architecture and feeding in a certain amount of data. (16) During the subsequent teaching and training phase, however, an "evolution" occurs. This evolution leads to a structuring and optimization of the data and may result in modifications to the algorithm's architecture. (17) The larger the degree of such AI evolution, the smaller the remaining human-AI connection. Ultimately, human input may still be discernible when looking at the Al's algorithm. Yet the Al's actual functions, operation, and productive output are determined entirely by its acquired capacities. With such an alteration of the process comes a loss of predictability, and AI autonomy ensues.
A cursory look...