15 April 2019

Moral Rights and Machines

'Australia’s experience of moral rights' by Elizabeth Adeney in (2019) 14(4) Journal of Intellectual Property Law and Practice 312–317 traces the history and outlines of moral rights legislation in Australia, including rationales, the concept of authorship and the issues this raises, and the construction of the rights and their limitations.

She comments
Moral rights were introduced into the Berne Convention in 1928. For many years, it was assumed in Australia, as in the USA, that a patchwork of existing common law and statutory protections rendered the country compliant with Article 6bis of the Berne Convention. This belief gradually eroded as the 20th century progressed. The main movement in Australia towards introduction of the rights as they now are began in 1977 and gathered momentum in the 1980s. It came to a temporary halt in 1988 when the majority of the Copyright Law Review Committee recommended against legislation for moral rights, resuming a couple of years later. A Discussion Paper followed, then an Exposure Draft, and, finally, a first amending bill in 1997. Following discussion of this bill, a second Copyright Amendment (Moral Rights) Bill was introduced in 1999. Moral rights came into force for Australia at the end of 2000. 
Four years later, in 2004, the Australian legislature enacted moral rights for certain performers in response to the WIPO Performances and Phonograms Treaty. The rights came into force in 2007, upon Australia’s accession to this treaty. In their words, the performers’ rights substantially mirror the authorial rights and exist side by side with the authors’ moral rights in the legislation. This account will focus on the author’s rights. 
... The primary reason given for the upgrading of Australia’s protection for authors was the need to bring the country into fuller compliance with Article 6bis of the Berne Convention. To this end, Australia legislated for rights that, as closely as possible, reflected the spirit and intention of the Berne negotiators, as understood in Australia. That intention was to protect the author as an individual personality. Therefore, the author is protected against certain types of personal or ‘intellectual’ harm that could come to the author through the work. The author is protected because the author as a personality is deemed worthy of protection and respect. The correct naming of the author in association with the work (with all its reputational advantages) is safeguarded by two rights: the attribution right and the right against false attribution, and the interest of the author in preventing prejudicial alteration of the work is safeguarded by the integrity right. Both rights express society’s regard for the author. As was mentioned at the Washington symposium, this individualist conception of the rights is supported by Article 27(2) of the Universal Declaration of Human Rights. It has been, internationally, the dominant conception of the rights in the post-WW2 period. Occasionally, moral rights are considered from a different angle. Instead of, or as well as, the ‘individual protection’ rationale, commentators and legislators have introduced an instrumentalist ‘public interest’ rationale or, more radically, a ‘cultural protection’ rationale into the discussion. In Australia, links were occasionally made between the safeguarding of artistic integrity and the policy of preserving a nation’s cultural heritage. At one point, the author was said in a Copyright Law Review Committee Discussion Paper to be ‘acting as a “trustee” for the public generally’.4 However, this type of thinking was resisted and has not subsequently been resurrected in Australia. 
 In referring to 'Litigation, domestic and international, and the floodgates argument' Adeney comments
The author Scott Turow’s testimony at the Washington symposium was amusing: ‘And so I called my lawyer, and I said, “Well, what are moral rights?” And he said, “Well, it has nothing to do with you. You’re an American.”’ 
Australians know from experience that moral rights do apply to American authors. Not long ago, the US musician Pitbull won a moral rights action in Australia against his Australian promoter. He was, pursuant to the Berne Convention, using Australian law—the law of the state where infringement had occurred and where protection was sought—to pursue his case. 
Nevertheless the floodgates did not open. To date, only the lowest level of the Australian Federal Court system has considered moral rights in any detail, and only substantially in two cases. In both cases, the rights of the applicant were upheld and infringement established. Modest remedies were ordered. The fact that these have been the only significant actions in a 17-year period indicates that even applicant success could not encourage a spate of litigation. 
There is no single reason for the paucity of moral rights litigation in Australia. Contributory factors may be: the impecuniousness of most artists; the likelihood that their most pressing concerns are economic rather than ‘moral’; the fact that moral rights issues are relatively easily settled before they reach the courts; and that authors, performers and practitioners lack confidence in these uncharted waters. The fact that no really authoritative case law has yet arisen means that, in most respects, understanding of the rights in their Australian context is no greater than it was in 2000. The lack of litigation may also be caused by the fear that the broad reasonableness defence will render actions on the attribution and integrity rights useless and that costs will be awarded against the plaintiff. It may also be the effect of a common law culture unaccustomed to thinking about ‘honour’ and for which ‘reputation’ is the domain of defamation law. In some sectors, moreover, authors are obliged to sign contracts curtailing their moral rights and are in a poor position to either resist, or to assess the validity of, these contracts.
Adeney concludes
So have moral rights been a success in Australia to date? For a few people they obviously have, and in many instances, users of copyright materials may have refrained from infringing behaviour because the rights exist. However, it is true to say that the Australian moral rights are not well understood, even in Australia. One reason for this is the transnational but non-uniform nature of the rights. The author or performer is likely to encounter contracts, in Australia, which are drafted in terms of foreign laws. The Australian Society of Authors moral rights clause begins with the words ‘The Author asserts his/her moral rights in relation to the work’. The Fairfax Publishing contract states ‘You waive Your moral rights (as set out in the Copyright Act 1968 (Cth)) …’.39 Who would realize, reading these documents, that assertion and waiver are not part of Australian moral rights law? Clearly, these contracts are drafted with an eye partially on the international dissemination of the material and the possibility of overseas litigation, but one cannot blame authors or performers in Australia for being confused. Their confusion will certainly grow if US moral rights become part of the mix.
Other work on the Australian moral rights regime is noted here.

'Moral rights and AI environments: the unique bond between intelligent agents and their creations' by Florian De Rouck at 299–304 in the same issue comments
‘Artificial Intelligence’ or ‘AI’ is set to become a key driver for value creation in the EU economy, with huge implications in a wide range of sectors. The current legal framework for data protection and ownership may be suboptimal for the (expected massive) flow of data arising from the use of robotics and AI. In the existing data-centred legal regimes, the importance of data quality is often overlooked. The moral rights provided to human authors under the international copyright regime may be interesting inspiration for EU legislators in developing new policies to safeguard the liens between intelligent agents and their creations. 
‘Artificial Intelligence’ or ‘AI’ is a broad term typically used to describe a whole range of software functionality designed through the cross-disciplinary effort of understanding, modelling and replicating human intelligence and cognitive processes by invoking various computational, mathematical, logical, mechanical and even biological principles and devices. 
AI systems differentiate themselves from simple automated machines which perform tasks in a structured and predictable way. Often, AI systems will be designed to perform human-like cognitive tasks, steadily improving their performance by learning from experience or external data. 
AI development has come a long way since Marvin Lee Minsky developed his neural network machine in the summer of 1951, which simulated the most basic electrical behaviour of a brain. His SNARC machine (Stochasic Neural Analog Reinforcement) used 3000 vacuum tubes and a surplus automatic pilot mechanism from a B-24 bomber to simulate 40 neurons, to solve a basic maze. Without being provided a specific program to run instructions, the SNARC started to make random connections, reinforcing each connection which brought the machine closer to its goal.  Independently from Donald Hebb, Minsky applied the first mathematical models which attempted to explain basic brain function, now known as Hebbian learning. Minsky’s machine was obviously inspired by the (limited) neuroscientific understanding of the brain in his day. 
Half a century—and a few ‘AI winters’—later, tremendous progress has been made in the six main subfields of AI (ie natural language processing, knowledge representation, automated reasoning, machine learning, computer vision and robotics) so that scholars are rekindling the idea of attaining human-level AI and even ‘superintelligence’. Using improved methodology, theoretical frameworks and approaches (such as Bayesian networks, deep learning and evolutionary algorithms), combined with access to massive data sets and computing power, many thousand (semi) autonomous AI applications which were research projects a decade ago are now deeply embedded in the infrastructure of every industry today.  McKinsey estimates investments in AI by private companies to have reached between 26 and 39 billion USD in 2016.  The European Commission’s fact sheet on AI estimates the economic impact of the automation of knowledge work, robots and autonomous vehicles to reach between 6.5 and 12 trillion EUR by 2025 on an annual basis. 
Notwithstanding the type of AI system, these intelligent agents will undoubtedly come to display increasing forms of creative thinking, as well as having the ability to learn from the content of other intelligent agents (humans). These systems will radically change the current market for books, music, paintings, films, computer programs, databases, advertisements, maps, technical drawings and other creative works. Traditional human-authored works will appear side by side with robotic works with little or no human intervention in the creative process. 
The pursuit of AI will certainly entwine with the international legal framework of copyright law established for human authors in relation to the creative expressions of their ideas. It is thus useful to explore whether these rules are able to govern appropriately AI works produced without a human author, as well. 
After having established that the originality requirement is particularly problematic for AI systems, this contribution observes a possible shift by the EU legislator from the concept of copyright for machines to alternative, data-based, sui generis regimes to regulate their protection and ownership. While such usage of machine-tailored terminology seems appropriate, it is argued that the increased focus on data should not fail to value sufficiently the importance of the unique bond between any intelligent agent and its creations. 
In doing so, this contribution advances the proposition that maintaining and promoting strong ‘liens’ between AI systems and their ‘works’ may ensure the high quality of information and knowledge in the information society, perhaps even more so than in the current, human-centred copyright regime. Given that the existing moral rights regime is tailored to humans, it proposes the development of functional equivalents to such rights for AI systems to safeguard the associated public interest goals. ...
 De Rouck considers 'the bond between intelligent machines and their works' -
Preserving the relationship between an author and his or her work has always been a key element and justification of European copyright law. The existence of moral rights is undoubtedly the strongest emanation of this bond, creating an inalienable link between the work and its creator. Moral rights of authors are historically well recognized in the legislation of the Member States of the EU and are internationally entrenched in the Berne Convention for the Protection of Literary and Artistic Works, the WIPO Copyright Treaty, and the WIPO Performances and Phonograms Treaty. In their most fundamental form, moral rights recognize an author’s right to have his or her own work attributed to him or her by name (ie the right of attribution), and to protect the work from harm (ie the right of integrity). 
Moral rights are by their very nature ‘human’ rights, in the sense that they rely on a human to wield them, so as to have works attributed to him or her, or to be able to object to an alteration, distortion or mutilation that is prejudicial to his or her honour or reputation. Although they belong to the author, the interests of these ‘moral rights’ are, however, for the general benefit of society as a whole: the right of attribution preserves the origin and authenticity of a work, while the right of attribution protects cultural works and knowledge from harm. In the knowledge preservation practice, the link between a work and its author not only aids in classifying works, but it also operates as an indicator of its subject, reliability, and quality. 
In a time where the amount of digital data already exceeds the zettabyte and information is increasingly being processed algorithmically, it would seem to be a mistake to simply discard this (albeit controversial) legal regime which protects the quality of content and information via the preservation of the bond between the intelligent agent and its work. According to Wilkinson and Gerolami, as the concern over efficient access to quality information grows, there should be greater recognition that legal protection for the moral rights of authors could play just as important a role in providing a balance between the interests of stakeholders as traditionally attributed to the economic rights regime in copyright.  
... absent a functional moral rights equivalent, how will we able to assess the origin, authority, and quality of the work provided to us by an AI if we do not know how it was programmed or which data was processed in creating it? How can we otherwise appreciate the creation of the intellectual agent? AI systems arguably provide us with the unique opportunity to truly understand the creative process of an intelligent being in all of its facets, much more than we could ever imagine for humans—to be able to know why an AI system wrote what it wrote, create the way it created. 
For all intents and purposes, the functional equivalent of the right of attribution would require some kind of self-identification of the ‘AI author’, providing the public with the means to ‘know’ the intelligent agent which has created the work (which may include the algorithmic method of processing data and a general description of the machine-learning process). Transposing the right of integrity would be more difficult, possibly requiring the provision of access to contextual data and the means to understand why certain creative choices were made by the AI system, which would include the information on the basis of which the work was created, and the intended audience. Such mechanisms arguably require further investigation.