Burkhard Schafer presents the results of research at the University of Edinburgh aiming to future-proof copyright law for the age of robotics and AI, for the Research Blog Series.
Project: Self-enforcing IP Law: Life after DRM & IP Dispute Resolution: Digital evidence, e-discovery and the changing costs of litigation
Investigators: Burkhard Schafer, Laurence Diver, David Komouves and Jesus Niebla Zattarain, University of Edinburgh
What did your research aim to do?
The aim was to future-proof copyright law for the age of robotics and AI, both in terms of substantive law, and of the software tools the legal profession will need for its implementation. In the olden days (that is, five years ago or so), two things seemed obvious: Copyright law regulates a quintessentially human activity – being creative and create art is what humans, and only humans do. By the same token, the practice of law in general, and copyright law in particular, requires uniquely human attributes, including an understanding of what creativity means, and also creative legal problem solving skills. This, many would have argued, protects both artists and the legal profession from the “fourth industrial revolution”, the replacement of labour through AI, These old certainties however are increasingly been called into question.
More and more, machines reach the market that take as their input the results of human creative activity, and then learn from it, modify it, transform it, and in some cases, create entirely new works on the back of it. AIs increasingly become consumers and producers of art, they write, paint and compose. In other applications, they interact with the output of creative works as an indirect consequence of their role, for instance when care robots read to the children they supervise. Is our legal system fit for purpose to allow robots access to the information they need, and does it give the right protection for the works they create? Can we build robots that observe copyright by design?
Creative robots do not just pose challenges for substantive copyright law, they also have the potential to disrupt the administration of law. Already now, the majority of data, often with significant commercial value, is generated not by humans but machines. This poses challenges for the law of evidence and proof, where ever increasing data size increases the costs of e-discovery. Robots as inventors could flood the patent system with applications. And automation of routine legal processes can enable undesirable litigation practice such as speculative invoicing. We therefore also explored how AI can be harnessed to face these challenges, by streamlining processes, facilitate e-discovery and potentially assist lawyers also to find “creative” legal solutions.
How did you do it?
The diversity of problems that our research tried to address also required a diversity of methods and approaches.
We worked with computer scientists and legal practitioners to develop formal models of copyright law at various levels of granularity. Using the Legal Knowledge Interchange Format LKIF and related legal ontologies, we modelled intricate but routine copyright licensing processes. Formal ontologies, but of a more demanding formal nature, were also used to model how copyright lawyers’ reason about fictional objects, such as the differences between Harry Potter and Tanja Grotter. Using link analysis and statistical tools from natural language processing, we helped lawyers uncover “hidden” connections between legal concepts that can assist in law reform, but also creative legal problem solving. And finally, we found a new way to use Petri Nets to model the interaction between copyright law and other legal domains (we chose data protection, and the Disability Discrimination Act) to assist in the development of robots that observe copyright by design.
On the second prong of our research, we combined doctrinal legal analysis with insights from media and cultural studies, economics and computing to chart the legal challenges that creative machines will pose for the copyright regime.
What are your key findings?
The much-derided DRM can get a second lease of life if not used in adversarial settings that pitch rightsholders against consumers, but if used as a tool that allows owners of creative AIs, care robots or similar systems to ensure that their machines do their job lawfully. However, to program robots to observe copyright law requires some new techniques to model legal knowledge, and can also benefit from a more rigorous theoretical grounding, to understand better what it means for machines to be law-compliant by design.
AI is still underused in copyright practice. Our work showed several proof of concept ideas how the technology can be harnessed to varying degrees of granularity and ambitiousness. They strongly indicate that legal AI can play a major positive role in addressing the challenges that the justice system and copyright administration faces if used responsibly, but also carry significant dangers and can be misused to exacerbate shortcomings of the present copyright regime.
Legal systems worldwide are ill-equipped to deal with a world where machines are increasingly both consumers and also (co)producers of creative works. This applies to legal systems across legal traditions and cultures. There are currently unintended and potentially harmful obstacles for AIs to access data due to IP restrictions (especially in machine learning). Even worse, the outputs of machine creativity are either not protected at all, potentially preventing investment in future technologies, or over-protected when they are treated as strictly analogous to human creators.
What impact has your work had so far/what impact do you anticipate it will have?
Our doctrinal work fed into a consultation by the Japanese Ministry of Internal Affairs and Communications, and a consultancy for a large commercial law firm on the use of AI in legal practice. Some of the software projects that we explored have by now been taken on and are further developed by commercial law firms. A follow-up project with an industry partner expands some of the ideas beyond copyright to other legal applications where license management and handling pays a crucial role.
We would hope that some of the other prototypes and worked cases that we developed are equally taken on by industry partners and developed to commercial strength.
Challenges encountered/Lessons learned
Highly interdisciplinary work remains a challenge, with some “up-front” cost to establish shared vocabulary, and an academic publishing environment and academic evaluation system that pays lip service to interdisciplinarity, but still strongly discourages it in practice.
Are there additional/new research questions still to be answered in this area?
Numerous, of varying degree of technicality and specificity. Legal reasoning is often very context sensitive, which is a challenge when it comes to build law compliant autonomous machines. How formal models of copyright law interact with other legal fields, and how this interaction can be modelled computationally, also remains an open issue.
Blockchain emerged as a new paradigm of automation only during the end stage of the project, and raises many issues along lines similar to the ones we tried to address.
The economic, and also wider socio-political consequences of the AI revolution on human creativity and the various professions that rely on IP law, from journalism to music and literature, video games to scientific research, are as yet ill understood and rarely substantiated with hard data.
Computational creativity research has made massive inroads in recent years, but has been ignored by researchers in legal AI. The reason for this raises numerous jurisprudential and also practical questions: how much creativity do we want/need in law, and the computer tools that assist its administration?
How has your association with CREATe helped to take things forward?
CREATE gave us access to a broad range of expertise, from AI to legal to economic, which we tried to harness in finding solutions to the questions we had asked ourselves. As importantly though, with a lead investigator who had 20 years of experience in legal AI, but not worked on copyright specific issues, the reiterative engagement with other network partners led to entirely new research questions. In valuable also the industry input, the made us realise that superficially highly abstract ideas normally discussed only in legal philosophy circles could with some modification have significant practical impact and answer pressing questions. The postgraduate community of CREATE finally allowed us to test some of our ideas on how the changing landscape of AI and (copyright) law should also change the training of the next generation of lawyers and make them “future proof”.
To find out more see the following papers: