The global pandemic has accelerated many aspects of life, perhaps none more obviously than the increasing amount of time that individuals have spent online. Following this spike, investment into virtual reality and augmented reality technologies has fast-tracked and we are now seeing the emergence of what many have coined ‘the metaverse’ – a shared virtual space that allows you to interact with virtual objects in real life with real-time information.
From the world’s first virtual fashion show through to the recent scramble by crypto-rich investors to snap up virtual land and digital-only creative works, the metaverse’s ascendancy into public consciousness seems inevitable. Whatever shape this emerging space takes, it will almost certainly be characterized by an overlay of unfathomably vast amounts of data, very large parts of which will be generated directly by machines rather than humans. Whereas in a pre-metaverse world almost everything is somebody’s creation and intellectual property, the metaverse calls into question whether creations amount to intellectual property at all. This raises a large question mark over the sustainability of AI applications in the space.
AI and IP
As AI applications rapidly gain the ability to behave as autonomous entities capable of generating creative works, they increasingly challenge the norms and assumptions underlying intellectual property protection. Can AI applications digest massive databases that include copyright works and use machine learning to author their own creative works without infringing copyright in the original data? Do the outputs of AI applications enjoy their own copyright protection?
Challenges to copyright raised by AI applications arise well before they create their first work, at their input or training stage. Particularly where they trawl public networks, aggregating vast amounts of ‘information’, they inevitably encounter copyright works such as videos, songs, novels and news stories. In this regard, any need for the authorisation of the owners of such content potentially restricts AI systems’ use of that information and exposes their operators to the dangers of infringement. While not all 'information' is protected by concepts such as ownership or intellectual property protection, large parts of it are. While the simple reading by a human of such information does not constitute a restricted copyright act, the acts of copying or reproduction carried out by a machine learning system often are.
Challenges of national copyright laws
Complexity arises here because the copyright exceptions on which operators of AI applications often rely, to steer clear of infringement generally, apply on a national basis, a notion which seems almost quaint in the connected world. So even in respect of information which clearly enjoys copyright protection, the doctrine of fair use in the United States, specific machine learning exceptions in jurisdictions such as Japan, or the more limited text and data mining exceptions in European law sometimes apply. An analysis of legality hangs on questions which have bedevilled previous creators and users of content in the Internet world – where is the information, where are the relevant acts being carried out, and which law applies? Even when the answer to the last question is clear, what the relevant law permits is another question.
Even if the creation and training of a machine learning model is not infringing, the outputs of an AI system raise their own problems. Having been trained on a particular type of data, if the system’s outputs are substantially similar to its inputs, then they may constitute unauthorised infringement of the copyright in the underlying inputs. Again, differences in national laws come to bear, where the concept of a derivative work in the United States is wider than the concept of modification in the UK.
Even if the outputs are sufficiently dissimilar to not be infringing, the crucial question arises whether the content created by an AI system is itself capable of copyright protection. International copyright law is founded in human-centric concepts of personal expression, authorship and originality as prerequisites for the existence of copyright. Outputs generated purely by AI systems challenge these norms. Even the UK’s unique provision governing computer generated works, where the person “by whom the arrangements necessary for the creation of the work are undertaken” is deemed the author, elide this debate by finding a human to whom authorship can be attributed. The UK Government is currently carrying out a major consultation and review of its intellectual property regime looking at precisely such points.
Rights of a machine creator
Likewise, traditional justifications for copyright protection, such as incentivising the creation of works or protecting the natural rights of creators, break down when the creator is a machine requiring no incentivisation. Even more pertinently, traditional notions of copyright ignore the very real question of incentives for the creators of AI systems. If the primary purpose of copyright law is to promote the production of creative works by providing an economic incentive to authors through the protection of their works, the same might – potentially wrongly – be thought to apply with regards to investment in technologies to create works which on the surface express extreme creativity and originality. Yet if an AI system cannot be an author, the most extreme conclusion is that the works it creates cannot enjoy copyright protection, and simply become part of the public domain. Their ensuing free distributiveness is an unpalatable prospect for an investor in an AI system seeking to monetise its results.
The certainty, and the key point, of the metaverse is that it will comprise massively greater volumes of information generated by and exchanged between machines with little to no human prompt, in the absence of which the intellectual property treatment of that creation and exchange is at best uncertain and at worst inapplicable. This in turn gives rise to the fear of a corresponding increase in intellectual property disputes, and the consequent uncertainty and risk introduced into business models and investment decisions. Potentially loading cost (for example of licensing and administration) into the input end of the business, while casting into doubt the ability to monetise outputs, has as negative an effect on the risk/reward balance as it is possible to imagine. The clearest point in this area is that intellectual property law will need to evolve with the burgeoning world of AI, whether by way of judicial decisions on the applicability of existing laws, or where necessary by specific decisions of policy makers to amend those existing laws to make them fit for purpose. That much will be necessary in a world where the transformative creativity and productivity of the machine promises to generate new content on hitherto unimaginable scales.
Tomos Jones is a senior associate at Reed Smith, where he acts across a wide range of media and technology work, with a particular focus on content rights and distribution.