Running from 29 October 2021 to 7 January 2022, the “Artificial Intelligence and IP: copyright and patents” consultation formed the latest round in an ongoing national conversation between the UK Intellectual Property Office (UKIPO) and interested stakeholders (see here). It followed the UKIPO’s previous, more exploratory inquiry into anticipated interactions between IP protection across the board and AI technologies, this time with limited, but more targeted questions. Therefore, the latest consultation sought mainly views on the prospects for facilitating patent and copyright protections for “inventions and creative works which are made by AI”, and for making easier the use of copyright-protected material by AI in innovation and research, through alternative approaches to licensing and text and data mining (TDM) exceptions.
On behalf of the University of Nottingham Commercial Law Centre (UNCLC), we submitted a response to the consultation that addressed the three areas of potential legislative change. Our contribution expressed our views as a research centre with direct interest in the development of IP and Commercial laws, and not as an organisation with technical experiences in AI or its applied uses as such.
Part 1 of this post focuses on the computer-generated works provisions (CGWs), while part 2 turns to the TDM exceptions.
The existing UK protection for computer-generated works
The first issue raised in the consultation was whether the copyright protection for computer-generated works (CGWs)without a human author, in the UK Copyright, Designs and Patents Act 1988 (CDPA), strikes the right balance between incentivising and rewarding investment in AI creativity. Under the CDPA, a CGW is a work “generated by computer in circumstances such that there is no human author of the work”; where this is a literary, dramatic, musical or artistic (LDMA) work, “the person by whom the arrangements necessary for the creation of the work are undertaken” is considered to be the author/first owner; and, different to the general copyright term for LDMA works, which is 70 years after the author’s death, CGW protection “expires at the end of the period of 50 years from the end of the calendar year in which the work was made”.
In view of revamping these provisions to incentivise AI investment in the UK, the consultation offered respondents three alternatives: (0) make no legal change; (1) remove the current CGW protection; (2) replace it with a new right of reduced scope/duration.
Coming to the more concrete legal points, the UKIPO sought opinions on resolving domestically the status of AI authorship. Lacking relevant solutions in international treaties and with the general topic still being debated in international fora, the UK is one of few national jurisdictions that have already provided for CGW protection. The big idea here is, apparently, that the UK aspires to take the lead in the race to becoming a primary destination for AI investment, by establishing guarantees for robust protection.
Our responses to the question about CGWs were grounded on the idea that the existing scope of protection for CGWs arguably adopts a one-size-fits-all take that does not cover the particularities of AI production. Expanding it blindly to include the incoming multifaceted wave of such works could have other, not yet in full view but easy to predict, negative repercussions.
Should the CGW provisions stay…
Therefore, despite appreciating the advantages of supporting the commercial setting with a domestic statutory provision, we considered that maintaining the status quo was the least favourable of the three options available. The CGWs protection provisions have not changed since the CDPA was originally drafted. Attempting to pre-empt the advancement of AI in the late 1980s, they reflect neither the contemporary technological reality of algorithmic and ML achievements nor our shared expectations from technological and market developments.
On the surface, the existing regime appears to be offering an easy, straightforward copyright solution. Substantially, however, we believe it puts in place an oversimplifying, perhaps inappropriate analogy: it equates complicated AI-generation – which in the example of deep neural networking entails training datasets, training algorithm, model architecture, neurons, weights and thousands of layers – with mere CGW output of conventional software. It furthermore falls short of matching the new basic standards of AI processes that involve many parties and multiple data inputs of diverse origins.
Additional flaws of significance (mentioned also in the Impact Assessment document that accompanied this latest consultation, and some even predicted back when the CDPA was debated in UK Parliament) include the long copyright term, high false attribution risk, and ambiguity over the originality requirement which is otherwise attached to all works of authorship.
…or should they go?
In our contribution we argued that removing entirely the current CGWs provisions becomes somehow more appealing in comparison to the previous option of keeping the provision. To start with, it would reduce noted ambiguities surrounding the current provisions’ application to AI-generated works. There are also numerous benefits to AI-generated works remaining in the public domain, including including enabling low-cost access to those works by others and their use for the generation of new (scientific) knowledge’.
A decrease in such copyright protections does not contradict the consultation’s policy objectives of incentivising investment and enabling competitive markets. As the Impact Assessment points out, it could actually result in AI service providers utilising more readily accessible CGWs to train AI, and thus it would increase the demand for AI. And – also confirmed in the consultation text itself – UK copyright law would still protect a wide range of AI-assisted LDMA works and entrepreneurial works made by AI.
Of course, to the extent that the market, international investment and involved industries might be seeking out the comfort of firmer legal guarantees as in AI-specific provisions, this could give the impression of a regulatory gap (something that, at least in appearances, the current CGWs provisions cover).
A third alternative: the related rights approach
Our most preferred option was to replace the current protection with a new alternative that would be suitable for AI-driven environments. We followed the approach of distinguishing between creation and dissemination of works, so as to propose transforming the current CGWs provisions into a neighbouring rights provision. We concluded that, in order to maintain a balanced system, only facsimile reproduction of the AI output (work) should be covered in the scope of rights, while adaptations ought to be excluded.
We, therefore, reflected upon relevant suggestions in scholarship, like those of introducing dissemination rights for AI outputs (resembling certain publishers’ rights) or granting tailor-made neighbouring rights to equitable remuneration (similar to the phonogram producers’ right under the WPPT). Proposals along this vein have aimed at providing disseminators with positive IPRs, so as to incentivise the sharing of AI-generated works and ensure these reach the public at acceptable costs. A turn of the focus on to related rights would also assist in avoiding noted uncertainties and shortcomings of the current regime, like the originality/authorship conundrum.
We favoured granting short protection for two years, which we believe would strike the right balance between incentivising engagement and the dissemination of AI-generated works. Considering the speed with which AI systems develop, it would not be surprising if soon most low-creativity works would somehow be the product of some sort of an AI system. Attaching to that output long term proprietary claims risks paralysing the public domain. The consultation text had raised similar concerns over the potential of AI to generate very large amounts of works extremely quickly and to flood the market with output. Therefore, the two years proposal reflected our position that IP should not be seen and treated as an instrument to regulate markets and instead encourage investment in a specific industry of interest.
Nevertheless, there is one important caveat to our support for the new alternative right. The investment narrative may be appropriate for new neighbouring rights only if stakeholders that participate in the UKIPO consultation have provided sufficient evidence on two key points: whether industries across the board would be inclined to enter into the business of generating works through AI systems if their output remained short of IPR protection; and whether, to this end, they would actually resort to the existing CGWs provisions.
Finally, we raised the question of how a potential new right requires very careful drafting, since other IPRs would inevitably be present when AI processes are utilised (e.g. potential copyright protection for software in use, database sui generis right in the training data, potential human intervention and co-creation etc.).
At the current conjuncture, it is not difficult to guess why the consultation has placed certain emphasis on the CGWs provisions’ development. The international legal community is still relatively uncertain and indecisive over AI authorship issues. Facing its own economic challenges in the interior (and yes: Brexit remains a factor), the UK has every reason to move quickly on settling its AI-related laws, so to become an attractive destination for AI investment and innovation. Therefore, if we could ever consider such a thing as a “race to regulation”, the UK could present its existing CGW provisions as an apparent advantage, so as to “get into business” as soon as possible. However, lying practically dormant for the past 30 years, without significant judicial engagement, these provisions do not offer desirable levels of certainty. The UKIPO has rightly proceeded cautiously on the matter. This is well-evidenced by the range of concerns identified and alternatives considered in the consultation documents.