WIPO’s revised paper on IP policy and AI

The World Intellectual Property Organisation (WIPO) has revised its draft issues paper on Intellectual Property (IP) and Artificial Intelligence (AI).

WIPO held its First Session of the WIPO Conversation on IP and AI in September 2019. It invited representatives from member states to discuss the impact of AI on IP systems, IP policies, IP rights management, and international cooperation on IP matters. In December 2019, it began a public consultation process during which feedback was requested on a draft issues paper designed to help define the most pressing questions likely to face IP policy makers as AI increases in importance. WIPO received over 250 submissions from a wide array of stakeholders across the world which were used to inform the revised issues paper, issued on 21 May 2020. Here are some key changes that have been made as a result of the submissions received during the consultation process.

1. Definitions

Many stakeholders observed that the issues paper lacked a clear definition of what was understood to fall under the terms such as “AI”, “AI-generated”, “AI-assisted”. In response, WIPO has set out a number of definitions to facilitate the conversation. “Artificial intelligence (AI)” is defined as “a discipline of computer science that is aimed at developing machines and systems that can carry out tasks considered to require human intelligence, with limited or no human intervention.” The paper adds that for these purposes, AI generally equates to “narrow AI” and it gives examples of machine learning and deep learning as two subsets of AI.

One thorny area will be the meaning of “data”. The revised paper proposes the term “data subsisting in copyright works” to distinguish between mere ideas and expressions of ideas that are protected under copyright. (The AIPPI’s UK working group on data has, similarly, wrestled with the scope of “mere data”, “data per se” and related terms.)

The revised paper also asks whether the law should attempt to define a line between “AI-generated” and “AI-assisted” outputs. The distinction may prove critical to copyright and patents and could to be the subject of considerable case law in the years to come.

2. Review of policy considerations for the patent system

The issue of inventorship in the context of patents has been hotly disputed in recent months. Indeed, the USPTO, UKIPO and EPO have each refused patent applications deliberately naming an AI as the inventor (see a discussion of the “DABUS” decisions here). The revised paper adds further considerations for policy makers. New questions include whether alternative protection schemes are available (such as trade secrets) for AI-generated inventions, whether a lack of patent protection would decrease the flow of information and technological advancement and whether it would incentivise concealment of the involvement of AI (and how to tackle that). These are undoubtedly important considerations and similar questions have been added in the revised paper for AI-generated copyright.

Further discussion arose on the topic of patentability. In particular, the revised paper identifies a discrepancy between the patent eligibility of computer programs and software in Europe and the US and queries whether there may be a need for harmonisation.

3. Copyright and datasets

The revised paper includes new questions concerning balancing incentives for human creation and technological progress, whether the potential flood of AI-generated works produced with a decreasing need for investment suggests such works should be public domain and whether special considerations should apply to “mixed training data sets” that contain both copyright works and works in the public domain.

4. Trade marks

Trade marks were not included in the initial draft issues paper. In the revised paper, WIPO notes that AI does not affect trade marks in the same way as patents, designs and copyright because trade marks do not have the equivalent of an author or inventor, but it acknowledges there may still be AI-related issues affecting trade mark law. Questions regarding ownership, marketing, brand recognition and unfair competition are identified as potential areas of issue.

5. Trade secrets

Trade secrets are also considered in the revised paper, particularly in relation to striking the right balance between the protection of innovations in AI and the legitimate interests of third parties in having access to certain data and algorithms. Striking this balance is also a focus of the European Commission and has been identified as a key issue by the UK’s upper legislative chamber (see the EC’s A European Strategy for Data and the House of Lords, Select Committee on Artificial Intelligence, AI in the UK: ready, willing and able?, page 5).

The Second Session of the WIPO Conversation on IP and AI is to be held from 7 to 9 July 2020 as a virtual meeting.

Many thanks to my colleague Tilly Berkhout for her work on this blog post.