WIPO Conversations on AI and Intellectual Property

From 7 to 9 July 2020, WIPO (the World Intellectual Property Office) held a three-day convention with interested stakeholders around the world, to discuss how the intellectual property (IP) community should address the development of artificial intelligence (AI), its impact on various aspects of IP law and policy, and whether new IP laws or guidelines are required. Over 2,000 people from 130 countries, including representatives of member states, academic, scientific and private organisations joined the meetings. The discussions involved contributions from more than 50 speakers, including presentations from senior members of the German and US judiciary (eg Judge Klaus Grabinski and Judge Kathleen O’Malley), interventions from representatives from member state IP offices (eg the UKIPO, EPO and USPTO), industry organisations (eg the International Chamber of Commerce (ICC), International Federation of Library Associations (IFLA) and Creative Commons) and industry (eg IBM and Tata Consultancy Services).

WIPO Conversations on artificial intelligence

On 27 September 2019, WIPO convened a first "Conversation on Intellectual Property (IP) and Artificial Intelligence (AI)". Following this, in December 2019, WIPO published a draft list of IP issues likely to be raised by artificial intelligence.

On 29 May 2020, WIPO published a revised issues paper on IP policy and AI, outlining the following 16 IP issues that will be affected by AI:

• issue 1: definitions;
• issue 2: patents: inventorship and ownership;
• issue 3: patentable subject matter and patentability guidelines;
• issue 4: inventive step or non-obviousness;
• issue 5: disclosure;
• issue 6: general policy considerations for the patent system;
• issue 7: copyright and related rights: authorship and ownership;
• issue 8: copyright infringement and exceptions;
• issue 9: deep fakes;
• issue 10: copyright: general policy issues;
• issue 11: further rights in relation to data;
• issue 12: designs: authorship and ownership;
• issue 13: trade marks;
• issue 14: trade secrets;
• issue 15: capacity building; and
• issue 16: accountability for decisions in IP administration.

The second set of WIPO conversations took place remotely from 7 to 9 July 2020 (the recordings are available online) and focused on the issues published in the May 2020 paper. The overarching theme was that existing IP laws are flexible enough to be applied to AI technology, but clearer guidance is required to ensure that there is an internationally consistent approach. Further, many cautioned against implementing legislative changes quickly, when AI technology is still developing rapidly. There was also a suggestion that WIPO should consider certain priority areas which require a more immediate response and others where further research should be undertaken prior to any action being undertaken.

The international discussions centred around the following five points:

1. Patent inventorship and ownership: It was unanimously agreed that a patent should be owned by a natural or legal person (and an AI machine should not be able to own an invention), but the issue of whether an AI system should be named as the inventor of a patent was widely debated. Ryan Abbott (Professor of Law and Health at the University of Surrey and part of the team that filed test cases naming DABUS AI as a patent inventor¹) argued that we have already reached a stage where AI systems are generating patents without human intervention and if we do not recognise these AI inventors, humans may be attributed credit for inventions that they did not actually devise. Further companies should be encouraged to invent efficiently. Some of the patent offices suggested that where AI was being used as a tool to create a patentable invention (for example in the case of narrow or weak AI), it was appropriate for a human inventor to be named. However, further harmonised guidance was necessary in the case of strong AI which was capable of creating inventions without human intervention. From a policy perspective, WIPO needed to consider whether the patent system was designed to incentivise the creativity of humans only or protect all technological advancements.
2. Inventive step: A distinction was drawn between AI-Assisted and AI-Generated inventions and there was a call for clear definitions to be adopted. AI-assisted inventions were generally thought to be those inventions developed with material human intervention and/or direction with the assistance of, for example machine or deep learning, whereas AI-generated inventions are those that are the output of AI without human intervention (and where AI is changing its behaviour during operation to respond to unanticipated information or events). In the case of AI-assisted inventions, where AI is used as a tool in the hands of the person skilled in the art, it would be important to consider whether the AI technology was available to the skilled person at the relevant date for assessing inventive step. In the case of AI-generated inventions, could an AI system become the equivalent of the person skilled in the art? Judge Klaus Grabinski (an eminent German patent law judge) commented that this would be more challenging, as it would be difficult to judge the inventiveness of an AI system.
3. Disclosure: Another issue identified as requiring clarification was the level of disclosure that is required to be included within patent applications. For AI-assisted inventions, it was argued that no disclosure of the AI algorithms or training data should be required, so long as there was sufficient and complete disclosure of the invention in the patent application. However, there was a debate over black box AI related inventions with the judges suggesting that patents should not be granted in those cases (as it would be difficult to explain how the AI system works, how it responds to multiple factors to solve a technical problem and so that the skilled person can reproduce the invention). However, representatives from industry suggested that there would always be “an iota of black box in an algorithm” and this was an unduly narrow restriction. It was suggested that a global depositary similar to that which has been used for the deposit of microorganisms under the Budapest Treaty may be useful for algorithms and related training data, although it would need to be internationally accepted if it was to have any credibility.
4. Access to training data: There was a broad range of support for greater access to high-quality and varied data or inputs for a variety of different reasons, such as improving AI outputs, preventing bias within AI systems, increasing access to documents to train library search systems and also leveling the playing field to prevent companies with huge piles of data from developing AI capabilities that will vastly outcompete those without access to such data. Research was presented by Sean Flynn (Chair of the Global Expert Network on Copyright User Rights) from a project examining copyright laws around the world to see whether they allowed data mining to a sufficient enough extent to enable AI machines to be trained. His team’s work showed that vast parts of the world, including the majority of South America, currently have laws in place that make this type of research and development illegal. WIPO identified this as an area for further research attention, potentially distinguishing between the situation where the data mining occurs to further research and scientific understanding generally, and where the data mining produces new works (eg music, art or literature) that compete with the original works that are mined.
5. Copyright and creativity: There were a number of submissions that human creativity is at the centre of copyright law. It was also emphasised that copyright is the economic foundation of creative sectors such as the music industry and that any weakening of copyright laws would be hugely detrimental to artists and authors. As such, we should not be quick to grant copyright protection to works created by machines. Whether there should be new rights granted to authors whose works are used to train AI machines was also discussed (and whether use of such works without permission would amount to an infringement of copyright in any event) and it was highlighted that current rules on joint ownership of copyright are not sufficient to deal with the complexities of AI systems.

Next steps

All parties are invited to submit any further interventions and comments by email by 24 July 2020. WIPO then intends to publish a document outlining the preliminary considerations from the many questions raised in the recent sessions and potentially a list of priority areas on which WIPO intends to focus. This will set the agenda for the development of clearer guidelines on how legislation should be applied to deal with AI in the intellectual property sphere. WIPO also intends to facilitate further discussions by holding a third session of the Conversation on IP and AI in November 2020.

¹To date, the patent applications filed by Dr Thaler, naming DABUS AI as the inventor, have been rejected by the UK Patent Office, the European Patent Office and the US Patent Office for failing to name a human inventor. In the UK, Dr Thaler has appealed the decision of the UK Patent Office to the English High Court. The appeal was heard on 15 July 2020 and a judgment is awaited.