Introduction

Artificial intelligence (AI) continues to be one of the most disruptive technologies of our time. With the growing availability of a host of commercial AI applications, businesses, organisations and public institutions are increasingly adopting AI in order to improve the efficiency of their operations. Even in these unprecedented times, AI technologies continue to provide important solutions to combat the ongoing pandemic. However, as AI advances, legal, technical and ethical questions arise.

Background
Due to its productive capabilities, AI challenges and poses important questions on the existing intellectual property (IP) regime. Some of these questions include:
• Whether inventions/works generated by or with assistance of AI (AI inventions) should receive IP protection.
• Whether AI inventions should be protected under the current IP system or under a sui generis system (of its own kind).
• Arising from the critical role that data plays in machine learning, whether IP rights, new or existing, should be preferred.
• Whether the resulting framework should allow or provide for free/uncontrolled use of data contained in copyright protected works/expressions for purposes of machine learning.

Cognizant of the need to map the precise impact of AI on the IP system, the World Intellectual Property Organization (WIPO) begun a conversation between its Member States and other stakeholders seeking to clarify the key issues for policy purposes.
At the initial stages, the conversation focused on understanding the practical uses of AI in IP administration. To achieve this, WIPO sent questionnaires on the use of AI applications or techniques in IP administration to national and regional IP offices (IPOs) in September 2017. Based on the responses, a meeting of IPOs was convened in May 2018 for a discussion on the ICT strategies of different offices.
The conversation then adopted a broader scope and WIPO convened the first session of the engagement in September 2019. This session was a general discussion on the broad points of intersection between AI and IP. The session had six panels covering the topics of patent, governance and development, copyright, data, and IPO administration.

WIPO then developed an initial paper containing the main questions raised to structure the conversation further. The draft issues paper was published in December 2019 and listed a total of thirteen issues for consideration. In addition, the paper called for comments from stakeholders with an aim to further correctly identify issues.

CIPIT’s Contribution and Support
CIPIT prepared and submitted a response from the global south perspective, advocating for inclusive and broad conversation that extends to the unique needs of developing and least developed nations. CIPIT is of the opinion that for the conversation to be as open and inclusive as possible, due and fair consideration should be had to developing countries’ institutional and infrastructural (in) capacities since this affects their ability to adopt AI inventions.

CIPIT found support in Professor Tom Ogada’s presentation during the first session. Prof. Ogada, the Executive Director of the African Centre for Technology Studies, noted that one of the potential consequences of AI development is the widening of the technology gap. The fact that developed countries are already leading in AI development and have greater capacity in terms of digital infrastructure means that the gap is likely to increase. In addition, developed countries have strong technological foundations for AI and investment incentives driven by a rising rate of ageing population necessitating workforce innovations. The combination of these factors highlights the need to actively consider the technological capacities of the developing countries when discussing AI and AI inventions.

Research ICT Africa, a policy think tank based in South Africa, underlined in its response the importance of the development issue. They encouraged the participants to consider the development agenda adopted by the General Assembly of the United Nations. In particular, they called for consideration of the likely impact of the grant of intellectual property rights (IPRs) for AI products on developing countries including creation of barriers to the transfer of technology.

Olugbenga, a participant from Nigeria, stressed the value of allowing each country to determine their own approach towards the interplay of AI and IP according to their community’s interests. This calls for adequate consideration for less developed and developing nations in the conversation in order to allow them accommodate the interplay in their systems.

After the first session, WIPO released the revised issues paper in May 2020 making minor amendments to the draft paper with the notable changes being the addition of sections relating to trademark and trade secrets. However, the paper made no revisions or additions to the capacity issue raised in the various interventions above.

In July 2020, WIPO convened the Second Session of the discussion based on the revised issues paper. The discussion gave rise to three pertinent approaches to the subject:

A. Caution over Urgency
One of the eminent themes in the discussion was the preference for caution over urgency. The general opinion was that a cautious approach should be taken in terms of introducing new rights within the IP framework to accommodate AI. The position was that a premature or hastened response in form of legislation would not be suitable. There were several reasons for this.

First, there was common agreement that AI is a rapidly growing technology with changing features. The continual development characteristic of AI therefore makes it difficult to adopt appropriate legislative and regulatory frameworks since these could quickly become outdated by the rapid pace of advancements. Premature approaches could risk becoming redundant in the face of these developments. Participants further suggested conducting more needs assessment studies to understand the nature and potential ramifications of AI before implementing any measures. The focus would be to identify, study and regulate only where necessary.

Second, AI development is not currently at a stage that warrants reconstruction of existing IP regimes. A number of commenters opined that autonomous generation of inventions, works or designs was still too far into the future and was not an issue that affected many jurisdictions today. Most participants seemed to agree that a degree of human intervention is still required for the AI-driven inventive/innovative process to result in a protectable invention or expression. Regulatory attempts may suffer a disconnection with the current state of technology therefore being rendered inconsequential. On the other hand, it was noted that the use of an AI application to generate an invention, work or design is a present phenomenon but one that could be adequately accommodated within existing IP regimes.
Finally, there may be need to have harmonized legislative approaches across different jurisdictions. A number of commenters were of the opinion that this new situation presents a valuable opportunity for the international community to develop consensus based harmonized IP approaches to AI and AI related technologies. A premature legislative approach risks causing legal uncertainty thereby hindering innovative activity. It was agreed that the need for international cooperation and harmony in such an approach is especially relevant since AI based products are used in a global setting with implications beyond national borders.

B. Retention of Status Quo
This is related to the call for caution over urgency. It is the view which considers the current/existing IP frameworks as sufficient for protection of AI related inventions and expressions. The specific proposal was that AI assisted inventions can be sufficiently protected under the current IP laws. It is based on the fact that the AI-based inventive process ultimately comprises of using tools that aid, albeit in a sophisticated manner, the traditional (human) process of invention.

It was also put forward that existing frameworks in some jurisdictions were flexible enough to accommodate AI-inclusive inventions/products. Examples were cited of possible patent protection of AI inventions as computer programs or mathematical methods if the invention has a direct (technical) effect. However, this was an area that was noted to be of great variation in different jurisdictions.

C. Broadening the rights of access to data
Arising from the fact that data plays an essential role in machine learning processes, the issue of access to data was one of the key discussion points. The main question was whether the use of data contained in copyright protected works without authorization for purposes of training an AI application should amount to infringement or should be allowed as an exemption.

The final speaker, Judge Kathleen O’Malley, presented on how fair use principles could potentially be used to qualify the use of data existing in copyrighted works. She discussed two cases in the US involving the use of large volumes of copyrighted literary works as starting data for machine mining. Despite the large scale nature of the copying, the Courts declared both instances to be fair use.
Majority of the participants agreed with broadening rights of access to data. The suggestion was that new exceptions to copyright (beyond fair use) should be introduced in order to allow for the copying and reproduction of protected works for purposes of data analysis. Singapore submitted that they were already proposing such exceptions. Stephen Wyber contended that this use could already be read into existing exceptions for instance in the EU’s Single Market Directive. Regardless of the legal avenues for broadening access, this view was supported by certain common arguments.

First, it was agreed that allowing unfettered access to data would serve to incentivize and promote further development of AI and AI technologies. Participants submitted that allowing broad access to data would be in everyone’s interests due to the promotion of beneficial activities such as education and research. Although a few commenters raised concerns that creating new exceptions would only weaken existing copyright regimes, others contended that such a move would be in line with the fundamental goal of the IP system to promote innovation and development for the benefit of society.

Second, it was argued that allowing such broad rights of access would not cause material harm to rights holders since such uses were non expressive and did not compete with the existing works. In addition, data mining/extraction for training purposes does not meet the standard of infringement as its processes and results do not interfere with the economic models of the works. This was a key turning point in the previously discussed cases as the Courts based their decisions on the fact that the data use did not provide to the public an alternate version of the copyrighted work. Along this line, other participants claimed that copyright does not apply at all with regard to training data as copyright protects creative works as opposed to mere presentation data.

Lastly, there was the argument that maintaining current limitations to data access would result in the development of biased AI applications. Limitations/protections have a direct effect on accessibility and availability of training products. The biases in AI inventions would arise from feeding insufficient or incomplete datasets into the applications. This would consequently prevent them from producing comprehensive and unbiased outputs.

The Discussion vis-à-vis CIPIT’s Response
CIPIT’s response proposed inclusive, gradual and incremental discussions to ensure a wholesome conversation. To this end, CIPIT aligns with the proposal for a cautious approach that was prominent during the session.

Moving forward, WIPO’s Director General suggested development of a list of priority issues drawn from the conversations to guide future discussions. Whilst CIPIT does not oppose the position that certain issues are more pressing than others, CIPIT believes that priority lists should be balanced and heterogeneous enough to allow participation of all parties. The specific concern here would be the possible exclusion of issues affecting participants in developing world. For instance, an issue that seemed to be earmarked as a priority matter during the session involved the requisite steps (for an IPO) to take when faced with a patent application listing AI as an inventor. However, as Ayokunle Adetuala from Nigeria noted, IPOs in global south nations are yet to receive such applications. This would thus seem to be an issue currently affecting developed nations whereas the issue of capacity did not receive any mention. CIPIT therefore opines that such a priority list should be balanced to cater for the needs of all Member States.

CIPIT continues to actively contribute to the conversation and shall endeavour keep readers updated on the progress of future discussions.
A recorded version of the Second Session is available.

Image Source: Flaticon