WIPO CONVERSATION ON INTELLECTUAL PROPERTY (IP) AND ARTIFICIAL INTELLIGENCE (AI)
The icon is generated from Flaticons
Recap on the third session, CIPIT’s contribution and future steps
On 4th November 2020, WIPO convened a third session as part of its continuous public consultation on IP and AI. This session focused mainly on issues in the revised issues paper not covered in the Second Session held in July. WIPO’s new Director General (DG), Daren Tang, open the session by highlighting the increasing importance AI plays in our daily lives. Using the examples of contact tracing and image screening, he noted how COVID19 has accelerated AI innovation and adoption.
In particular, the session covered definitions, trademarks, capacity building and accountability as highlighted in the revised issues paper.
1. Session One
Here, the most pressing issues were defining AI and the distinguishing AI-generated outputs vis-a-vis AI-assisted outputs.
a) Definitions: It was noted that clear and precise definitions form the foundation to establishing any constructive conversation at normative and technical levels. The key note speaker, Mr. Jean-Marc Delton, submitted that confusion within the AI field on the theories and goals of intelligence is a timely reminder of the consequences of a lack of a common ground on definitions. In addition, the definition provided by a regulator would directly determine the material scope of application. Mr. Delton attributed the difficulty in defining AI to the fact that such a definition would rely on the meaning of the term ‘intelligence’ which is difficult to distinguish from the concept of human intelligence.
The proposed broad definition of AI was noted by some participants as seemingly categorizing it as general AI.1 This is in contrast with the particular uses of AI contemplated by the revised issues paper which were application-specific and which fell more closely within the category of narrow AI.
b) Distinguishing AI-generated outputs vs AI-assisted outputs. The revised issues paper categorized these two separately with the differentiation being material human intervention. To many participants, it was however not clear what would to amount to material intervention.2 What is the threshold? How would it be assessed to determine contribution/intervention noting the varied nature of human input? This is an important issue as unclear thresholds would blur the line between human and machine inventorship/authorship. Further, participants noted that AI development has not yet attained a stage where AI technologies could be said to generate output autonomously.3 On this basis, participants seemingly suggested that focusing primarily on the consequences of AI-generated inventions would be misguided.4
Others suggested that material human intervention should not be assessed with reference to presence or absence of such interventions but rather by the degree of involvement required5 and the context i.e. the IP right under review.6 To those opposing this, their position was that the absence of significant boundary between AI-generated and AI-assisted outputs would create inconsistencies and uncertainties.7 Based on these difficulties, some suggested introduction of an intermediate stage of AI-supported outputs8 between the two categories.9
2. Session Two
The focus here was the impact of AI on Trademarks. The key note speaker, Ms. Tiki Dare,10 noted that AI poses conceptual and practical challenges to the fundamental principles of trademark law. For instance,
• Who is the average consumer in an AI controlled purchasing environment?
• Is consumers’ recollection improved or lessened by the use of AI?
These questions directly relate to how AI could potentially impact the way consumers search for, encounter, select and interact with products (goods and services). To illustrate, Ms. Dare used an example’ of a ‘smart refrigerator’ which made automatic purchases or in case of a defect suggested or chose its own repair options. Through this illustration, she described how consumers’ right and ability to choose, one of the functions of trademarks, was limited by AI. It is notable that at the point of interaction with products, AI applications can control brand information available to a consumer. Examples of AI assistants such as Amazon’s Alexa or Apple’s Siri at times act as gatekeepers between consumers and the market.11
In addition, whilst trademark law is predicated on human notions such as likelihood of confusion, brand recognition and perception, AI can reformulate this model since it cannot distinguish what would make a trademark distinctive in human terms.12 Due to the ‘blackbox’ nature of algorithmic processes, it is difficult to understand the criteria through which an AI application selected/suggested a product.13
On liability, the question is on apportioning appropriate legal responsibility when an AI application makes biased suggestions or suggests counterfeit products. One potential positive element is the adaptability of trademark law. As evidenced historically since 1920s, trademarks law has adapted from the traditional shopkeeper model, growth of modern supermarket, emergence of the internet and rise of social media. On this basis, it was cited that there was no reason why trademark law would not further adapt to recognize new issues raised by AI technologies.14
3. Session Three
a. Capacity Building
This session focused on capacity building and accountability for the use of AI in IP administration. CIPIT was honored to make its oral interventions on both of these issues through Ms. Caroline Wanjiru Muchiri.15 Ms. Wanjiru submitted that while the goal of IP systems is to promote innovation by offering protection, Africa’s innovation models exhibit preference towards open collaborative models. The incentive and capacity to innovate is not only dependent on the degree of protection offered but also on the circumstances and context of the innovator.
While some participants noted that the conversation should focus on incentivizing the people behind the machines,16 it is important to consider that for countries in global south, existing developmental constraints mean that innovators have to draw their innovation from technology spillovers and absorption from developed countries. In this regard, adopting a restrictive IP framework may counter innovation goals by hindering requisite access to technical knowledge that innovators in developing countries (DCs) require to jumpstart their innovative activities. In echoing these concerns, Mr. Patrick Mugisha17 compared AI ecosystems in different States and noted that it would not be advisable to regulate them at the international stage due to their different levels of development.
In addition, Ms. Wanjiru advocated for greater access to quality data as the availability of comprehensive datasets is vital for training and proper functioning of AI applications. Protectionist IP policies may hinder access to data required by developers where they impose stringent requirements for access. This may serve not only to disincentivize developers but also deprive them of the necessary ‘tools’ on which to base their innovative activity.
Regarding data subjects, Ms. Wanjiru submitted that data should only be shared within frameworks which would ensure utmost transparency, consideration and protection of their fundamental human rights. In line with this, CIPIT in its first intervention submitted that the registration/recognition of private rights arising out of data mined from data subjects especially DCs should always be secondary to the protection of the public interest. CIPIT argues that data should be treated in a similar manner as genetic resources arising from DCs to mitigate the high risk of predatory registration in these countries. Such private rights, whilst being enshrined in legal systems, serve to protect individual holders and at times at the expense of public interest. Mr. Yuri Zubov,18 echoed this and stated that developing AI systems should be guided by and adopt a people-centric approach with the ultimate goal of applying them to ensure protection of rights, freedoms and raising welfare standards of citizens.
b. AI in IP Administration
On AI in IP administration, Ms. Wanjiru noted that most intellectual property offices (IPOs) in DCs require infrastructural support particularly on human resource. To this end, she suggested inclusion of human capacity building activities for staff members of the IPOs to enhance institutional capacities in effectively adopting AI systems. This view was shared by several participants including Ms. Maria Covadonga19 who stated that training programmes and skill building seminars could assist in reducing the capacity gap. Mr. Konstantinos Georgaras20 noted that continued cooperation between IP offices would contribute in this regard. Mr. Patrick Mugisha called for the introduction of exchange programmes between practitioners in different countries for purposes of capacity building.
Mr. Andrey Sekretov21 urged WIPO to set up an electronic forum so as to enable experts to exchange information and findings on the subject and also organize training courses at the local levels. While supporting such initiatives, CIPIT urges for localization and customization of the trainings in a manner that reflects the individual country’s ecosystems. In addition, the specific AI systems used for the training and IP administration should be country specific and developed using local data sets to avoid instances of bias and prejudice. This is in line with the key note speaker, Ms. Nta Ekpiken who called for adoption of efforts which takes into account the practical realities of each country.
On cultural, socio economic front, Ms. Wanjiru noted that while use of AI offer significant gains in terms of efficiency, automated processes should be introduced cautiously as in some societies, automation of human led processes is associated or seen as a sign of ‘laziness’. Application of AI in may have the opposite effect of dissociating users with the IPOs. Mr. Yuri Zubov noted that a potential impediment of lack of trust in AI systems would arise among system users and general society due to lack of understanding of AI and related procedures. This may also discourage users from seeking services from the national IPOs. Caution should therefore be had to the cultural specificities in different countries especially during the adoption processes.
Mr. Patrick Mugisha suggested developing clustering initiatives on AI innovation as seen in Uganda, which other countries could replicate with WIPO’s support. Mr. Andrey Sekretov suggested the sharing of trained AI systems in use by WIPO, with partner offices receiving already trained AI models to implement. CIPIT supports such measures as they could greatly address the overreliance by national IPOs on international offices especially in the processing of international applications. However, caution must be taken in such instances to ensure shared AI applications are modified/adapted to suit local realities to prevent unintended discriminative or biased outcomes.
The rich discussions in the third session showed that there remain several issues arising at the nexus of IP and AI. On defining AI and distinguishing AI-generated and AI-assisted outputs, more work lies ahead in terms of coming up with precision and creating clarity among them. There is need to continue with the international and multidisciplinary conversations within WIPO’s framework to achieve comprehensive and inclusive conclusions.
On capacity building and from contributions made, it is clear that there are a number of ways to support DCs in efforts to reduce or contain the existing gaps. Suggestions on collaborations through training and skill building seminars22 should be explored to strengthen the capacities of national IPOs. The distinct challenge for IP policy is in balancing incentivizing innovation by offering reasonable degree of protection and facilitating a level of open collaboration to support innovative activities.
Lastly, at the close of the session, the DG noted that the ongoing discussions have been quite conceptual and suggested narrowing the focus to certain concrete issues. In particular, there was the suggestion that next steps should be to organize discussions with policymakers and IPOs on the aspects of AI that affect their operations. Whilst CIPIT is aligned with the call to narrow the focus of the conversation, we submit that restricting participants may take away from the holistic approach required by this conversation. This is especially because the issues raised have potential implications on various stakeholders beyond IPOs. The multi-stakeholder and multidisciplinary contributions have played an important part in enriching the ongoing conversation. Further, as shown by WIPO’s previous meetings, a limited number of national offices from DCs engage in such sessions but there are participants from other sectors.
As noted by Mr. Vitor Ido,22 the views from IPOs and governments from DCs need to feature more prominently since their status may not be the same as that of the IPOs in developed countries. In structuring the conversation, engaging participants from different fields will make the discussions inclusive ad comprehensive.
CIPIT continues to encourage a holistic approach to development of a legal (IP) framework that maximally benefits all stakeholders.
See CIPIT’s oral intervention here. The full video of the third session can be found here.
1 Mr. Stefan Ditmer, Mr. Jonathan Osha, Mr. Leopoldo Belda-Soriano
2 Mr. Stefan Dittmer, Mr. Takeshi Ueno, Ms. Komal Kalha,
3 Mr. Leopoldo Belda-Soriano, Mr. Jonathan Osha
4 Ms. Daniela Simone
5 Mr. Stefan Dittmer, Ms. Luo Li
6 Ms. Daniela Simone
7 Mr. Claes Hedlund
8 Ms. Luo Li suggested the division of AI outputs into the categories of AI-supported, AI-assisted and AI-generated outputs which differ in terms of substantial contribution, the degree of human intervention and AI’s capabilities in producing those creative outputs.
9 Mr. Jonathan Osha also proposed a three tier classification of AI output with the categories being Core-AI outputs, application-specific AI outputs and AI- generated outputs. Mr. Takeshi Ueno also suggested a three part division but using inventor eligibility as the determinant factor rather than material human intervention. Mr. Takeshi’s three tier classification was: invention generated by AI with the intervention of an eligible human inventor; inventions generated by AI with the intervention of an ineligible human inventor and inventions generated by AI without human intervention.
10 President-Elect, International Trademark Association (INTA) and Assistant General Counsel, Oracle, Burlington, United States of America
11 Mr. Lee Curtis
12 Mr. Stefan Dittmer, Mr. Lee Curtis
13 Mr. Leopoldo Belda-Soriano
14 Mr. Lee Curtis
15 Ms. Caroline Wanjiru Muchiri, Research Fellow and IP expert, CIPIT.
16 Ms. Daniela Simone
17 Innovation & Intellectual Property Management, Ministry of Science, Technology & Innovation (MoSTI), Kampala
18 Federal Service for Intellectual Property (Rospatent), Moscow
19 Department of Legal Coordination and International Relations, Spanish Patent and Trademark Office, Madrid
20 Canadian Intellectual Property Office (CIPO), Gatineau
21 Eurasian Patent Office, Moscow, Russian Federation
22 See interventions by Mr. Konstantinos Georgaras and Maria Covadonga during the third part of the session.
23 Intellectual Property and Biodiversity Programme (HIPB), South Centre, Geneva