Skip to main content

Clifford Chance

Clifford Chance
All
Intellectual Property<br />

Intellectual Property

Talking Tech

Hong Kong government proposes to introduce copyright exception for text and data mining activities

Encouraging the development of generative AI

Intellectual Property Artificial Intelligence 25 July 2024

In the Hong Kong government's quest to promote the development of artificial intelligence  (AI) and innovation and technology more generally, it issued a public consultation paper entitled "Copyright and Artificial Intelligence" on 8 July 2024. The consultation paper proposes to introduce a new copyright exception for text and data mining activities, covering both commercial and non-commercial contexts. This will be supplemented by appropriate safeguards for copyright owners such as rendering relevant activities unauthorised if licensing schemes are available or an express reservation of rights (or opt out) has been made.

The consultation paper sets out an overview of the current regime and how issues around AI generated works may be addressed in existing Hong Kong copyright law. In addition, the consultation paper discusses deepfake and transparency issues, but the government has decided to leave these issues for another day, since they do not exclusively or directly relate to copyright.

The consultation period will end on 8 September 2024.

In this article, we will discuss the potential impact of the proposals from the perspective of AI developers and AI users.  

Impact on AI developers

Proposal to introduce specific copyright exception for text and data mining activities

The large amount of training data used for an AI model to learn is likely to involve a data mining process reproducing digital copyright works. Various ongoing lawsuits have ignited a global debate on whether use of works for training AI models is subject to permission of copyright holders under local legislation and whether law reform is necessary to provide legal clarity or certainty on the issue.

The government intends to introduce a new copyright infringement exception to allow reasonable use of copyright works for text and data mining and computational data analysis and processing for the training of AI models (text and data mining activities / text and data mining exception). The text and data mining exception is proposed to extend to both commercial and non-commercial contexts.

Safeguards for copyright owners have also been put forward:

  • requiring lawful access to the copyright work
  • rendering relevant activities unauthorised if licensing schemes are available or an express reservation of rights has been made (copyright owners are given an option to opt out)
  • restricting further dealing of the copy made under the exception
  • other possible conditions such as sufficient acknowledgment, secured storage, retention for limited purposes and destruction after certain period of time/upon request.

The Hong Kong proposal is a mix of the EU, Japan and Singapore approaches in the sense that all of these three jurisdictions have copyright exceptions for text and data mining activities for both commercial and non-commercial uses. The EU exception has an opt-out option available to copyright owners. It is noted that the UK copyright exception for text and data mining activities is limited to non-commercial research.

In terms of the consultation, the government is particularly looking for views on the appropriate conditions to the exceptions to be imposed including practical difficulties in compliance with the proposed conditions and information as to fields and industries in which licensing for text and data mining activities is common and the effectiveness of licensing as a solution. Copyright owners will be concerned as to what will constitute reasonable use of copyright works within the text and data mining exception, and how it might affect copyright owners' entitlement to remuneration for use of their copyright works and undermine copyright owners' legitimate interests in exploiting their works and market practice in licensing schemes and arrangements. On the other hand, a licensing driven approach creates additional costs and risks for the AI industry (especially without access to any collective rights management system).    

Impact on AI users

Copyright protection for AI generated works

Copyright laws around the globe generally require human authorship for affording protection to copyright works. When it comes to output from AI, various jurisdictions have stated that copyright protection does not extend to works generated by AI without a creative process under human authorship. The UK and Hong Kong copyright legislation is unique in this respect. The Hong Kong position follows the UK Copyright, Designs and Patents Act 1988 which provides that work produced by a computer, or with the assistance of one, can be afforded copyright protection and the copyright is owned by the person who enabled the generation or creation of the work.

The Hong Kong government considers that this provision in the current copyright law is sufficient to protect AI-generated works, whether through literary, dramatic, musical or artistic works (LDMA works), or sound recordings, films, broadcasts, cable programmes or typographical arrangements of published editions (non-LDMA works).

For LDMA works, the specific protection for computer generated works would cover AI-generated works with no human author, albeit with shorter duration of copyright (50 years from which the work was made) and more restrictive moral rights. For non-LDMA works, both human created and computer generated works are protected in the same manner. The government considers that the common law can be left to develop in terms of originality requirements and otherwise contractual arrangements can offer a pragmatic solution.

For an AI-generated work, potential candidates as first copyright owners may be the developer/programmer/trainer of the AI model, operator of the AI system, or user who inputs the prompts to create the work depending on the facts. The government's view is that determination of copyright ownership may in practice be solved contractually such as through investment, employment and/or commissioning contracts, assignments, or user terms, as contracts and user terms may provide for AI system owners to have copyright ownership whilst granting a non-exclusive license to users or vice versa. The government is seeking views as to whether challenges or disputes have been experienced in claiming copyright protection for AI-generated works, and if so, are there suggestions for enhancements or clarifications to the existing copyright regime to protect AI-generated works and encourage the use, development and investment in AI, or whether contractual arrangements provide a practical solution.  

Copyright infringement by AI output

Issues of infringement arise when outputs generated by AI systems resemble existing copyright works. Copying of the copyright work may be shown by evidence that the AI system was trained using the underlying copyright work. There are two relevant questions here:

  • can AI outputs be infringing works?
  • who is liable for infringement?  

The government's view is that the basic legal principles on infringement in the current Copyright Ordinance are generally applicable to determine copyright infringement arising from the creation and use of AI generated works and that these issues have all along been determined based on facts and circumstances. 

In terms of infringement liability, the government again considers that existing copyright law is sufficient to deal with the same and who is to be held liable varies depending on the circumstances. The illustrations provided show that they consider an element of culpability is required for liability for copyright infringement, namely, an AI developer/programmer/operator having control of the AI system creating the AI-generated work constituting copyright infringement should generally be the party held liable for copyright infringement. An end-user should only be held liable if he or she prompted the copying or further distributed the infringing AI-generated work.

The infringement of moral rights in AI-generated works involving LDMA works and films is similarly provided in existing statutory provisions, notably requiring attribution to the author or director be made in communication of the AI-generated works to the public and providing for the right to object to derogatory treatment of their LDMA works or films.

The consultation paper talks about market practices of employing contractual arrangements to allocate risks and liabilities and how the coverage of liability clauses may vary depending on the nature of AI systems. One of the issues sought for consultation is whether laying down rigid rules that assign infringement liability to specific persons would fail to account for the factual context of each infringement and could lead to unfairness or what changes should be made to the law to address the liability issues on copyright infringement arising from AI generated works.    

Conclusion

The government's intention in proposing the copyright exception for text and data mining activities is to promote the development of AI and innovation and technology more generally in Hong Kong by affording legal certainty and achieving an appropriate balance between protection for copyright owners and allowing reasonable use of copyright works.

The government's decision not to act in the areas of copyright protection and infringement hinge on maintaining the current sufficiently broad provisions, which should remain adaptable to changing circumstances given the fast pace of technological change. Other major jurisdictions have also found it unnecessary to make legislative reform in the area of copyright infringement involving an AI-generated work and in the area of copyright protection, a non-uniform approach has been taken and the government is taking a wait and see approach.

In terms of next steps, the government has commissioned a local research centre specialised in generative AI to examine and suggest appropriate rules and guidelines on the accuracy, transparency and information security of generative AI technology. It will also continue to closely monitor policies, regulations and initiatives by other jurisdictions and international standards.

For more on this topic, see:

Toolkits & Client Log-in