The UK IPO has launched a consultation on IP protection for Artificial Intelligence (AI) to further understand the balance between the incentives for AI development and innovation while continuing to promote human creativity. The consultation focuses on three main subject areas:

1.      Copyright protection for computer-generated works without a human author. Currently such works are protected in the UK for 50 years from date of creation, but should they be protected at all?

2.      Licensing or exceptions to copyright for text and data mining, which is often significant in AI use and development.

3.      Patent protection for AI-devised inventions. Should we protect them, and if so, how should they be protected?

The ALPSP Copyright Committee would appreciate any feedback on your view of how the IPO should approach protecting AI works and what its policy on text and data mining should be. Below is an abridged summary of the IPO consultation which explains the options the IPO is investigating, and asks specific questions for industry feedback. Please feel free to respond directly to any or all of the questions included below, or to share any general thoughts you may have by emailing leslie.lansman@springernature.com. All viewpoints, perspectives, and opinions will help us to highlight the challenges AI/TDM pose for the publishing industry, and are greatly appreciated.

Responses requested by Friday, Dec 03, 2021 to leslie.lansman@springernature.com

AI/Protection of Computer Created Works

The UK is one of only a handful of countries to protect works generated by a computer where there is no human creator. The “author” of a “computer-generated work” (CGW) is defined as “the person by whom the arrangements necessary for the creation of the work are undertaken”. Protection lasts for 50 years from the date the work is made.

Criticisms have been made of the specific provision for computer-generated works.

From a legal perspective, a computer-generated work must be original if it is to receive protection. But the legal concept of originality is defined with reference to human authors and characteristics like personality, judgement and skill. It has been argued that the law is unclear and contradictory.

From an economic perspective, some argue that copyright protection for computer-generated works is excessive. This is because computers do not need to be rewarded to produce new content, but IP rights have costs to third parties. They reason that this protection should be removed or limited to the minimum necessary. Others believe that computer-generated work protection may incentivise investment in AI technology, though they disagree on the ideal scope of this protection.

From a philosophical perspective some argue that copyright, with its roots in human authorship and creative endeavour, should only apply to human creations. They maintain that protecting computer-generated works may promote these works at the expense of human creations and devalue human creativity.

False attribution

It has been suggested that works generated by AI should be automatically tagged in some way to show their origin. Others suggested introducing a specific sanction for falsely claiming AI generated work was created by a human. There are already provisions in law that may be relied upon if works are being falsely attributed to humans. For example, the Fraud Act 2006 includes provisions to penalise people who make false representations for gain. The IPO thinks this is sufficient to address any existing or near-future issues relating to false attribution and AI. But the IPO would welcome further views on this issue.

Proposed options for protection of computer generated works

Option 0: Make no legal change

One option is to do nothing and maintain the status quo. The IPO is keen to understand the value of copyright in computer-generated works and the effect on AI investment decisions. They are also keen to understand stakeholder experiences in territories where there is no copyright protection for works without a human creator.

This option would be justified if the current approach to computer-generated works were shown to have an incentive effect in encouraging new AI-generated works and investment in AI technology. It would also be necessary for this to come without unreasonable costs to third parties, including users of these works and human creators.

Option 1: Remove protection for computer-generated works

Under this option the IPO would remove the computer-generated works provision, with the intention to limit copyright protection to human creators. Works generated by a computer would not be protected by copyright.

However, AI-assisted works with a sufficient level of human intellectual creativity would continue to be protected, as described above. Sound recordings, films, broadcasts or published editions made by AI would also continue to be protected (as these do not have an originality requirement).

This option would be justified if granting copyright for computer-generated works is not necessary to incentivise their production or has an unreasonable cost to third parties.

Option 2: Replace the current protection with a new right of reduced scope/duration

Under this option the IPO would remove the existing copyright protection for computer-generated works as described under Option 1. The IPO would provide a new type of protection instead. The duration of protection of works would be chosen to reflect the effort or investment put into their creation. In light of arguments that the present term of protection is too long, a shorter term of protection, for example 5 years, could be considered. The duration would aim to reflect the capacity of computers to generate works quickly, with little effort or human input. The term should be no longer than is needed to encourage the production of AI-generated works. A shorter term of protection would allow third parties to benefit from free use of the work once the protection had expired earlier than the current 50-year term.

The new form of protection would sit alongside any other rights that subsist in the work. This means it could subsist in a work which is a combination of human and AI creativity. At present, the special protection given to computer-generated works only applies where there is no human author. This means there is no joint authorship in a co-creation by a human and an AI system.

Under this option, the “author” of the computer-generated work would be the same as now – the “person by whom the arrangements necessary for the creation of the work are undertaken”. This is similar to how the producers of sound recordings and films are identified.

The IPO would also maintain existing entrepreneurial rights for these works. So, where AI generates a sound recording, rights in the sound recording would exist alongside the rights in the computer-generated work.

As for Option 1 there would be no change where AI has been used as a tool to assist a human creator, as this would fall under general copyright protection.

This option would be justified if there is evidence that protection for computer-generated works incentivises their production or investment in AI technology. The evidence would need to point to more limited protection than at present setting a better balance between right holders and third parties.

Questions for Feedback

Please rank these options in order of preference (most to least preferred) and explain why.

1. Do you currently rely on the computer-generated works provision? If so, please provide details of the types of works, the value of any rights you license and how the provision benefits your business. What approach do you take in territories that do not offer copyright protection for computer-generated works?

2. If the IPO introduced a related right for computer-generated works, as per option 2, what scope and term of protection do you think it should have? Please explain how you think this scope and term is justified in terms of encouraging investment in AI-generated works and technology.

3. For each/any option, what are your views on the risk that AI generated works may be falsely attributed to a person?

Text and Data Mining

Text and data mining (TDM) is the use of automated computational techniques to analyse large amounts of information to identify patterns, trends and other useful information. TDM may be used to develop and train AI and has a range of other uses including enabling research. This includes the analysis of medical and scientific data, business intelligence, and data analytics.

TDM usually requires copying of the material to be analysed. Some of this material will be protected by copyright. To data mine, including for training AI, it is often necessary to acquire a copyright licence or rely on a copyright exception.

Existing copyright framework

The UK has a specific copyright exception for TDM, which was introduced in 2014. This exception has the following features:

1. It permits the making of copies of any copyright work for the purpose of TDM for non-commercial research;

2. Researchers must have lawful access to material (for example, via subscription or permission by way of terms and conditions);

3. Publishers and content providers may apply reasonable measures to maintain their network security or stability;

4. Contract terms that stop researchers making copies of works for TDM to which they have lawful access are unenforceable;

5. Acknowledgement of the works and rights holders is required unless impractical.

Guidance for researchers was published when the exception was introduced.

TDM is also possible in other ways:

• copyright expired: If the term of protection has ended, TDM can be used on those works without permission
• temporary copying exception: This allows temporary copies to be made of most copyright works for lawful use of the work. This is only permitted as long as the copy has no independent economic significance
• licensing: As the exception covers non-commercial research TDM, licences are likely to be required when TDM is for a commercial purpose. Some right holders charge a fee to make content available under a licence. A licence might be in the form of a subscription with associated terms and conditions, or individually negotiated agreements. Alternatively, some material may be available to mine when the right holder has set up free licensing conditions in advance. Most commonly, this would be subject to a generic licence condition, such as Creative Commons or the Open Government Licence. Each licence will have conditions attached, such as attribution
• copyright does not exist: It should be remembered that not all sources of text or data will be protected by copyright or the database right. In such cases these will not restrict TDM, however, other areas of law may be relevant, including data protection, general licensing, or other contractual restrictions.

The copyright exception for TDM does not apply to the database right, so currently all TDM on databases which qualify for that right would require a licence. 

Proposed options for TDM policy

Option 0: Make no legal change

The current TDM exception would remain unchanged, as outlined above. Updated guidance on the definition of non-commercial research and what might constitute fair dealing could be produced. This would give both researchers and rights holders greater clarity about what can and cannot be done under the exception. The Post-Implementation Review (PIR) on copyright exceptions published in 2020 concluded that the TDM exception was valuable and being used for non-commercial research, as intended.

Some drawbacks of the exception were raised in the PIR, specifically the application only to non-commercial research and the fact that databases are out of scope. These could not be addressed directly through the PIR process which only looks at the impact of the exception as introduced.

This option would be justified if the current law is not an impediment to accessing material for text and data mining, in particular when training AI systems.

Option 1: Improve licensing environment for the purposes of TDM

This option would seek to ensure best practice in licensing of rights for TDM. Ways to implement this could be, for example, through educational materials, model licences (such as the Community Data Licence Agreement), or codes of practice. These could be used to assist negotiations between users and rights holders.

This option could be justified as licences can provide more legal certainty than relying on an exception. They enable copyright owners to benefit through licensing income from data mining by others, using their works. But they also have a corresponding licensing cost to those doing the mining. There are also transaction costs associated with locating rights holders, drawing up licensing agreements, etc. These costs will vary depending on who the rights holder is and whether they are used to licensing for this purpose. The opportunity to license may also encourage some rights holders to provide value-added data-mining products to their services.

As there is limited data on the use of licensing for TDM, the IPO encourages respondents to submit more information on this area to help identify the best option.

Option 2: Extend the existing TDM exception to cover commercial research and databases

This option would extend the existing exception to also allow commercial scientific research outcomes and allow TDM of databases. This would support the variety of research projects which use TDM and would allow all types of funding models to promote new discoveries.

Contract terms attempting to override the exception would be made unenforceable, and application of technological protection measures would be restricted, as with the existing exception.

Protections for rights holders would still be in place, including requiring lawful access to material (for example through subscription) and allowing reasonable measures for network security and stability.

Option 3: Adopt a TDM exception for any use, with a rights holder opt-out

This new exception would permit TDM for any use by anyone, commercial or non-commercial. It is modelled on the exception recently introduced by the EU, which was advocated by some respondents to the Call for Views.

It would not be restricted to scientific research and so would allow TDM to enable training AI systems for any purpose. It would support commercial uses of research and machine learning, as well as other TDM uses like business analytics, journalism and citizen engagement.

The exception would cover both copyright works and those protected by the database right. Users would still require lawful access to the works, which means that subscriptions and licences would still be viable for rights holders. However, once access was granted, and paid for if necessary, data mining would be permitted. Users would be able to retain the copies for as long as necessary (to verify results, for example).

However, rights holders would be able to opt-out individual works, sets of works or all their works that they did not want to be mined. An opt-out could operate through a machine-readable opt-out, so that computers mining significant numbers of works can identify works that can be lawfully mined.

Advantages of this option are that it would remove the high costs associated with mining works where licences are difficult to agree. For example, works available on the open internet, where it may be difficult to track down and negotiate with thousands of separate copyright owners. This option would still enable rights holders to license works when they wish to.

Option 4: Adopt a TDM exception for any use, which does not allow rights holders to opt out

This new exception would permit TDM for any use by anyone and would allow TDM for both non-commercial and commercial purposes, such as research, machine learning, business analytics or journalism. It would cover both copyright works and those protected by the database right.

This option would be like Japan’s exception for information analysis or Singapore’s proposed exception. These were recommended by some respondents to the Call for Views.

Lawful access would still underpin the exception and licences and subscriptions to allow such access would be permitted. However, there would be no opt out or ability to override the exception by contract.

This option is likely to be most favourable to researchers, those training AI systems, and others making use of TDM. But it would be the least favourable to rights holders, who would have their ability to license TDM restricted.

Questions for Feedback

Please rank the options in order of preference (most to least preferred) and explain why

4. If you licence works for TDM can you provide any information on the costs and benefits of these? For example, availability, number and types of works covered by such licences, and any benefits that TDM may provide you.

 5. Do you think there is a specific approach the government should adopt in relation to licensing?

 6. Do you have any experience with the EU exception with opt out for rights holders? If so, what are your thoughts?

 7. How would any of the options above positively or negatively affect you? Please try to quantify this if possible.