March 16, 2021
In its call for views, the UKIPO sought to understand the implications that AI might have for intellectual property (IP) policy as well as the impact IP protection might have on commercial aspects of AI in the near to medium term. The call for views considered the legal framework relating to patents, copyright, designs, trade marks and trade secrets in the UK.
Wiggin LLP submitted its response on 30 November 2020, a summary of which is set out below.
There are three broad types of AI inventions: (1) inventions of which the subject is, or part of it is, the AI itself (“AI Focussed Inventions”); (2) inventions involving the application/use of AI, which are devised by a person using AI as a tool (“AI Assisted Inventions”); and (3) inventions that are devised by an AI (“AI Generated Inventions”).
- State of Innovation: Many academic experts do not consider that AI technology is close to being ‘sentient’ or acting independently, nor do they consider society close to developing an AI whose contribution to a patentable invention is so great that there is no identifiable human contributor. AI is still at the stage where its use by humans is as a tool (albeit a complicated tool with various applications).
- Patent Activity: Globally there are approximately 150,000 active patent families (granted or pending) relating to AI technology, 13% relate to AI Focussed Inventions and 87% relate to AI Assisted Inventions. Despite challenges related to excluded subject matter (including business methods and computer programs), the rate of AI-related patents being granted is increasing, with Intel, Google and a number of Chinese institutions coming top of the AI patent league table.
- The Patent Bargain: the ‘patent bargain’ is designed to foster innovation by providing a balance between the interests of inventors who obtain a monopoly right and of society as a whole which is entitled to use the invention after the monopoly has expired. It is the inventor that is primarily entitled to a grant of a patent by benefiting from its value as a property asset. Without this benefit, the prospect of obtaining patent protection is unlikely to provide any motivation or incentive to innovate. It is difficult therefore to see how AI Generated Inventions fit within the patent bargain. AI would receive no such benefit and AI would not be motivated to innovate by the prospect of obtaining patent protection.
- Inventorship: Inventorship and ownership are closely linked. It is the inventor who is primarily entitled to the grant of a patent under UK patent law, the inventor being the devisor and ‘natural person’ who came up with or formulated the inventive concept. There can be no transfer of ownership of the invention from the AI system to the AI’s owner since the AI system does not have a legal personality so under UK law cannot itself hold any property. In relation to AI Generated Inventions, it is therefore currently not possible for an AI entity to be an inventor or co-inventor of patent applications in the UK, whether national or applied for as part of a European patent (confirmed in the Dabus decisions). Some commentators have suggested giving the AI a legal personality so that it may pass title to its owner. However it is far from clear what constitutes an AI owner and whether there are others that may reasonably claim ownership in the inventive output of the AI, for example, the developer of the underlying algorithm or the data selector/trainer. What if the AI being used was licensed from a third party or was created using open source software?
- Sufficiency of disclosure: Sufficiency is a critical part of the patent bargain and requires that disclosure is ‘sufficiently clear and complete’ so the invention may be carried out by a person skilled in the art. As it stands, technical experts do not believe there is any particular challenge in communicating a new and inventive AI technology in a way that allows a skilled person to make the product in question or work the process, and that it should be achievable without the requirement of prolonged research and without undue burden. But much depends on the nature of the invention. If the AI-related invention (whether AI Assisted or AI Generated) is complex, then it might require significant technical support in the description of the specification. But how much is enough? The skilled person should not undertake a ‘gigantic project, even if routine’ but it must be expected that any inconsistencies in training data or variations in implementation will only hinder those seeking to re-work the claimed invention. Some have suggested that the complete AI with training data could be lodged with a deposit scheme similar to that established by the Budapest Treaty which may be of particular utility where there are explainability or “black box” problems. However, software is very different to the naturally occurring micro-organisms which are the subject of the Budapest Treaty and in practice such a scheme may be difficult to operate.
- Inventive Step: It is unlikely that the level of inventive step will need to change in such a way that it cannot be accommodated by current patent law. Similarly there seems little need to extend “the person skilled in the art” to “the machine trained in the art”. Using AI as a tool may allow a skilled person to access the state of the art in a more efficient manner, but only in the same way that tools such as internet search engines have done since their inception.
- Infringement: As a non-human, AI lacks the necessary legal personhood to be held liable for the tort of patent infringement. The only party that may be so liable is that who has control of the AI itself. For AI to create something more than just information output, it is likely that there will need to be some form of human intervention that would give rise to infringement. In most circumstances loss suffered will be due to the commercialisation of an infringing product or process, and it is difficult to see how facilitating infringing sales could be done without human direction. Proving infringement may become more difficult where an AI’s operation is obscured because of its ‘black box’ working, however, efforts relating to explainability may help in this regard as may existing legal mechanisms (such as experiments).
Copyright and related rights
It is generally accepted that most AI software will be protected by copyright as a literary work. Issues may arise where AI generates work that would usually be protected as copyright, for example, in an artistic work or a literary work (such as a computer program).
Broadly speaking, however, the UK’s legal framework should allow for such AI generated work product to be protected so that it may be appropriately commercialised and licensed. Though the lack of case law means that some uncertainty exists around how some provisions should operate, caution should be exercised when considering the need for any further legislative amendments.
- Ownership/authorship: UK copyright law provides that ownership of a computer-generated work (i.e. a work which has no human author) belongs to the legal person who undertook the arrangements necessary for its creation. Commentators question what test of originality should apply to AI generated copyright works, if any, but such principles may be determined through developing case law rather than introducing further legislative provisions. Where a human cannot even be said to have “made arrangements necessary” for the creation of the copyright work, then such a work risks being authorless and will in effect have no copyright protection. However, much depends on what act might qualify as ‘making arrangements necessary”, e.g. programming, training, etc, and such a problem may not arise if the definition is drawn broadly. In any event, currently AI technologies are primarily used as tools by human authors and as such the resulting work is owned by the human creator, as it is their “own intellectual creation”.
- Infringement: Copyright can only be infringed by a person who does or authorises another to do (directly or indirectly) any of the acts restricted under UK copyright law. Typically, AI will be used as a tool by a natural person to inter alia copy a copyright work and it is difficult to envisage circumstances where it is used without at least some form of human direction or authorisation. Accordingly, it appears that the existing principles for allocating liability can apply equally in the AI context to ensure that a copyright owner may enforce its rights against a human infringer.
- Exceptions to infringement: In the UK the current fair dealing and text and data mining (TDM) exceptions to copyright infringement appear to strike the correct balance between allowing AI technologies to grow and develop for important research purposes and preventing AI systems from using the copyright works for commercial gain without remunerating the rightsholder.
- Ownership/authorship: It is not currently possible for AI to be the author/designer or owner of a UK or Community design as the statutory provisions clearly contemplate that these roles may only be fulfilled by natural or legal persons. Where AI is used to assist in the process of creating or developing a design, the existing UK legislation provides that if a human author or designer can be identified, that person will be the author or designer and entitled to be recognised as the owner of the design. Like UK copyright law, deeming provisions exist in respect of computer-generated designs. Specifically, where a human author or designer cannot be identified, the person who made arrangements necessary for the AI’s creation of the design will be deemed to be the author or designer.In a situation where the AI creates or develops the design autonomously, i.e. where a human cannot even be said to have “made arrangements necessary” for the creation of the design, then UK legislation will render the design authorless, and in effect without protection. However, as with copyright, should the provisions relating to “made arrangements necessary” be read broadly in respect of computer-generated designs then arguments may exist that the creator of the AI system itself, or an individual that uses the system and has provided training data, could be recognised as the author. Irrespective of which individual is deemed to be the author, recognising the AI itself as an author does not accord with the traditional justification for the grant of design rights as without legal personhood, one cannot own or exploit the resulting monopoly right.
- Infringement: Liability for acts of infringement provided for under UK and Community legislation in relation to registered and unregistered designs expressly rests with a ‘person’ or ‘persons’. While aspects of the acts of infringement may be automated by the use of AI, it is envisaged that a human will need to have some input for example, in regards of ‘offering or putting on to the market’. In such circumstances, the operator of the AI that is facilitating those acts may logically be liable.
Where a design is created or infringed by AI, in our view there is no reason why the notion of the “informed user” would need to change in order to assess whether there was infringement. The informed user is a legal fiction and is deemed to be a user of the product in which the design is intended to be incorporated, which is unlikely to ever be the AI system.
- Infringement: Concepts relating to trade mark infringement are founded on human interaction with branding and human involvement in the purchasing process. The legal framework, and the way that the relevant grounds for infringement and other relevant tests are set are rooted in that underlying assumption. Humans will presumably always retain choice and the ability to select an alternative product. Only if the nature of that transaction was to change, for example to involve AI in purchasing decisions in some way, would the framework of trade mark law need to adapted. UK law only envisages a natural or legal person as the infringer of a trade mark and there is no scope for an AI to be held liable for infringement. While, the actions of an AI could result in activity which might otherwise infringe a trade mark, the law does not provide for the actions of an AI to be capable of infringing a trade mark. Instead such actions are likely to give rise to liability for the operator of the AI e.g. where it is the ultimate seller of goods.
- Branding and the human consumer: It must be noted the impact of a brand on the human consumer and that human’s perception of it will remain a core part of the policy framework around which trade mark law is based. Those humans will retain, for example, their perceptions of preference, quality, and origin.
- Benefits to AI innovation: Trade secret protection is an important part of the package of rights available to those developing AI technology, which may provide significant advantages over other forms of protection in terms of effort, cost and time. Not only is it economically efficient, it can potentially last indefinitely. Moreover, innovations that are commercially valuable to AI businesses are often iterative developments or scaled versions of existing technology, which present challenges when seeking patent protection.
- Challenges with trade secrets: However, trade secrets are not without their own challenges. They do not offer the monopoly over technology that patents do and the owner must demonstrate that it has taken reasonable steps to keep the relevant information secret. Further, the attraction of indefinite protection is diminished where innovative products and technologies move on quickly or where similar results can be achieved without use of the trade secret in question.
The current array of intellectual property rights in the UK is sufficient to reward innovation in AI in much the same way as other technologies. While the scope of what is considered excluded subject matter for software (and AI) related inventions remains uncertain for many that seek patent protection, many AI related patents are nonetheless being granted throughout the world. Accordingly, the current principles of UK patent law seem sufficiently well equipped to deal with the patentability of AI. If an AI invention is not capable of being patented, for example it cannot be sufficiently explained or disclosed in a patent specification (e.g. it operates in the opaque environment of its own black box), then trade secret and copyright protection can still operate to protect aspects of the AI’s underlying technology.
While there is uncertainty around protecting IP generated solely by an AI without human intervention, it is possible that copyright and design law in the UK already provide a legislative framework that may be developed to enable innovators to own and commercialise such rights. Regarding patent protection, it is unclear whether AI, as a matter of fact, is currently or will soon be able to generate inventions that are capable of patent protection, or at least would be if they were generated by a real person. Should such a situation become a reality, the link between inventorship and ownership presents a particular challenge to patentability because of the AI’s lack of legal personhood, but it is not clear why such inventions should, as a matter of policy, be capable of benefiting from a patent system that provides very significant monopoly rights over a given technology.