With the expanding popularity of ML and AI systems, developers and investors expect those systems to be protected from unauthorised use by others.
Providing suitable protection to encourage and reward investment in the creation of intellectual property – but balancing that against the rights of the public to use those creations – is one of the core aims of the intellectual property system. Striking this balance correctly has proved difficult in this emerging area of technology, and considerable uncertainty remains in legal systems around the world as to the extent of protection afforded to ML and AI systems.
- For new and inventive AI systems and methods, patent protection is theoretically available. However, obtaining patent protection for computer-implemented inventions (CII) such as AI systems has been met with differing levels of success in different jurisdictions, and this remains the subject of continued debate worldwide.
- The key challenge for patent systems is distinguishing between protectable CII and unprotectable abstract concepts that are merely being implemented using standard computing technology.
- Copyright protection in the source code underlying AI systems, and the protection of key aspects of the method by way of confidential information or trade secrets, may present immediate and perhaps more reliable forms of protection than patents. That is because patents must be applied for and granted, whereas copyright arises on the software's creation and can be protected as confidential information/trade secrets if the right steps have been taken by the business.
- However, each form of protection has particular requirements that differ across jurisdictions, and each has its advantages and disadvantages in terms of enforcement. For example in relation to demonstrating subsistence of the copyright or trade secrets, and proving infringement, copying or breach of confidence must be shown (respectively) whereas patent protection is a monopoly right and those exploiting the invention may infringe even without knowledge of the patent.
- The considerable ongoing investment into AI will no doubt result in continuing pressure from developers and investors for suitable protection of that investment. Intellectual property regimes worldwide will likely develop to meet this demand, both through court processes and via legislation.
Patents protect inventions. While the requirements for a patent to be valid differ between jurisdictions, of key relevance for AI and ML systems is the more fundamental requirement that the subject matter of the invention be of a kind that is appropriately protected by patents. In most jurisdictions, mere ideas or abstract schemes cannot, of themselves, be patentable. This has raised particular difficulties in seeking patent protection for CII, as various jurisdictions have sought to distinguish between genuine inventions involving the use of computer technology, and the mere implementation of an otherwise unpatentable idea or some process that is already publicly available on a general-purpose computer.
In Australia, the relevant criteria for patentable subject matter is set out in s 18(1)(a) of the Patents Act 1990 (Cth), and requires that an invention is a “manner of manufacture”.
Generally speaking, an invention is a “manner of manufacture” and therefore patentable subject matter if it results in an “artificially created state of affairs” that has “economic utility”. Mere “intellectual information” such as discoveries, ideas, and business schemes are not a manner of manufacture.
There is currently some uncertainty in Australia as to how this criteria should be applied to a claim for CII. This is particularly so following the recent decision of Australia’s highest court of appeal, the High Court, in Aristocrat (read more here). In that decision, the bench was evenly split 3:3 as to whether the CII in the case – an electronic gaming machine with a particular feature game –was a “manner of manufacture” and therefore patentable. Three of the judges determined the invention was not patentable, since the invention lay entirely in the game (rather than the computer technology) and there was no “adaptation” or “alteration” of that technology to “accommodate the exigencies” of the new idea. The other three judges considered that the analysis regarding patentability should remain focused on whether the invention as claimed – an electronic gaming machine that plays a particular game – produces an “artificial state of affairs” of economic utility.
The decision in Aristocrat emphasises the importance of how a claimed invention is characterised: a key difference in the two competing approaches taken by the High Court was whether the invention was a computing apparatus (which played a particular game), or a game (which was implemented on a computer). As a consequence, developers and inventors of AI and ML systems should consider carefully how the nature of their inventions, and the ways they are separated from the prior art, can be characterised at the time they seek to obtain, or enforce, patent protection in Australia.
In the UK certain things are not patentable. These include business methods and computer programs. However the use of a computer program to effect a process can be patentable in some circumstances. In determining patentability, whilst elements of the EU/EPO test are applied (see below) the main issue is whether the "invention" is a technical advance in itself (beyond the normal interaction of software and hardware) or is simply the "digitisation" of a process that would otherwise be non-inventive.
The assessment carried out by the UK courts in relation to processes involving a computer program, is whether the process involves a “technical effect”. This is assessed by considering:
- whether the claimed “technical effect” is upon a process carried on outside the computer;
- whether the claimed “technical effect” occurs at an architectural level, irrespective of the data being processed or the applications being run;
- whether the claimed “technical effect” results in the computer operating in a new way;
- whether there is an increase in the speed or reliability of the computer; and
- whether the perceived problem is overcome by the claimed invention as opposed to merely being circumvented.
If there is a technical effect in this sense, it is still necessary to consider whether the claimed technical effect lies solely in excluded matter.
A digitised improvement to a known business process therefore might not be patentable unless the improvement has a technical effect. The results of and impact generated by AI as a CII will be the determining factor in its patentability.
In the European Union, there is a longstanding debate as to whether, and how, AI systems can enjoy patent protection.
The main issue is that CII, such as AI systems, receive differing treatment across the various national patent offices within the EU. For a complete understanding, it is necessary to examine the approach of each patent office on a jurisdictional basis.
At the European Patent level, the European Patent Convention expressly excludes "computer programs" from patentability (Art. 52(2)(c)). However, CII may be patentable insofar as the CII has a “technical” character (i.e. can provide a solution to a technical problem).
Similarly, the Guideline for Examination in the EPO (the "EPO Guidelines"), recently amended, provides that computation models and algorithms underlying AI and ML are not patentable in the abstract, but may be patentable if they have a “technical” character to serve a technical purpose.
Despite the fact that the technical character requirement is not clearly defined at EU level, the EPO has adopted a "two hurdle approach" to the patentability of CII involving AI. Under the first eligibility hurdle, it is key to identify the specific technical features in the patent claims. The second hurdle requires that that the features, which contribute to the technical character, meet the inventive step requirement.
The same EPO Guidelines clearly state that the technical features incorporating a mathematical method must contribute to the technical character of an invention, i.e.:
- by contributing to the production of a technical effect that serves a technical purpose, by its application to a field of technology; and/or
- by being adapted to a specific technical implementation.
It is therefore important for European inventors of AI applications to precisely draft the patent claims in a manner that emphasises the technical application of the CII that solves a technical problem. The claims must also disclose the invention in a sufficiently clear manner, so as to be carried out by a person skilled in the art.
Similarly to the UK and Australia, patent applications for AI systems have also encountered obstacles in the examination phase in China, on the basis that they are not appropriate subject matter for patent protection.
Under Article 2.2 of the Chinese Patent Law (CPL), an invention means any new “technical solution” relating to a product, a process or the improvement thereof. This definition for patent eligibility focuses on three elements: solving technical problems, utilizing technical means, and producing technical effects.
In addition, Article 25 of the CPL, serving as an exclusion clause, stipulates that rules and methods for mental activities are not the subject matter of patent protection. As such, simple rules or algorithms without application scenarios or technical means are excluded from patent protection.
When an AI algorithm only involves mathematical methods, it is likely to be deemed as a set of rules for mental activities and thus cannot be protected by a patent under the CPL. According to the Guidelines for Patent Examination (for public opinion) issued by the China National Intellectual Property Administration (CNIPA), the solution defined by a claim is a “technical” solution as provided for in Article 2.2 of the CPL if at least one of the following conditions is met:
- the data processed by the algorithm is data with definite technical meaning in the technical field, and execution of the algorithm can directly reflect the process of using the laws of nature to solve a certain technical problem and obtain a technical effect;
- the algorithm has a specific technical connection with the internal structure of a computer system, and can solve the technical problem of improving the hardware computing efficiency or execution effect, so as to obtain the technical effect of improving the internal performance of the computer system in conformity with the laws of nature; or
- the solution of a claim deals with big data in a specific technical field and uses algorithm tools to mine the internal associations that are in conformity with the laws of nature in the data, so as to solve the technical problem of improving the reliability or accuracy of big data analysis in a specific technical field and obtain the corresponding technical effects.
Therefore, as long as the innovation involving AI algorithms can be associated with “technicality”, it may be protected by patent.
Obtaining the maximum scope of protection for patent applicants of AI algorithms under the existing patent system and related regulations with a high grant prospect is an important challenge that patent attorneys are currently facing.
In addition to patent protection, copyright protection may be afforded to certain components of AI systems – most notably its source code. The benefits of copyright protection over and above patent protection include that:
- it arises automatically on the creation of protected subject matter (in many countries it does not require registration to be effective) and so can reduce the upfront cost of protection, and
- it arises without any requirement to prove that the work is “novel” or “inventive”, although copying must be proved for there to be infringement.
However, it is important to bear in mind that while copyright can protect the code of a computer program or AI system, it cannot protect the functions achieved by that system per se. The law of copyright is concerned with the “form of expression” of the work, not the ideas behind that expression. In the context of an AI or ML system, this does not mean that copyright will only be infringed where there has been verbatim copying of the system's source code. For example, in both the UK and Australia, it seems settled that the protection offered by copyright is not confined to the text of the code and can apply to matters at a higher level of abstraction such as its structure, sequence and organisation. In the EU, although the structure, format, sequence, and specific organisation of a work might enjoy copyright protection, the actual application of these principles can differ between single EU Member states as there is no real harmonisation in the EU for copyright. Nonetheless, the protection must be, at some level, confined to the way in which the developers chose to implement the functionality of the system, rather than the functionality itself.
Confidential information and trade secrets
More flexible protection may be provided by the law relating to confidential information or trade secrets. The information that could be protected in this manner may extend to the way an AI system operates, the data on which it is trained, and any algorithms used in its operation.
In the UK, there is a common law system of protection of confidential information which runs alongside legislative provisions on trade secrets. In the EU, in contrast to copyright, the protection of trade secrets is widely harmonised, and grants protection against unlawful misappropriation of trade secrets, provided that adequate measures to preserve the confidentiality of the information are put in place. In the UK and Australia, courts will readily assist in the restraint of unauthorised use of a wide variety of information, provided it can be established that the information is indeed confidential.
However, those seeking to rely on confidential information or trade secrets protection must be vigilant in protecting that information and guarding it against unauthorised use. Once information becomes public, it can be difficult (or impossible) to “put the genie back in the bottle” and restrain further use of that information. This can be so even if the disclosures occurred in the course of pursuing other rights – most notably patents – and so developers and owners of AI and ML systems should carefully consider which protection routes are the most practicable for their particular circumstances.
Where to from here?
Throughout history, intellectual property systems have adapted to technological frontiers, striving to provide the balance between innovators and users. AI is no different, and the continued investment in, popularity of and demand for AI will undoubtedly place pressure on national intellectual property systems to develop rapidly in order to provide suitable protection for AI and ML systems.
In the meantime, there remains a degree of uncertainty about how AI and ML systems can be protected. Inventors, developers and investors in AI should consider carefully what it is that makes their AI systems valuable, how these features can be best protected, and in what jurisdictions – perhaps using multiple IP rights in tandem to the extent achievable.
|For more on the developing area of intellectual property protection and risks for AI and ML systems, follow our blog series The IP in AI.