Rapid expansion in the capabilities and availability of AI in recent years has left the legal and regulatory worlds struggling to keep up. Legislators and regulators have sought to address many of the issues raised by AI through new law – the EU’s Artificial Intelligence (“AI”) Act being a prime example, which we have discussed previously here. However, questions remain around the extent to which traditional intellectual property protections may interact with this brave new world of AI.
Two cases due before the English courts this year may provide much-needed guidance – Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Mark and Getty Images (US) Inc & Ors. v Stability AI Ltd.
These cases are expected to create some of the first appellate court authorities on how traditional principles of patents, copyright and trademarks may apply in the context of AI. Indeed, the English Supreme Court is no stranger to landmark AI decisions: it was the first national apex court to hold that an AI machine could not be an “inventor” for the purposes of present patent law (see Thaler v Comptroller-General of Patents, Designs and Trademarks [2023] UKSC 49). While the Thaler decision was viewed as a blow to industries which rely heavily on AI innovation, given the increasing importance of AI, it remains to be seen whether the upcoming cases may indicate a more AI-friendly approach from the English courts.
Given our shared common law tradition and the similarities between our IP regimes, these English judgments will likely be influential in Ireland.
Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Mark
This case revolves around the patentability of an “artificial neural network” (“ANN”) designed to provide improved media recommendations based on emotional response. ANNs are a fundamental component of most modern AI technology and so, any judgment here may have wide-reaching consequences.
At a very basic level, an ANN is a synthetic ‘brain’: a web of inter-connected artificial neurons capable of receiving inputs from outside the system, processing those inputs in defined ways and, ultimately, producing an output.
Section 1(2) of the English Patents Act 1977 excludes “a program for a computer . . . as such” from patentability unless that program makes a “technical contribution”. The key questions for the English courts here, therefore, were whether an ANN was a “program for a computer” and, if so, whether the ANN in question made a “technical contribution”. Importantly, section 1(2) of the Patents Act 1977 implemented Article 52 of the European Patent Convention and so, the same considerations would apply to patent applications in all other signatory States, including Ireland.
In the first instance, the High Court held that the ANN was not a “program for a computer” but, even if it were, it made the requisite technical contribution for patentability. This was on the bases that:
(a) a computer program is, in essence, something that is “implementing a series of instructions pre-ordained by a human”, whereas, once initial training is complete, an ANN “operat[es] according to something that it has learned itself”. This distinction, the court concluded, meant that an ANN could not be considered a “program for a computer” for the purposes of the exclusion; and
(b) the media recommendation ultimately sent by the ANN to the end user is a sufficient technical contribution as it represents a final product which is independent of the ANN system itself and which is selected by the application of technical criteria.
The Court of Appeal disagreed. Overruling the High Court’s judgment, the Court of Appeal held that:
(a) a “computer” is “a machine which processes information” and, whether presented as hardware or software, an ANN is, therefore, a computer;
(b) there is no justification in law for making a distinction between human versus ANN-generated instructions; a computer program is “a set of instructions for a computer to do something”, regardless of whether the instructions are created by a computer or a human. By that logic, the artificial neurons of an ANN system (which control how the ANN system processes inputs) fall within the definition of a “program for a computer”; and
(c) while the ANN in question may, like all computers, have applied technical criteria to create its ultimate output (ie, the media file recommendation sent to the end user), this does not make that output itself a technical contribution. Instead, the ANN’s output relies on semantic qualities (ie, how the output file could be linguistically described as compared in the input file), something which is inherently subjective and cognitive, rather than technical.
Many proponents of AI, not least Emotional Perception AI Ltd itself, were disappointed by the court of Appeal’s treatment of ANNs (and ANN ‘neurons’) as no different to traditional computers and / or computer programs, as well as its refusal to characterise the real-world output produced by the ANN in question (which the court itself described as a “concrete task”) as a sufficient technical contribution.
On 29 November 2024, the Supreme Court granted Emotional Perception AI Ltd permission to appeal and a hearing on that appeal is expected in mid-2025. Whether the Supreme Court will implement a more generous approach than the rather high bar now set by the Court of Appeal remains to be seen.
Getty Images (US) Inc & Ors. v Stability AI Ltd
This case focusses on the use of copyrighted and / or trademarked material by AI models.
The plaintiffs are the owners or exclusive licensees of a large database of copyrighted “Visual Assets” (ie, photographs, video footage and illustrations, together with associated captions and key words), each of which bear a “Getty Images” watermark (which is also a registered trademark). These Visual Assets are licensed by the plaintiffs and / or their group companies, via their websites, for use by “creative, corporate and media customers in over 200 countries”.
The defendant is the owner and distributor of a deep-learning AI model which produces synthetic images in response to linguistic and / or visual prompts from users.
The plaintiffs allege that, without their consent and in breach of their copyright and database rights, images have been “scraped” from their database of Visual Assets and used as input to train and develop the defendant’s AI model. The plaintiffs further allege that certain of the synthetic images produced by the defendant’s AI model reproduce a substantial part of their copyrighted material and / or their trademarks (including, but not limited to, where the end user has themselves uploaded a Visual Asset as a prompt) and, therefore, represent further infringements of their intellectual property rights.
While the defendant has admitted that “at least some images from the Getty Images Website were used during the training of [its AI model]”, it is denying liability through a myriad of defences, including that no training of the AI model occurred in the UK (and so, English copyright law has no jurisdiction), that any copying / reproduction of Visual Assets would, in the context of image-to-image outputs, be attributable to acts of the user only and that any use of the Visual Assets would constitute ‘fair dealing’.
Liability issues are scheduled to be heard in June 2025. Given the defences being raised, the resulting judgment may provide important guidance on, inter alia, the territorial nature of copyright, the division of liability between providers of generative AI systems and the users of those systems, exceptions to copyright protection (and, in particular, exceptions around pastiche, text and data mining and / or temporary acts of reproduction) and use of trademarks in synthetic images – all matters of equal relevance to Ireland’s copyright and trademark regimes.
This judgment will also represent one of the first legal attempts at balancing the often competing interests of developers and content creators over how copyrighted works are utilised and / or monetised in the context of AI. It may, therefore, have significant influence on commercial relationships far outside of the individual parties to the proceedings.
Conclusion
The use of AI raises novel and complex questions for traditional intellectual property regimes worldwide. While the Irish courts are yet to confront these questions in a meaningful way, continued enthusiasm for AI use and expansion indicates that this may be only a matter of time. In those circumstances, should they find themselves facing similar legal issues, the Irish courts will likely glean significant guidance from the approaches of their English counterparts (who continue to find themselves at the forefront on various AI-related issues), including in response to the cases highlighted above.
Matheson will continue to monitor the development of the Emotional Perception and Stability AI cases, including any authoritative guidance arising.
For more information on this or on matters of AI and / or intellectual property more generally, please contact Commercial Disputes and Investigations partner, Karen Reynolds or senior associate, Naoise Cosgrove or Technology and Innovation partner, Carlo Salizzo or your usual Matheson contact.