CHINA (HONG KONG BAR): An Introduction to Intellectual Property: The Bar

Navigating IP issues in the AI era 

Generative AI systems and IP law 

Since their launch a few years ago, generative AI systems such as ChatGPT have been rapidly growing in popularity and their field of application. They have also thrown up a host of difficult IP legal issues. While some of those issues have long been anticipated and foreseen, specific national legislation has been typically slow to follow. Until legislative solutions are found, the task will fall on the courts to grapple with conundrums thrown up by new technology. In this article, we will survey a sample of common IP issues that are likely to surface in the next few years and examine possible solutions within the existing legal framework. We will also look into the suitability and adaptability of various modes of dispute resolution that may be employed in relation to AI-related disputes in the IP arena.

Ownership of generated content 

Determining the ownership of content generated by AI systems is a complex issue. Copyright laws generally attribute authorship to human creators, but AI-generated content blurs the line between human and machine creativity. This challenge becomes more significant when AI systems are capable of creating highly original and valuable works.

In the case of artistic, musical and literary works, the copyright laws of most countries that adhere to the Berne Convention require originality on the part of a human author. If the requirement is not met, copyright cannot subsist in the work. In a recent case before the US Copyright Office (2023), a question arose as to the registrability of a graphic illustration of a comic book titled Zarya of the Dawn. The artist created a vivid graphic image of a young woman using one of the commercially available AI platforms known as Midjourney. However, the Office declined to register the work, holding that the skill and labour in devising image prompts as such did not amount to any original contribution that would qualify for copyright protection. It opined that there was an insufficient causal relationship between the prompts and the graphic output generated via the AI system.

While somewhat controversial, the reasoning is in line with the case law of the Court of Justice of the European Union expressed in Cofemel v G-Star Raw (2019). The decision underscores the need for the work to express the free and creative choices made by the author herself. However, these cases leave room for arguing that where the AI system acts like a tool such as a paintbrush or conventional computer-aided design (CAD) software, the creative force and authorship should be attributed to the person behind the tool. This was indeed the argument voiced by Jason M Allen, the author of the visually stunning artwork, Théâtre d’Opéra Spatial. After it had been awarded the first prize in the 2022 Colorado State Fair’s annual fine art competition, Allen disclosed that the work had in fact been generated by Midjourney albeit after many hours of fine-tuning with the help of prompts.

Should copyright law recognise the person involved in the use of AI systems as the author (or joint author) or right-holder in some way? If so the current law may not be an adequate regulatory framework for generative AI works. A related question is who should be regarded as the owner of such works: should the owner of the AI system be recognised as at least a co-owner? It may be recalled that under the current UK law, the ownership of computer-generated works vests in the person who made the necessary arrangements for the work to be generated. At least, as between the user of an AI system and its owner, the issue of ownership is likely to be determined by express agreement, ie, the standard terms and conditions of use, which is likely to be designed in favour of the owner of the system. Whether this is a fair or acceptable outcome in all cases is a different question.

Copyright infringement 

Typically, generative AI models are trained on vast amounts of copyrighted material, leading to concerns about potential copyright infringement. If an AI system generates content that closely resembles an existing copyrighted work, it may raise legal issues and infringe upon the rights of the original creator. Already this has become a hotbed of copyright disputes.

In June 2023, Getty Images filed lawsuits in the USA and the UK against Stability AI Disputes over graphic content spawned by the latter’s Stable Diffusion platform. Some of the images generated by the platform mimicked Getty’s stock photos, complete with a modified version of Getty’s iconic watermark. Despite this, Getty did not get any payment for the use of its photos. Then, in December 2023, New York Times sued several AI platforms such as OpenAI and Microsoft over the use of the former’s news materials without permission or payment, thus free-riding on its massive investment in journalism.

Confronted with the explosive growth of generative AI technology, policymakers around the world face a common dilemma. How far should the law favour innovation at the expense of owners of existing works? There is no easy answer to the question, nor has any consensus been reached among the major countries. In early 2023, the UK government rejected a broad exception to allow greater access to materials needed for machine learning and to train AI systems. In August 2023, the Interim Measures on the Administration of Generative AI Services came into force in the PRC. They imposed obligations on generative AI service providers to ensure that the use of training data did not infringe on third-party IP rights. The regulations, however, did not spell out any details regarding the practical implementation or the consequences of non-compliance.

Dispute resolution 

If the recent cases are anything to go by, it seems inevitable that we are going to see more and more disputes arising from evolving AI technology. Another thing that can be said with certainty is that the legislative response will lag woefully behind the development of technology. Therefore, the need for an effective and efficient mode of dispute resolution is paramount in order to ensure that disruption to the operation of business and the progress of technology is kept to a minimum.

Given its well-known and unique advantages, arbitration enjoys excellent potential to be embraced by businesses as a standard and streamlined dispute resolution process. First, compared with conventional litigation, arbitration offers much-needed speed and flexibility in the resolution of AI- and IP-related disputes. In principle, the constitution and procedure of an arbitral tribunal depend entirely on the agreement of the parties, who are not constrained by issues of geography or territorial jurisdiction. Secondly, an arbitral award is final and internationally enforceable. Finality stems from the fact that the scope of an appeal is strictly limited, while international enforceability is guaranteed by the New York Convention, to which over 150 countries are signatories.

Nevertheless, the availability of arbitration is not without its limits. Here it suffices to mention two of them. First, the essential foundation of arbitration is the parties’ agreement to arbitrate. In practice, this means that, unless the parties have signed a prior arbitration agreement, litigation is still the default or fall-back option. This would be the case in most infringement disputes between the owner of an IP right and a total stranger.

Besides, the laws of some countries regard IP disputes, at least in so far as validity issues are concerned, as non-arbitrable since such issues are considered to rest solely within the exclusive jurisdiction of the sovereign state. In those countries, of which China is one, IP disputes can only be resolved through the courts or designated state agencies, such as the domestic patent office. To overcome this difficulty, some jurisdictions, such as Hong Kong and Singapore, have introduced specific legislation that allows IP disputes to be settled by arbitration. This is satisfactorily achieved on the clear understanding that the outcome of an arbitration would only bind the parties and would not affect any third parties such as state agencies.

Conclusion  

In our view, generative AI holds untold potential to improve our lives and drive innovation. As we embrace the advantages engendered by AI, lawyers and businesses alike must stand ready to face the legal challenges that come with new technology. Policymakers and other stakeholders should strive to foresee and evaluate the socio-economic harm, as well as likely benefit, of AI technology. In the meantime, we believe that arbitration promises to be one of the most business- and user-friendly options for resolving AI- and IP-related disputes.

引领人工智能时代的知识产权问题 

生成式人工智能系统与知识产权法 

近年来，以ChatGPT为代表的生成式人工智能系统自面世以来迅猛发展，被广泛应用于各种场景。然而，此类技术同时亦引发了一系列棘手的知识产权法律问题。尽管其中一些问题早已在业界的预料之中，但是不同国家在有关方面的法例却迟迟未能出台。在此背景下，在用立法作为解决方案之前，法院将肩负解决新兴技术带来的法律问题这一重任。本文将探对在未来几年中可能会出现各类典型的知识产权问题，并研究如何利用现有法律框架寻找可能的解决方案。同时，我们还将探讨在知识产权领域与人工智能相关的争议中各类争议解决途径的适用性和适应性。

人工智能生成内容的所有权 

如何确定人工智能系统生成内容的所有权是一个复杂的问题。版权法通常将著作权归属于人类创作者，但人工智能生成的内容模糊了人类与机器在创造力之间的界限。当人工智能系统能够创造出具有高度原创性及价值的作品时，这方面的问题就显得更加突出。

就艺术、音乐和文学作品而言，大多数《伯尔尼公约》（Berne Convention）缔约国的版权法都要求人类创作者具有独创性。如果这个要求不能被满足的话，那么该作品的版权将无法存续。在美国著作权局2023 年受理的一起案件中，一幅题为《Zarya of the Dawn》（黎明的扎利亚）的漫画的可注册性问题受到质疑。一位艺术家利用一个名为Midjourney的商用人工智能平台，创作了一副生动的年轻女性图像。然而，美国著作权局在审查后拒绝该作品的注册申请，认为该作者通过设计图像提示词以及通过人工智能创作出该作品的技能和工作并不构成任何原创性贡献，因此不符合版权注册的条件。此外， 美国著作权局还指出，作者输入的提示词与通过人工智能系统生成并输出的图像之间没有足够的因果关系。

尽管存在争议，该局的论证与欧盟法院在 Cofemel v G-Star Raw (2019)一案中作出的判决一致。该判决强调，作品必须表达作者本人的自由意志和创造性选择。不过，这些案例也留下了争议的余地，即如果将人工智能系统作为工具使用，例如画笔或传统计算机辅助设计（CAD）软件等工具，那么创造力和著作权则应归属于工具背后的人类创作者。艺术家杰森·艾伦（Jason M Allen）就持有该观点，他的作品“Théâtre d'Opéra Spatial”展现了卓越的视觉效果，并赢得了2022年美国科罗拉多州博览会年度美术（Fine Art）竞赛的第一名。然而，艾伦透露，该作品实际上是由人工智能工具Midjourney在提示词的辅助下经过长时间的微调后才生成的。

版权法是否应该以某种方式承认参与使用人工智能系统进行创作的人是作者（或共同作者）或权利人？如果在这样的情况下，现行法律可能并不完全适用于监管生成式人工智能作品。另一个相关问题是，谁应被视为此类作品的所有人——人工智能系统的所有人是否至少应被视为共同所有人？值得注意的是，根据英国现行法律，计算机生成作品的所有权属于为作品生成作出必要安排的人。至少在人工智能系统的使用者和其所有人之间，所有权问题很可能由明确的协议来决定，即标准的使用条款及条件，而该条款很可能被设计为有利于系统的所有人。至于该条款是否在所有情况下都能达到公平的或可接受的结果，则是另一个需要探讨的问题。

版权侵权问题 

一般通常的情况下，生成式人工智能模型依赖于用大量受版权保护的材料进行训练，因此导致潜在的版权侵权问题。 如果人工智能系统生成的内容与现有版权作品非常相似，就可能侵犯原创作者的知识产权，因而引发法律问题。这个情况已经成为引发诸多版权争议的导火线。

2023 年 6 月，盖蒂图片社（Getty Images）分别在美国和英国对 Stability AI提起诉讼，指控其旗下的 Stable Diffusion 人工智能平台生成的图片内容侵权。 该平台生成的一些图片模仿了盖蒂图片社的照片，并配有Getty标志性水印的修改版，但是盖蒂图片社并未因这些照片被使用而获得任何报酬。随后，《纽约时报》于 2023 年 12 月起诉了 OpenAI 和微软等几家人工智能平台，指控这些平台在未经许可或未支付费用的情况下使用《纽约时报》的新闻资料，免费从《纽约时报》在新闻领域的巨额投资中获利。

面对生成式人工智能技术的爆炸式增长，各国的立法机构都面临着共同的挑战：法律应当如何在鼓励创新与现有作品所有者的利益之间取得平衡？这个问题并没有非黑即白的答案，各国之间亦尚未达成共识。 2023 年初，英国政府否决了一项例外规定，该规定允许进一步放宽取得机器优化及训练人工智能系统所需材料。 2023 年 8 月，中国内地实施了《生成式人工智能服务管理暂行办法》（“办法”）。该办法规定，生成式人工智能服务提供者有义务确保训练数据的使用不侵害第三方的知识产权。 然而，该法规并未明确说明具体实施细节以及违规的后果。

争议解决方式 

近期的各种案例显示，人工智能技术的不断演进不可避免地将会引发更多的争议。另外可以肯定的是，法例规定将远远落后于技术发展的步伐。因此，寻求一种高效且实用的争议解决方式显得尤为重要，以确保对业务运营和技术发展的干扰降至最低程度。

有鉴于仲裁具有众所周知的独特优势，仲裁极有可能被企业接纳为一种解决该类争议的标准简化程序。首先，与传统诉讼相比，仲裁在处理与人工智能和知识产权相关的争议时，提供了当事人急需的效率和更大的灵活性。在仲裁中，原则上仲裁庭的组成和程序完全取决于当事人之间的的协议，这使得当事人可以不受地域或领域管辖权（即属地管辖权）问题的限制。其次，仲裁裁决具有终局性和国际可执行性。终局性源于上诉范围受到严格限制，而国际可执行性则由《纽约公约》提供保障。目前，已有150 多个国家签署了该公约。

然而，仲裁的可用性并非没有限制，我们在此仅提及其中的两个主要限制因素。首先，仲裁的重要基础是当事人同意仲裁。在实践中，除非当事人已事先签署了仲裁协议，否则诉讼仍然是默认的或备选的争议解决方式。这对于侵权人是陌生人的知识产权争议而言，可能会造成问题。

此外，一些国家或地区的法律认为知识产权争议，至少在有效性问题上，是不可仲裁的， 因为这些问题被视为主权国家的专属管辖范围内的事项。 在这些国家或地区（包括中国内地），知识产权争议只能通过法院或指定的国家机构（如专利局）解决。为解决此问题，部分司法管辖区，如香港和新加坡，已经制定了专门的法律，允许通过仲裁解决知识产权争议。 同时，在明确仲裁结果仅对当事人具有约束力，而不影响任何第三方（如国家机构）的前提下，通过仲裁来解决知识产权争议被认为是可接受的。

结论 

我们认为，生成式人工智能在提升生活品质和驱动创新方面拥有难以估量的潜力。在享受人工智能带来的优势时，律师和企业都需要充分准备，以应对新兴技术带来的法律挑战。政策制定者和其他利益相关者则应努力预测和评估人工智能技术可能对社会经济造成的危害以及可能带来的利益。与此同时，我们相信仲裁有望成为解决人工智能和知识产权相关争议的便捷且有利营商的选择。