Discover how Artificial Intelligence (AI) intersects with copyright law as recent court cases and global developments shape the ongoing debate. Gain insights into how different countries, including the UK, India, the US, the EU, and Japan, navigate the authorship of AI-generated works within their copyright frameworks. Explore the fair use argument put forth by AI companies and delve into four distinct schools of thought on fair use. Stay informed about the evolving legal landscape and find practical steps for corporations and individuals to protect their copyright interests in the age of rapid AI advancements.
According to a finding by PwC in its Global Artificial Intelligence Study, Artificial intelligence (AI) could contribute a staggering $15.7 trillion to the global economy by 2030. While AI has triggered many discussions, the confluence of AI and intellectual property rights (IPRs) is one of the most hotly debated topics today. On the one hand, AI can be a great enabler for IPR management and protection measures. On the other hand, the recent court cases emerging in the United States indicate that AI may be a threat to monetizable human creativity which is protected by IPR laws. In one such case, artists Kelly McKernan, Sarah Andersen, and Karla Ortiz sued AI companies, Midjourney, Stability AI, and DeviantArt for using their art to train the AI technologies of these companies. The case shot to such fame that in September 2023, Kelly McKernan was listed among the 100 most influential persons in AI in 2023 by the Times Magazine. In another case, American comedian and actress, Sarah Silverman sued OpenAI for using her published works for training AI models. More than 5,000 authors have signed a petition asking tech companies to get consent from and give credit and compensation to writers whose books were used in training data. It was submitted that mass scraping of art, writing and movies from the internet for AI training was never considered or consented to. And a few days before this article was going to print, the New York Times has sued OpenAI and Microsoft over the use of its copyrighted works in millions of articles for training chatbots that, according to the New York Times, are competing with it as a source of reliable information. These are just a few examples, and from the mounting numbers of such cases, it appears that AI is going to keep copyright lawyers busy in the coming years.
As is evident, there is widespread apprehension among artists, authors, and content creators about their future in the wake of generative AI (Gen AI) tools like ChatGPT. While the United States is witnessing an increasing number of law suits on copyright violation, European Union’s legislation to curb these conflicts is nearly out of the anvil. There are similar developments in other parts of the world as well where courts, policy makers and legislators are grappling with various issues such as human contribution versus AI contribution, copyright infringement of the works used as training material for input by the AI tool to get the desired output, issues of copyright ownership in prompts, the argument of fair use raised by AI companies, whether AI can own copyright etc.
This article seeks to examine the issues raised by the confluence of AI and copyright and the relevant legal developments around the world in that context. In the end, the authors attempt to list out certain interim practical measures that may be adopted by corporations and individuals to safeguard themselves from copyright violations till such time the law catches up with the dizzying pace of technological developments in this area.
An overview of the global developments
Based on the degree of human involvement, WIPO in its session on “WIPO Conversation on Intellectual Property and Artificial Intelligence”, categorized works created using AI as “AI-generated works” (where the output is generated without human intervention) and “AI-assisted works” (where the output is generated with material human intervention and/ or direction). Since copyright law globally recognizes authorship and protection for works conceived and created by humans, the advent of AI-generated and AI-assisted works has created a new set of challenges and uncertainties in this area.
So who is the author of works created by AI whether it is “AI-generated” or “AI-assisted”? The answer to this question varies in each jurisdiction. Below is a discussion of how this question may be answered in some parts of the world including some of the developments that have been taking place in these countries.
The United Kingdom
Section 9(3) of the UK Copyright, Designs and Patents Act, 1988 states that, the author, in the case of a literary, dramatic, musical, or artistic work, which is computer-generated, shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken. Further, Section 178 thereof states that “computer-generated”, in relation to a work, means that the work is generated by a computer in circumstances such that there is no human author of the work. There are two issues that are debated in respect of the UK law in this context. First is that AI-generated works is a recent phenomenon and hence the law did not contemplate it in the expression “computer-generated”. Secondly, the UK law also contradictorily insists on originality in literary, dramatic, musical, or artistic works, which is argued to be absent in “computer-generated” works.
Reports indicate that the Intellectual Property Office of the UK (UKIPO) is reviewing the current law to position itself as a global leader in AI. The UK government is also in the process of creating a code of practice on copyright and AI. The code aims to make licences for data mining more available. It is also hoped that the code will help to overcome barriers that AI firms and users currently face and ensure there are protections for rights holders. The code reflects the intent of the UK copyright framework to promote and reward investment in creativity.
India
India, an erstwhile colony of Britain, has its copyright statute, the Indian Copyright Act 1957, largely based on the UK statute. Section 2 (d)(vi) of the Indian statute defines an author “in relation to any literary, dramatic, musical or artistic work which is computer-generated” to be the person who causes such work to be created. The Indian Copyright Office’s Practice and Procedure Manual (2018) states that only the details of natural persons be provided as author of the work in any copyright application. Based on section 2(d)(vi) it can be argued that, in those cases where AI output is with significant human input and involves skill and creativity, such output may be eligible for copyright protection. However, when it comes to AI-generated outputs with no human intervention, the Indian legislative framework for copyright ownership is lacking.
In 2020, the Indian Copyright Office, for the first time recognized an AI tool, “Raghav”, as a joint author of an artwork produced by Raghav along with its human developer and granted a registration. However, subsequently the Copyright Office issued a notice withdrawing the grant of registration, stating that the onus was on the co-author and human developer to inform the Copyright Office about the legal status of Raghav. This move appears to arise out of the subsequent realisation of the Copyright Office that there is no provision for non-human authors to own copyright under the Indian statute.
The United States
The United States maintains the stance that copyright can be asserted only when a work is original in nature and that for a work to be qualified as original it requires independent creation plus a modicum of creativity and should be the fruits of intellectual labor which are founded in the creative powers of the mind. Also, Chapter 300 of the Compendium of U.S. Copyright Office Practices titled, “Copyrightable Authorship: What Can Be Registered”, clearly states under clause 306 dealing with ‘The Human Authorship Requirement’ that the U.S. Copyright Office will register an original work of authorship, provided that the work was created by a human being. Further, clause 313.2 that deals with ‘Works That Lack Human Authorship’ states that, to qualify as a work of “authorship” a work must be created by a human being.
In February 2023, the US Copyright Office decided one of the first cases on Gen AI involving the work “Zarya of the Dawn” a comic book by Kristina Kashtanova. The graphics of the said comic book were generated by Midjourney, an AI image generator. The Office held that the graphics were not eligible to be safeguarded but considered Kashtanova as the author of the work’s text as well as the selection, coordination, and arrangement of the work’s written and visual elements. The Office reaffirmed that human creativity is an essential requirement for copyright protection and that the most prominent AI systems are unlikely to produce copyrightable work. According to the Office, users do not have complete creative authority over how the Gen AI technologies comprehend prompts and develop work; what would be genuinely important is how much creative control the human had over the work’s expression and how well the conventional aspects of authorship were developed.
In June 2023, the US government announced a set of voluntary AI guidelines that several prominent companies agreed to follow. The commitments in these guidelines, namely, safety, security, and trust mark a critical step towards developing responsible AI. On October 30, 2023, the US president Biden signed an Executive Order which sets out several guiding principles and priorities, including standards to promote AI safety and security, consumer and worker protection, data privacy, equity and civil rights, innovation, competition, and responsible government use of AI.
The European Union
In April 2021, the European Commission proposed the first EU regulatory framework for AI, namely, Artificial Intelligence Act (AI Act), that can be analysed and classified according to the risk they pose to users. In December 2023, the European Parliament (representing the citizens) and the European Council (representing the Governments) reached a political agreement on the AI Act.
The objectives of the AI Act are reportedly to create a regulatory framework for AI technologies, mitigate risks associated with AI systems, and establish clear guidelines for developers, users, and regulators. It is intended to aim to ensure the responsible use of AI by protecting people’s safety, security and fundamental rights and promoting transparency in AI applications. The AI Act, it is said, classifies AI systems as high-risk and prohibited AI practices to ensure that AI systems that are not likely to cause serious fundamental rights violations or other significant risks are not captured. AI systems presenting only limited risk would be subject to very light transparency obligations, for example, disclosing that the content was AI-generated so users can make informed decisions on further use. Subject to a set of requirements and obligations to gain access to the EU market, the Act is reported to authorise a wide range of high-risk AI systems. These requirements have been clarified and adjusted to be technically feasible and less burdensome for stakeholders to comply with. Systems used exclusively for military or defence purposes, research and innovation, and persons using AI for non-professional reasons are exempt from the applicability of the AI Act. A full text of the new Act was still unavailable for further analysis when this article was going to print.
That being so, the legal protection of works produced with AI technology is governed by EU copyright laws. A work must be “original” and clearly and objectively distinguishable to be eligible for copyright protection under EU law. In its Final Report on Trends and Developments in Artificial Intelligence: Challenges to the Intellectual Property Rights Framework, the European Commission identifies four interrelated criteria to be met for an AI-assisted output to qualify as a protected work. Under these criteria, output must be:
- A production in the literary, scientific and or artistic domain;
- the product of human intellectual effort;
- the result of creative choices; and
- is expressed in the output.
The report states that since most AI artefacts belong to the “literary, scientific or artistic domain, and are the result of at least some “human intellectual effort” (however remote), in practice the focus of the copyright analysis is on the third and the fourth criteria. That, whether the AI-assisted output is the result of human creative choices that are “expressed” in the output.
It is interesting that several Copyright Offices across the globe recognizes human author and human creativity to bestow copyright registration for a work and hence AI generated work has still not been provided copyright protection.
Japan
A different stand was taken by the Japan government in May 2023 that it would not enforce copyrights on information used in AI training. Its policy generally permitted copyrighted material to be used for training AI models. However, in June 2023, in response to public feedback, the Government decided to address a clearer definition of copyright infringement associated with Gen AI technology to safeguard and protect the rights of creators. A recent report in December 2023 states that, in a draft report compiling the views on Gen AI and copyrighted material, a panel under Japan’s Agency for Cultural Affairs has indicated that unauthorized use of protected texts, images and other material to train artificial intelligence may qualify as a copyright violation. It is reported that the panel plans to wrap up discussions by March 2024 towards updating related guidelines.
Argument of fair use as a defence by AI companies
Most of the foundation models are trained on billions of data sourced from the internet like news articles, books, images, videos and blog posts scraped from the internet. A significant portion of this data is likely protected by copyright. However, most providers do not clarify the copyright status of the training data. The use of large amounts of copyrighted data is raising serious issues everywhere. Some of these instances are already discussed at the beginning of this article. Interestingly, AI companies are arguing that the use of copyrighted works to train AI falls under fair use, a concept in copyright law that creates an exception if the material is changed in a “transformative” way. Fair use is the right to use copyrighted work under certain conditions without permission of the copyright owner and is an affirmative defense to a claim of copyright infringement, meaning that the alleged infringer has the burden of proving that their use was a fair use.
Fair use is used as a strong defense by AI companies on the basis that most outputs from AI models do not and may not explicitly resemble the work of any specific human being and the output may be argued to be ‘transformative work’. The test to see if a work is transformative is to see if it is different in character, serves a different purpose than the prior work, and is not a mere substitute for superficial changes made in the work. Hence developers could argue that the outputs of responses to prompts are transformative in nature.
It will be interesting to see how the question of fair use will play out in the courts in the times to come as all information that is being taken to train the AI models from the internet without the permission of the copyright owner were never intended to be utilized by AI models. Artists and other creative professionals who can establish that their copyrighted works were used to train the AI models could have an argument against the entities using these.
Getty Images sued Stability AI in Getty Images (US), Inc. v. Stability AI, Inc., alleging that the latter illegally used millions of the former’s photos to train its image-generating bot as the watermark of Getty images was reflected in the output work. Though no order has been passed in the case yet, Stability AI CEO has been quoted as saying that he believes that Gen AI “transforms” the work product and hence is protected by fair use.
There are also discussions of four broad schools of thought regarding the scope of the fair use defence that consider whether outputs generated from the input used by Gen AI tools infringe copyright in the inputs. The four broad schools are fair use minimalism, fair use maximalism, conditional fair use maximalism and limited fair use maximalism. Fair use minimalists consider all output generated through Gen AI tools to be unoriginal and therefore derivative works of the inputs, even if the output work is not an exact copy of any single input work. Fair use maximalists argue that the fair use defence should cover output works because each output work is unique and is created by a sufficiently transformative process. The third school that advocates conditional fair use maximalism takes a middle path between the first two schools above. They argue that fair use protection should be extended only to those works generated from sufficiently diverse input works that do not copy the “heart” of the expression contained in the input works. The fourth school of thought, namely, limited fair use maximalism, requires Gen AI companies to comply with license restrictions of applicable works protected by copyright before using them as input works.
The jurisprudence and policies in this space are yet to evolve and it would be interesting to watch the related developments that are emerging globally.
Some general measures to ward off copyright liability and ensure copyright protection
Clearly, the world has just begun to grapple with the effects of AI on intellectual property rights. Courts and law makers around the world are in the process of churning the right approach and policies to accommodate these developments. But it would be some time before the dust settles on the legal side of things because historically technology has always been ahead of the law. And in this instance, the pace of technology seems a bit too fast, and law has clearly been caught off guard around the globe. However, corporations and individuals that are at the receiving end of these developments can take certain measures to mitigate the effects of the fast-paced developments in the AI field while the law and policies are still getting shaped on the anvil globally.
How to mitigate corporate liability
As far as corporations are concerned, the issue could be manyfold.
The employees of a corporation may use Gen AI tools for performing client services though the employer-corporation has not authorised it. The easy availability of Gen AI tools in the hands of employees who are ignorant of copyright laws must concern corporate entities, especially for corporates in the areas of entertainment, research and innovation, publishing, software, and other technical fields. When employees use the relatively new Gen AI tools unbeknownst to the employers to create work products to be delivered to the customers of the employers, it could put the employers at the receiving end of third-party IPR liability. Another situation is that the corporation may be having its own copyrighted content on its website. Despite a copyright notice, such content may be used by the Gen AI companies for training purposes. Then it is possible that a company could be at the risk of receiving products from their vendors which are created using Gen AI. Employers must be watchful of IPR liability arising from service and work products offered to them by their vendors as these could unwittingly land the employers as defendants in IPR lawsuits. Yet another issue could be arising from a corporation’s plans to use the content available with them for training of AI tools without verifying the copyright ownership of the works – this could be true for large consulting companies who may wish to use their existing materials for training of Gen AI tools. Such existing materials may also include work products delivered to clients, the copyrights of which would belong to the clients. Unless there are mechanisms to identify these, the corporation could be potentially infringing copyrights.
The following are some of the preventive measures that could be taken by corporations to mitigate the risk of copyright infringement in such situations:
- Incorporate provisions in employment contracts which prohibit employees from using Gen AI tools for creation of work products. This prohibition can be a general one as well because in some countries (for example, India) the copyright ownership in any work created in the course of employment vests with the employer. In addition to such clauses on prohibition, the employer could consider obtaining representations and warranties coupled with indemnification provisions from employees to cover such situations.
- Incorporate a provision in the terms of use of the website of the corporation using specific language that the content on its website shall not be used for the purposes of training Gen AI tools.
- In service contracts with outside vendors, corporations must get a warranty from their vendors that they or their employees will not employ Gen AI tools for the purpose of delivering the work products. There must also be an indemnification provision coupled with such warranties to protect corporations from legal liability.
- If a corporation intends to use its existing content for training Gen AI tools, they must first conduct a due diligence of all content which they propose to use for such training to ascertain their copyright title.
- Yet another point to bear in mind for companies creating AI tools is to consider protecting their uniquely original “prompts”. “Prompt engineering” is the practice of designing inputs for Gen AI tools that will produce optimal outputs. Recently, the word ‘prompt’ has been in the news for making it to the Oxford Dictionary as one of the words for 2023. The Dictionary considered ‘prompt’ for its relevance to AI, and defined it as “an instruction given to an artificial intelligence program, algorithm, etc., which determines or influences the content it generates.” While the question of copyrightability of prompts is currently hotly debated, there is no harm if copyright protection is employed on uniquely original prompts. If the jurisprudence that is evolving now favours copyright protection for original prompts in the future, then such a protection would come in handy.
How to mitigate individual liability
Artists, writers, composers, or those involved in any other kind of creative arts express their work in physical form. Besides communicating these works in the physical realm, they often exhibit their works to the world through social media platforms which heavily permeate our world today. Unlike corporations, individuals are often less informed about the means and ways to protect their copyrights online or offline. As such, they are seldom diligent to employ copyright protection methods on their works before exhibiting them to the world at large, online, or offline. This puts them in a highly disadvantageous situation vis-à-vis Gen AI companies who are looking for works to train AI tools. The fact that legal solutions currently existing are inadequate further exacerbates this situation.
One simple measure they could adopt is to indicate on the work that the copyright therein is owned by them and that all rights are reserved including the specific rights to use the work for training AI tools. Additionally, if the national laws require registration of copyright in the work, they may register the work for protection with the relevant office. This could help in blunting any defence by Gen AI companies that the work is not copyright protected in that specific jurisdiction.
Conclusion
AI and technological developments around the same cannot be wished away. The implications of AI for IPRs in general and copyrights in particular are manifest through the judicial and legislative developments emerging around the world almost daily. In this evolving and still fluid landscape, the rights of stakeholders of copyright need some interim protection, which currently appears to be in their hands alone. Taking preventive measures like the ones listed above should mitigate the risks of copyright protected works being accessed by Gen AI companies and entities and individuals incurring copyright infringement liability through their day-to-day business operations.