Abstract
This paper dives deep into the connection between copyright laws and artificial intelligence (AI). There is an increasing need to comprehend how current copyright rules relate to these works and how they might need to be modified to address new difficulties resulting from the advancement of AI as Artificial Intelligence technology is used more and more in the production and distribution of creative works. The introduction of the paper covers the fundamentals of copyright law and how it relates to AI-generated works. The forthcoming sections examine a number of important topics, including the evolution of copyright laws in the global perspective, ownership, infringement, fair use, derivative and transformative work and the effect of AI on the established balance between creator and user rights. The report addresses various ongoing cases,policy implications and potential future paths in the final section.
Keywords
Copyright laws,AI generated images, Intellectual Property,Deep Learning, Neural Networks, Fair use, Derivative work , infringement.
Introduction
During the past few months, an aberration was in discern .The notion of creativity being solely a human capability began to dismember as the computer scientists reached a tipping point of innovation with the creation of generative artificial intelligence.
Millions of professionals who indefatigably perfected an artform throughout their lives are now faced with this gigantic conundrum of whether all their hard work has been for nothing.
With these tools ,now anyone with absolutely no artistic aptitude can easily claim to be the creator of spectacular and astounding works of art.
This gives rise to several conspicuous wringers which include how these machines are able to accomplish such arduous tasks,who owns these work, who has the right to use these works,etc.
Research Methodology
This paper is doctrinal in nature and the content is derived from secondary sources like news articles, academic papers, webpages,journals etc in order to analyze the current scenario surrounding copyright ownership of AI generated images.
Literature Review
This particular topic has also been focused on by a number of recent studies. One such text includes
Daniel J. Gervais, The Machine as Author, 105 Iowa Law Review. 2053 (2020) .
The article investigates whether or not computers qualify as writers of original works. The author contends that the current copyright laws fall short in their attempts to solve the problem and that a new legal framework is required to take into consideration the growing influence of artificial intelligence (AI) in the creative process.
Before exploring the current level of AI technology and its possible impact on copyright law, the article first offers an overview of the history of copyright law and its guiding principles. The author then takes into account a number of various definitions of authorship and assesses their applicability in the context of machine-generated works.
The study comes to the conclusion that copyright laws should be changed to reflect the fact that robots can be regarded authors.
Another study regarding the same includes VK Ahuja , Artificial intelligence and copyright : Issues and challenges, Winter issue,ILI law review,2020. The paper highlights the various challenges that arise when using AI in copyright-related activities, such as content creation, distribution, and enforcement.
The author argues that AI poses a significant challenge to the traditional copyright framework, as it raises questions about the ownership of works created by AI, the infringement of copyright by AI, and the liability of AI for copyright violations. The paper discusses the legal and policy implications of these issues and suggests possible solutions, such as the adoption of a sui generis legal framework for AI-generated works and the development of AI-specific copyright enforcement mechanisms.
The paper concludes by emphasizing the need for a balanced approach that takes into account the interests of all stakeholders, including creators, users, and developers of AI. The author argues that by addressing the challenges posed by AI, copyright law can evolve to better reflect the realities of the digital age and foster innovation in the field of AI.
The study emphasizes the different difficulties that can occur when AI is used in copyright-related tasks such as content production, distribution, and enforcement.
The author contends that AI significantly challenges the established copyright framework since it calls into question who owns works produced by AI, whether AI violates copyright, and if AI is responsible for such violations. The paper explores the ramifications of these challenges for law and policy and offers potential remedies.
Copyright: past and present
[1]Copyright is a legal concept that gives the creator of an original work exclusive rights to control how that work is used and distributed. The work can be a literary, artistic, musical, or other creative work that is fixed in a tangible form, such as a book, painting, song, or software code.
The purpose of copyright Is to incentivize creators to produce original works by giving them the exclusive right to control how their work is used and distributed. Copyright law grants several exclusive rights to the copyright owner, including the right to reproduce the work, distribute copies of the work, create derivative works based on the original, publicly display the work, and publicly perform the work.
Copyright protection typically lasts for a fixed period of time, which varies depending on the type of work and the country in which the work was created. [2]In India, for example, copyright protection generally lasts for the life of the creator plus 60 years.
Copyright law also includes several exceptions and limitations to the exclusive rights of copyright owners. These exceptions allow certain uses of copyrighted works without the permission of the copyright owner, such as “fair use” which is defined under section 52 of the Copyright Act of 1975.[3]”Fair use” includes activities such as criticism, review,reporting, research,etc. The purpose of these exceptions is to balance the interests of copyright owners with the interests of society in promoting creativity, education, and access to information.
The history of copyright dates back to the 18th century Britain when with the Statute of Anne, monopolies were granted to the publishers and printers of books .
[4]The Statute of Anne ,also called the Copyright Act of 1710 ,enacted by the British Parliament, was the first law passed to provide Copyright to original works regulated by the government. Previously, copyright regulations were carried out with the Authorization of the Licensing of the Press Act 1662 and the enforcement of the said act were enforced by the [5] Stationers’ company. The statue was christened after Queen Anne ,under whose regime the act was passed. This censorship resulted in incessant public protests. The Stationers’ monopoly was finally halted when the parliament refused to renew the Licensing Act in 1694.
One of the most integral personalities contributing to this eventualities were John Locke and John Milton. [6]This endeavour which started with the intention of protection of authors’ labour has influenced the copyright laws enacted for years to come thenceforth. The lockean theory of authors' ownership of their works is therefore a very rudimentary foundation of copyright ownership.
The primordial conception of the need to construct an international coordination for the protection of the writers’ and artist’s right was originated in association called the [7] Association Littéraire et Artistique Internationale (ALAI) in 1878,Paris with Victor Hugo as honorary President. This initiative was achieved eight years thereafter with the Berne Convention on the 9th of September 1886. [8]In order to handle the administrative tasks of the Berne Convention, a separate bureau was set up. After merging with the administrative handle of the Paris Convention for the protection of Industrial Property 1883,this bureaus in 1967 were rechristened the World Intellectual Property Organization (WIPO) ,which in the current times is a major global player in the protection of intellectual property. IN 1974 WIPO officially became an organization within the United Nations.
Development of Ai models
The key to untangle the Gordian knot of “who owns the copyright of AI generated images” lies in the way the AI models are trained.
[9] Around 2015 some researchers were able to develop a model by which computers were able to automatically generate captions from given images. An image caption generator as it was called could produce syntactically correct sentences to accurately depict the on goings of a given image in English. [10]This was achieved with the help of a Neural Image Caption (NIC) generator which is a deep learning model.
Inspired by this progress,few other researchers set on developing a model which could be given a description of a scene and the model could then generate an image based on the given prompt.
In the course of the next seven to eight years artificial intelligence has gone from creating distorted images with blobs of paint to effectuate stunning,awe-inspiring works of art.
[11]These impressive advancements ostensibly give rise to apparent questions as to how the models are able to perform such actions. The answer lies in the way they are trained in.
The AI models are trained on hundreds of millions of images found on the internet called datasets .The aforementioned text-to-image generator AI was trained on a dataset called Microsoft Common Objects in Context (COCO) . Other popular dataset includes [12] Large-scale Artificial Intelligence Open Network (LAION)which released its first dataset in August of 2021 with four hundred million image-caption pairs and subsequently released LAION-5B with over five billion pairs in March of 2022.
When the models are fed with a prompt , one might assume that the models would then search for a similar caption in their training dataset and create a new picture out of it.But that is often not the case . The output actually comes from [13] the ‘latent space’ of the deep learning model. A latent space can be described as a mathematical space with coordinates which helps the models accurately group similar objects together .
The computer looks for metrics in each picture which separates them from one another , eventually grouping them into a particular space in the multi dimensional coordinates.
[14] If we do a deeper analysis of this very topic we could tell that the models work in a very similar fashion to our brain and that process is called deep learning .
The main motive of these processes are to convert pixel values of images into a representation identifiable by the learning system.[15] Computers are able to process raw data with the help of Artificial Neural Network, a biologically inspired method in which the structures send signals to each other in the same way the neurons in the human body does.
[16]The process that generates these images is called diffusion models. These models , being generative in nature, distort the original training data with successive addition of Gaussian noise and then recover the data by reversing the noising process. Due to disruptions in the noise retrieval process, the images obtained are always different from any other one.Thus,the same prompt fed to the same model multiple times create different result each time and same prompt fed to different models tend to generate very different outputs due to the difference in the datasets that they were trained in.
Another type of model that has gained popularity in recent times is called [17] Generative Adversarial Networks (GAN) which differs from diffusion models in its usage of two neural networks -Generator and Discriminator- as opposed to the likelihood based models of diffusion.
Some other generative AI models that has gained immense popularity in the recent times include:
✓[18]Stable diffusion – a latent diffusion model created by CompVis group at LMU Munich , trained on LAION-5B dataset.
✓[19]DALL-E – This deep learning model is developed by a silicon valley based company called OpenAI with Sam Altman as CEO.
This model uses a version of ChatGPT-3 to create images.GPT or Generative Pre-trained Transformers , developed by OpenAI are a group of language models trained on huge sets of data in order to produce human like text.First revealed in a blogpost by OpenAI in 2021,the model was later modified into DALL-E 2 which is capable of generating images of Better quality and higher resolution.The company’s aim to amalgamate art and technology is facilely visible in it’s inspiration behind naming the model after the famous artist Salvador Dali and the robot WALL-E from the Pixar movie with the same name. [20]With Microsoft recently investing $10 Billion in OpenAI , it can be predicted that the market for Ai images is set to expand immensely in the coming days .
✓[21]Midjourney – Similar to OpenAI’s DALL.E , Midjourney is a research lab that has produced a text-to-image model where users can create high quality, stunning visual art within seconds through a Discord bot.
The conflict
With the advancement of every new technology, at least some degree of the public life fills up with panic as to how their lives might change.
These predicaments sometimes involve legit concerns and other times they don’t.
Now that creativity , previously thought of as exclusively human ,is being incorporated by machines in their plethora of skill sets , the people whose livelihoods are dependant on creativity are discerningly concerned.Not to mention the fact that these machines are able to perform such creative functions is particularly possible because they have been trained upon the works created by these artists.This gives rise to several legal ,ethical and moral question as to who owns these works – The machine? The person who entered the prompts ? Or the Artists who formed the base for these works to be created.
The first and foremost query that needs to be addressed in order to be able to see some light at the end of a long tunnel seems to be if “ machines can own a copyright” . In the current global scenario,the answer seems to be “ NO” .
In the case of [22] Thaler v. Vidal , Dr. Stephen Thaler created an Artificial Intelligence system called “ Device for the Autonomous Bootstrapping of Unified Science” or DABUS for short. [23] In July of 2019, Dr. Thaler filed for patent protection for two inventions invented by DABUS . His intentions were to claim ownership of the parents , simultaneously giving credits to DABUS as their inventor. The United States Patents and Trademark Office (USPTO) refused to grant the patent to Dr. Thaler on the grounds that, under the Patent Act , “a machine does not qualify as an inventor” . Henceforth,Thaler made an appeal to the director of the USPTO which was denied.
Then, he sought a judicial review of the decision of the USPTO in the United States District Court for the Eastern District of Virginia where his request to reinstate the application was denied as well. According to the court,the inventor must be an individual under the Patent Act .
Subsequently, Thaler approached the United States Court of Appeals for the Federal Circuit where the decision of the district court was affirmed. Thaler made the argument that in Section 271 of Title 35 of the United States Code the term”whoever” is used to refer to corporation and other non-human entities and the Section 101 of the same title states that “whoever invents or discovers any new and useful process , machine,manufacture,or composition of matter,or any new and useful improvement thereof,may obtain a patent therefor,subject to requirements and conditions of this Title.” Thus,the definition from section 271 must also apply to section 101 similarly.
The court stated that only a natural person can be an inventor,so AI cannot be.
Some other cases that deals with the ownership of copyrights by AI include:
✓[24]Univ. of Utah v. Max-Planck-Gesellschaft - which held that “inventors must be natural persons and can not be corporations or sovereigns”
✓[25]Rural Telephone Service v. Feist Publications -Rural telephone service was a telephone cooperative providing services in the US state of Kansas.Feist publications specialized in the compilation of telephone directories in Kansas.Fiest copied 4000 entries from Rural’s directory for which Rural sued First for copyright infringement. In a landmark decision,the United States Supreme Court ruled that information alone without minimum creativity cannot be protected by copyright
✓[26]Asia Pacific Publishing Pte Ltd v Pioneers & Leaders - upheld that copyright protection can only be granted to human authors.
✓[27]Naruto v. Slater, – This case is probably the most popular case involving the ownership of a work by a non-human author.
Naruto was a seven year old macaque living in the reserve island of Sulawesi, Indonesia. In 2011 A cameraman left his camera unattended in the forest and Naruto took several photographs of itself with the camera. Slater along with Wildlife Personalities ltd. then published these photographs in a book . In 2015 , People for the Ethical Treatment of Animals (PETA) Filed a lawsuit of copyright infringement by Slater and others claiming that the maqacue was the copyright owner .
The court upheld that all animals lacked statutory standing under the Copyright Act, 17 U.S. Code § 101 er seq. since they were not human .
Thus in order to claim Copyright of something,the author must be a natural person. Also there is a minimum threshold of creativity required in order to claim a copyright as discussed above in the case of Rural v. Fiest. A very similar situation was observed when cameras first became a thing. Though it is something that we take for granted today,that a photographer has rights to protect their work,it wasn’t an established fact until 1884 in the case [28] of Burrow-Giles Lithographic Company v. Sarony .
A lawsuit was filed against Burrow-Giles Lithographic Company by photographer Napoleon Sarony claiming that the company has marketed unauthorized lithographs of Sarony’s pictures of the famed writer Oscar Wilde called “ Oscar Wilde No.18” .The company argued that photographs couldn’t be considered as “writings” or works by an author and that the § 4952 of the Copyright Act of 1965 was unconstitutional.
Justice Miller while ruling stated that all writings,paintings,engravings,maps,charts,etc ,which required the ideas in the mind of the author to be visualized were to be provided with protection.
Thus, an AI image which requires the author to only type in a prompt and press a button may not be considered to be minimally creative so as to provide it with protections.
Ongoing cases
✓ [29]Getty Images (US) Inc v. Stability AI Inc, – On the 3rd of February 2023, stock photo provider Getty images sued artificial intelligence company Stability AI Inc in a Delaware federal court, claiming that the AI company indulged in brazen copyright infringement by using about 12 million images from their collection. In one instance infact an image of a football match generated by the AI model had a distorted logo of Getty Images on it.
Getty claims that Stability AI has copied their pictures and associated metadata in order to build a competing business through its interface called DreamStudio.
The lawyers representing Getty have requested for a jury trial and statutory damages reaching $ 150,000 for each infringed material. Earlier this year Getty filed a similar lawsuit against the company in the United Kingdom.
✓[30]Andersen et al v. Stability AI - In January of 2023, artists Sarah Andersen,Kelly mcKernan and Karla ortiz filed a suit against AI image generators Stability AI, Midjourney and DeviantArt in the U.S District Court of California,San Francisco division, claiming that the defendants have infringed their copyright by using images created by them and other thousands of artists like them in order to train their generators to produce [31] derivative works.
Derivative work is a creative work based on a pre-existing, copyrighted , original work that is fixed in a tangible medium.
For example, The movies and the translations of JRR Tolkien’s “The Hobbit” are derivative works of the original book.
Common derivative works include translation,fictionalization,sound recording,motion picture,Abridgement,Condensation or any other form in which the work can be transformed or adapted.
The right to create derivative work is known as the Adaptation Right which exclusively means that a derivative work can not be created without the proper authorization of the original work’s author.
On the other hand transformative works are those works which alter the original work with new angles to such an extent that it's familiarity with the original work becomes negligible. Transformative works, thus no longer qualify as infringement as it serves a completely new purpose and gives the work a totally different meaning. So, it is for the court to decide whether the AI image generators’ usage of the artist’s works constitute [32]“ fair use “ . Section 107 of the 1976 Copyright Act of the United States has 4 elements to determine whether fair use is an applicable defense in the given case which include :
A) Nature of the copyrighted work
B) Character of the copyright work
C) Amount of portion used in the transformed work
D)Effect on the copyrighted work
It is to be noted that Fair Use is an affirmative defense, meaning,it can only be argued by the defense and not the plaintiff to initiate a lawsuit.
A judgment in favour of the defendant will have far reaching consequences that will be carried forward into the future.
It would essentially mean that AI models will be able to produce works based on any kind of pre-existing work without the consent of the creators.The court might rule based upon some precedences involving the fair use defense which include:
✓ [33]Perfect 10, Inc. v. Amazon.com – It involved a claim of copyright infringement against Amazon and Google. The court held that hyperlinking and framing of original images in image searches did not constitute infringement,as the work was extremely transformative .
✓[34]Authors Guild v. Google, Inc. – Google was creating a massive library with digital copies of a plethora of books for public search but included only snippets of the books and not the whole book in itself . The plaintiff sued Google,Inc. for copyright infringement in the U.S District Court for the Southern District of New York. Google’s attorneys argued that their use constituted “fair use “ which is not a copyright infringement under 17 U.S.C. § 107. The court held that the defendant’s use of the plaintiff’s work was non-infringing fair use.
Conclusion
In conclusion, the field of intellectual property law governing copyright rules for AI-generated images is complicated and dynamic. As AI develops, serious issues about the ownership and protection of artificial intelligence-generated art are raised. There is no one-size-fits-all answer to these problems, which the existing legal system is still trying to resolve. But it is obvious that copyright laws need to change to accommodate the particular difficulties brought on by AI-generated images.
The best strategies to safeguard and encourage innovation in this field should continue to be discussed by legislators, legal scholars, and business leaders in the future.
References
[1]. Copyright
https://en.m.wikipedia.org/wiki/Copyright
[2]. Hand Book of Copyright Law https://copyright.gov.in/documents/handbook.html#:~:text=It%20is%20protected%20for%20a,the%20death%20of%20the%20author
[3]. Copyright Office: Exceptions To Infringement Under Copyright Act, 1957 https://copyright.gov.in/Exceptions.aspx
[4]. History of Copyright: Statute of Anne, 1710 https://www.copyrighthistory.com/anne.html
[5]. Worshipful Company of Stationers and Newspaper Makers https://en.m.wikipedia.org/wiki/Worshipful_Company_of_Stationers_and_Newspaper_Makers
[6]John Locke’s Labour Theory: A Justification of IPRs https://www.legalservicesindia.com/article/2536/John-Locke%E2%80%99s-Labour-Theory:-A-Justification-of-IPRs.html
[7] Association littéraire et artistique internationale
https://www.alai.org/en/information/history.html
[8]. Berne Convention
https://en.m.wikipedia.org/wiki/Berne_Convention
[9] (Kiros et al., 2014a; Karpathy & Li, 2015; Vinyals et al., 2015; Xu et al., 2015)
[10] Image Captioning Encoder–Decoder Models Using CNN-RNN Architectures: A Comparative Study
https://dlnext.acm.org/doi/abs/10.1007/s00034-022-02050-2
[11] arXiv:1511.02793v2
[12] LAION
https://en.m.wikipedia.org/wiki/LAION
[13] Latent Space in Deep Learning
https://www.baeldung.com/cs/dl-latent-space
[14] What Is Deep Learning?: How It Works, Techniques & Applications
https://www.mathworks.com/discovery/deep-learning.html
[15] What are Neural Networks?
https://www.ibm.com/in-en/topics/neural-networks
[16] Introduction to Diffusion Models for Machine Learning
https://www.assemblyai.com/blog/diffusion-models-for-machine-learning-introduction/
[17] Copyright Storm - Authorship in the age of AI
https://va2rosa.medium.com/copyright-storm-authorship-in-the-age-of-ai-baba554aa617
[18] Stable Diffusion Online
https://stablediffusionweb.com/
[19] DALL·E: Creating images from text
https://openai.com/research/dall-e
[20] Microsoft to Invest $10 Billion in ChatGPT Maker OpenAI (MSFT)
https://www.bloomberg.com/news/articles/2023-01-23/microsoft-makes-multibillion-dollar-investment-in-openai
[21]Midjourney https://www.midjourney.com/home/?callbackUrl=%2Fapp%2F
[22] Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022)
[23]Thaler v. Katherine K. Vidal, 43 F.4th 1207 (Fed. Cir. 2022).
http://www.ipbrief.net/2022/10/27/thaler-v-katherine-k-vidal-43-f-4th-1207-fed-cir-2022/
[24]Univ. of Utah v. Max-Planck-Gesellschaft, 734 F.3d 1315, 1323 (Fed. Cir. 2013)
[25]Rural Telephone Service v. Feist Publications, 957 F.2d 765 (10th Cir. 1992)
[26]Asia Pacific Publishing Pte Ltd v Pioneers & Leaders Pte Ltd, [2011] SGCA 37
[27]Naruto v. Slater, No. 16-15469 (9th Cir. 2018)
[28]Burrow-Giles Lithographic Company v. Sarony, 111 U.S. 53,4 S. Ct. 279 (1884)
[29]Getty Images (US) Inc v. Stability AI Inc, U.S District Court for the District of Delaware, No.1:23-cv-00135
[30]Andersen et al v. Stability AI Ltd.,3:23-cv-00201
[31]Copyright & Copywrong: What are Derivative and Transformative Works?
https://foundrylawgroup.com/copyright-copywrong-what-are-derivative-and-transformative-works/
[32]U.S. Copyright Office Fair Use Index
https://www.copyright.gov/fair-use/#:~:text=Fair%20use%20is%20a%20legal,protected%20works%20in%20certain%20circumstances
[33]Perfect 10, Inc. v. Amazon.com, Inc., 508 F.3d 1146
[34]Authors Guild v. Google, Inc. – 804 F.3d 202 (2d Cir. 2015)
Author - Suprana Chakraborty
Comments
Post a Comment