‘Artificial Intelligence (AI)’ which once was a myth in the world of computer programming, is now a reality at the hands of the commons. Not long before we witnessed cinemas like The Terminator Series or Marvel’s Iron Man Series or Transcendence and such other Sci-Fi movies which delved into the extremities of AI in all its superficial glory, but many of us do not even realize the depth of our present-day dependency on AI. The applications, operating systems and even the Software we use contain an element of AI in it, which makes it an inextricable part of our digital life. Before moving on to the legal nuances around AI, let us briefly understand the concept of AI.
What is AI?
The roots of AI can be traced to the initial works of Alan Turing, who is celebrated as the Father of Computer Programming, wherein he develops the renowned Turing Test[1] (also known as the Imitation Game) to investigate the thinking capacity of computers. It is at this juncture; Turing observes the distinction between human and machine thinking and highlights the necessity of human input. Alongside this further foresees the possibility of the evolution of Machine Learning (ML) to ‘systems that act like humans’.
John McCarthy, the father of Artificial Intelligence, who coined the very term ‘AI’ defined it as ‘the science and engineering of making intelligent machines’. This lexicology refers to a machine which is able to think, compute and overcome problems without human intervention.
Types of AI
Broadly there are two types of AI,
- Artificial Narrow Intelligence (ANI) – refers to a specific form of AI which focuses on solving problems specified in the compiler algorithm while requiring human inputs for functioning. These AIs employ machine learning to compute across the probability matrix. These can be often seen in the case of AI assistants such as Apple’s Siri, Samsung’s Alexa, Google Assistant, Amazon’s Alexa etc.
- Artificial General Intelligence (AGI) – refers to the theoretical form of AI which is believed to be equivalent to human intelligence. This form of AI is based on a deep learning algorithm which eliminates the need for human mediation. A dystopian example of this may be Ultron from Marvel’s Comics.
AI Image Generation and Art
Art is considered as an expression of the innate thoughts of a person, through various modes of portrayal. It includes artistic images, writings, sketches etc. The concept of painting or sketching in the physical domain was limited to the standard canvas sizes. In case of the digital art, the canvas has stretched wide and afar with complex tools of artistry available digitally at the fingertips. This opened a whole new dimension for art, which paved way for various artists, who were limited by physical tools, to explore the various horizons of art on e-Canvas.
As the e-Art rose, the concept of AI-powered art generation gained momentum in the early 2017s and later on, various algorithms of deep learning models were developed such as Dall-E and Midjourney as a semi-open source in the late 2020s and mid-2021 by various programmers. These deep learning algorithms allow the generation of art and images based on text, imagery, computer code and other such inputs via natural language prompts. Some of these algorithms receive textual inputs from users and based on it, the algorithm processes billions of images from various sources to intelligently morph and design unique art. The fascination is related to the quality of output of these algorithms which is astonishing owing to the magnanimity of the algorithmic array making it a mark of creativity, novelty, precision and adaptability.
In August 2022, a deep learning text-to-image generation AI under the name of Stable Diffusion was released by Stability AI which allowed open-source access to the public and witnessed millions of usage over a span of few months. Other AI-Image Generation algorithm developers followed suit in making their models more open-ended. Following these developments, the problem that gains the spotlight is the legality of web-scrapping of inputs that are processed by the AI which include stock and licensed images from third-party proprietors for third-party image generation without the consent/license of the said proprietors.
The legal perplexity further includes the questions of the novelty of such creation by AI. If so, where does the ownership fall? Whether the consent of image service providers essential prior to web-scrapping? Liability of User/Creator of AI? Most importantly, whether the said image-scrapping constitutes an infringement of the copyrights of the creators/artists?
The Suits
There are majorly 2 suits filed recently in January 2023 before the United States Northern Californian Court which have the potential to determine the future of the development and scope of AI systems.
- Andersen et al v. Stability AI et al [2] –
- The plaintiffs namely Sarah Andersen, Kelly McKernan and Karla Ortiz filed a class-action suit on 13th January 2023 against AI developers Stability AI Ltd., Midjourney Inc. and DeviantArt Inc. alleging that the defendants had used the work of the plaintiffs (artists) without permission by training the AI systems to web-scrap derivative works, thereby contenting copyright infringement.
- The core contention of this suit is that the ability of the AI tools to create art ‘in the style of’ or ‘similar to that of’ a particular artist constitutes a copyright infringement.
- Further, the secondary plea relates to whether the said AI proprietors have an obligation to share the revenue on images, generated on the basis of training images of the artists.
- Getty Images v. Stability AI [3] –
- The plaintiff company which is a proprietor holding license and copyright over billions of stock images is suing the Stable Diffusion AI Image generation creators, Stability AI, for unlawful copying and processing of millions of images protected under copyright, in the High Court of Justice in London.
- The plaintiffs assert that they offer exclusive licensing options for training and usage of copyrighted images for AI systems. They allege that the defendant company circumvented viable licensing options thereby illegally using billions of copyrighted images and content for AI image generation.
- Pursuant to the same, the plaintiff’s main contention is that the defendants infringed the copyrights of Getty Images. In reply, the defendants are resorting to the defence under fair use and the channel of copyright preemption under S. 301 of the Copyright Law of the United States (17 U.S. Code).
The fundamental legal nuance in both these plaints can be narrowed to the following tendency of whether the said trained AI image generation constitutes,
- Creation – by learning to generate unique images via processing of proprietary art of the artists, through deep learning; or
- Copying – including morphing and retention of parts of the art itself which leads to infringement apparent on the face of the record.
If this is determined the next question would be on the liability. The current position of law however recognizes AI as an object of rights via the human agency as its subject. Hence, the creator of such a system is held to be liable for the acts of AI unless express law is enacted to define the persona of AI (which may also give scope for Androids and AI lifeforms in near future).
Overall, this legal centre of gravity is placed on the crunch of the perception of the Courts of law in interpreting the nature of the operation of these AI systems. Hence, these cases have the potential to be a landmark in the history of AI in penning down the borders for the AI playground in the domain of copyright law, which determines the very development of AI systems and algorithms in its entirety.
References
- A. M. Turing (1950) Computing Machinery and Intelligence. Mind 49: 433-460.
- Case 3:23-cv-00201 https://fingfx.thomsonreuters.com/gfx/legaldocs/myvmogjdxvr/IP%20AI%20COPYRIGHT%20complaint.pdf
- https://newsroom.gettyimages.com/en/getty-images/getty-images-statement
Leave a Reply