Do AI-generated images deserve legal protection? – Part I

Introduction

Creativity and innovation can be regarded as the powerhouse of intellectual property. Intellectual property is the value inherently embedded in creative and innovative works. In the same vein, intellectual property rights confer exclusive legal ownership of a creative and innovative work for a specified period. These rights, including copyright, patent, trademark, industrial designs, and geographical indication, protect creative and innovative works from exploitation by unauthorized persons.

The digital age can be considered the most fluid era of human civilization due to its constant innovative advancement. A leading innovative advancement in the digital age is artificial intelligence (AI). Artificial intelligence is a simulation of human intelligence. Simply put, it is the replication of human abilities in machines. As a disruptive technology, artificial intelligence impacts human activities in various walks of life, and the creative industry is no exception. AI is projected to increase productivity in the creative sector to 40 percent by 2035.

The remarkable progress of AI has resulted in AI-generated images. These are artificial images are generated by artificial neural networks or algorithms. According to a recent report, AI-generated images have infiltrated various sectors such that over 15 billion images are reported to have been generated by AI. Notwithstanding its profound impact in the creative industry, there are potential threats and challenges emanating from AI-generated images, such as determination of authorship, authenticity, misinformation, violation of privacy, and increased ease in intellectual property infringement. The relationship between AI-generated images and intellectual property is multifaceted.

It is therefore important to consider the impact of AI-generated images on intellectual property protection, challenges to intellectual property protection posed by AI-generated images and strategic measures for a sustainable relationship between intellectual property and AI-generated images.

Understanding the Concept of AI-generated Images

AI-generated images can be created from scratch using trained artificial intelligence neural networks. Image generation through AI relies on textual inputs. These textual inputs are made in natural languages, which utilize text-to-image algorithms to fuse concepts and attributes to generate artistic images.

To generate AI-generated images, the AI generators are trained using an extensively large dataset of images. Through the training process, the algorithm is familiarized with various characteristics of images. The AI generator then relies on these characteristics to generate new images with similar features. There are various models of image generators using AI. They include the following:

• Generative Adversarial Networks: This model utilizes duo neural networks to generate an image that is closely similar to images in the training dataset. The duo neural networks in the generative adversarial networks are the generator and the discriminator. The aim of this AI generator is to generate the most realistic images based on the original images in the training dataset. This method relies on an adversarial approach in which the generator is trained to produce sample images that are closely alike and nearly indistinguishable from the original images while the discriminator evaluates the image generated to distinguish between the fake sample and the original image. The generator relies on the evaluation of the discriminator to improve its image generation. It is through this method that deepfake images are generated.#
• Text Understanding Using Natural Language Processing: This model transforms textual inputs into numerical representation. These numerical representations are thereafter relied on to generate AI images. When textual prompts are imputed, there are translated into a machine-friendly language using a natural language processing model. These numerical representations serve as a navigational map depicting the potential intention of the textual input to generate images.
• Neural Style Transfer: Neutral style transfer generates images through the fusion of two different images, such that the concept of one image is infused into another image. This model relies on a content image and a style image. The content image often includes the subject of the image, while the style image includes features that are adopted into the content image to create a new image.
• Diffusion Models: This model generates images that are a replication of images in the training database through a reverse process.

The Legal Dilemma of AI-Generated Images

Intellectual property aims to safeguard creative and innovative works. However, the introduction of AI-generated images threatens to blur the line between human creativity and machine-generated works. Over time, this has resulted in legal disputes arising from copyright infringement and the determination of authorship in AI-generated images. An example of such a legal dispute is the recent case of Li v. Liu in which the Beijing Internet Court ruled that an AI-generated image qualified for copyright protection. In this cited case, the Court recognized input prompts and parameter modifications as intellectual investment, thereby satisfying the requirement of originality for copyright protection.# This position is however in conflict with the position of the US court in Thaler v. Perlmutter in which the court ruled that human authorship is a requirement for copyright protection.# Copyright protection of AI-generated images differs according to jurisdictional rulings. In certain jurisdictions, AI-generated images have been ruled as ineligible for copyright protection, therefore relegating them to the public domain, while certain other jurisdiction recognizes copyright protection for AI-generated images.

Additionally, AI-generated images raise legal concerns about copyright infringement during the training process of AI models. AI-generated images rely on massive data including original images, texts and other creative works, some of which are legally protected by copyright. According to a study carried out in August 2023, many generative AI models such as BloombergGPT, are partly trained through copyright infringement of well-known authors. The indiscriminate use of copyrighted materials in training generative AI models has resulted in various legal actions by aggrieved parties.#

Conclusion

As artificial intelligence continues to reshape the creative landscape, the legal complexities surrounding AI-generated images remain a subject of intense debate. While some jurisdictions recognize AI-generated works as eligible for copyright protection, others maintain that human authorship is a fundamental requirement. This inconsistency raises crucial questions about ownership, originality, and the ethical implications of AI in creative industries. It is therefore expedient that AI is reconciled with intellectual property protection through legal and non-legal techniques. While AI-generated images pose potential risks to intellectual property, the implementation of properly structured legislations, industry initiatives and collaborations, technical solutions and ethical considerations would foster a sustainable environment for creativity and innovation.