Page 241 - Kaleidoscope Academic Conference Proceedings 2024

Innovation and Digital Transformation for a Sustainable World

content based on a simple prompt. So, you can provide a user prompts, adapting content and style for a human-like
word, a phrase, a question, or any other kind of starting quality.
point, and the AI tool will use its knowledge to generate
something new and original. These examples showcase the growing capabilities of AI in
various creative fields, blurring the lines between human
These works such as text, art, videos, computer code, etc and machine creation.
are all subject of copyright. The crux of the copyright issue
with AI-generated content lies in how these systems learn. 5. PROTECTION OF AI-GENERATED
Similar to other machine learning models, AI for creative CONTENT UNDER COPYRIGHT
tasks functions by uncovering and imitating patterns within
data. This means that to create an original sentence or Copyright law often relies on the Lockean labor theory,
image, the AI must first be trained on existing works by which states that people own the fruits of their labor. This
humans. For instance, if an AI art generator produces art theory suggests that ownership arises when someone mixes
reminiscent of Georgia O’Keefe’s style, it likely means it their labor with something from nature, creating something
was trained on a dataset that included her artwork. new.
Likewise, an AI content generator capable of writing in the
style of Toni Morrison would need to be trained on a large This theory might not apply directly to AI-generated works
corpus of her written works. This dependence on pre- because the Lockean theory emphasizes human labor. AI
existing creative works raises copyright concerns, as it blurs creations, however, result from programmed algorithms, not
the lines between inspiration and potential infringement. human emotions or feelings and because AI outputs are
There are many examples of works generated by AI. A few based on the data they’re trained on and the specific
of them are mentioned below:
prompts provided, they tend to be predictable and
repetitive. This raises questions about originality, a key
The “Next Rembrandt”: In 2016, a bank collaborated with requirement for copyright protection.
an ad agency to push the boundaries of technology and art.
They used deep learning and facial recognition to create a In essence, the Lockean theory might need reinterpretation
3D-printed painting in the style of Rembrandt – over 300 in the age of AI to determine if, and how, AI-generated
years after his death! The project involved a team of content can be copyrighted.
specialists who analyzed Rembrandt’s existing works to
capture details like brushstrokes and canvas textures.
Some people suggest that the data that is used to train the
AI is already copyrighted data of other people and therefore
AI-Composed Music: Sony’s research lab developed the result is an infringement of the copyrighted material.
software called Flow Machines that can create music by But, even if we take humans, for example, the data
analyzing vast libraries of songs. The software can generate collected by them compiling or arranging it in any original
new compositions by blending elements from various manner is a subject matter of copyright under the database
tracks. Users simply choose a desired style or artist, and the category and there is a possibility that it is a copyrighted
software creates a unique piece. A human musician then material of other people. Along similar lines, the output of
adds lyrics and refines the arrangement.
the AI is also an arrangement of the data fed into it.
Furthermore, humans also create work taking inspiration
AI News Stories: The Press Association (UK/Ireland) from others, similarly, we can say that AI has also taken
partnered with Google to experiment with AI-powered inspiration from the data made available to it and with the
news reporting. The project aimed to generate localized help of machine learning has created a new work.
news stories using AI.
There is one more popular theory in copyright- Idea-
AI Co-Authored Novel: In Japan, a short novel co-written Expression Dichotomy. Similar to human creativity,
with the help of an AI program made it through the initial copyright law for AI-generated content protects the unique
screening for a national literary award. The project involved expression, not the underlying idea. Two AI systems could
setting parameters and allowing the AI to “write” be trained on the same dataset and receive the same prompt,
autonomously. yet generate different creative outputs. Copyright
protection would apply to the original way each AI system
AI-Generated Music and Poetry: Google’s DeepMind lab expresses the idea, not the idea itself.
created WaveNet, a software that can compose music after
being trained on existing recordings. Additionally, Google Just as with human works, the core principle here is to
AI experimented with feeding large amounts of text data safeguard the free flow of ideas. AI development thrives on
(including romance novels) to an AI and prompting it to access to a vast pool of information and creative concepts.
write poems based on specific starting points. Restricting ideas through copyright could hinder the
advancement of AI technology. Copyright law should
OpenAI’s GPT-2 language model is another example. It can allow AI systems to be inspired by existing ideas and data,
generate realistic-looking news articles or stories based on enabling them to build upon that foundation and generate
new expressions. In essence, copyright protection should

– 197 –