Intellectual Property Issues in Computer-Generated Art
With each technological advance, it becomes easier not only to produce new creative works, but also to copy the creative work of others. Computers, the Internet, and mobile devices have greatly facilitated not only the creation of new content and ideas, but also their dissemination. As the barriers to sharing and creating ideas become lower, so do the barriers to protecting intellectual property and basic concepts.
Concerns about generative AI output IP: It all depends on the training data
The biggest intellectual property concerns about generative AI center around GenAI systems producing outputs that appear too close to the training data they were trained on. GenAI systems do not generate content, images, music, or other output from scratch. Reproduce Different recombined versions of the data used for training. In effect, the GenAI system collages together inputs from multiple sources and creates a new output that is a near fusion of those inputs.
The problem is that many of these inputs come from sources protected by various intellectual property rights. Recent lawsuits by intellectual property rights holders have highlighted the fact that AI systems rely on huge data sets to learn patterns from the data and generate new content. However, these data sets often contain copyrighted or trademarked material, and much of it is used without the explicit consent of the intellectual property rights holders. This has raised concerns about the legality of using such data without explicit permission.
The process of indiscriminately scraping images, text, and other content from the web to train AI models often incorporates copyrighted works, leading to potential IP infringement. Not surprisingly, many of the largest foundational models and large-scale language models (LLMs) do not publicly disclose the sources of their training data for these reasons.
The main challenge with this regenerative collage style of data use is that the output can be very similar to the training data input. In some cases, these GenAI outputs may even contain watermarks or other features that indicate the origin of the training data. In other cases, with clever use of prompts, it may even be possible to generate an output that is identical to the training data input. When such an output closely imitates or incorporates an existing copyrighted work, it blurs the line between original creation and reproduction in what would otherwise be considered an original creation.
However, GenAI output often mimics the style or specific elements of protected works. In non-AI contexts, simply mimicking the style or using another work as inspiration or influence does not infringe on the work of others. In fact, most works of art, music, and writing are often inspired by or imitate other works. However, when those outputs come from machines purposefully built to generate new output based on patterns from specific inputs, it is no wonder that legal disputes arise over ownership and the right to distribute such works. As AI capabilities increase, so does the need for nuanced approaches to IP law that can address these new complexities.
Can the GenAI output be protected?
Humans are born with creativity, but translating that creativity into real-world artifacts often requires skill and talent. You have to learn to write, paint, play an instrument, or otherwise translate your ideas into real-world artifacts. But AI is now dramatically lowering the barrier from idea to artifact, bypassing the need for skill, training, and talent.
With AI, not only will artists and writers be able to generate their vision with the help of AI, but anyone else will be able to create deliverables of a similar quality level. Anything that has a digital output, or even a physical output, will be augmented with the help of an AI system trained on every creative achievement to date by humanity.
Augmenting skills with AI is a boon for those who can accelerate and enhance their capabilities, but it also brings as many problems as it does solutions. In many ways, people may start to doubt creativity itself. AI blurs the lines between human ideas and human artifacts, making it difficult to know what should be attributed to humans and what should be attributed to the existing trained data used to generate those artifacts. Will people start to doubt human creativity? Will they demand some kind of “proof” of human creations or human-assisted work?
This brings the very concept of intellectual property into question. Global intellectual property bodies are opposed to the idea of providing intellectual property protection to AI-generated works. Even if humans are involved in coming up with the concept or curating the deliverables, intellectual property bodies argue that there is insufficient human involvement in the deliverables generated. This is a major problem for the industry.
GenAI changes the IP environment
With today’s swarm of AI-generated tools, people are seeking to acquire intellectual property rights in these creative works. If you are using AI tools to enhance your creative work, you want to be able to obtain some ownership of the work and ensure that it is not left unprotected. People want to copyright AI-generated text, music, and art, trademark AI-generated logos, and patent AI-generated ideas. Currently, global intellectual property organizations are pushing back and demanding more human authorship and creation. But in an AI-enabled future, this may be unrealistic.
When AI is used every day, in every activity, as part of every generative step, separating humans from machines will be a very difficult, if not impossible, task. In the past, copying a work required cut and paste, which was simply plagiarism. With AI, no one will need to plagiarize anymore, as it will be easier to have an AI system generate something new from existing work. Will the concept of intellectual property itself become a relic of the past? The concept of patents, like the concepts of trademarks and copyrights, is only a few hundred years old. Perhaps this concept was only important in ecosystems where skill and talent were required to translate ideas into reality, but in a future filled with AI-generated output, this may no longer be the case.
GenAI: A more sophisticated and nuanced approach to IP issues
When it comes to intellectual property, clearly we are at a crossroads and the answers are not straightforward. It may not be possible to simply prohibit IP protection for AI-generated works, and preventing AI systems from exploiting existing IP works may be a Pandora’s box that cannot be closed. We need to find new approaches that balance innovation and IP protection. This means more sophisticated and nuanced approaches to clarify the legal status of data used in AI training and develop mechanisms to ensure that AI-generated output respects existing IP rights, while also providing an aspect of human creativity in curation and prompting. Clearly, we are in the early stages of a continuing evolution of the meaning of intellectual property.