With every new iteration of technology, it becomes increasingly easier to not only generate new creative works, but also to copy people’s creative outputs. The computer, the Internet, and mobile devices have all greatly increased not only the creation of new content and ideas, but also their more rapid spread. As the barriers to idea sharing and creation come down, so too do the barriers for the protection and even fundamental concepts of intellectual property.
Generative AI Output IP Concerns: It’s all about Training Data
The biggest IP concerns with generative AI are focused on GenAI systems generating outputs that seem too close to the training data that they have been trained on. GenAI systems aren’t really generating content, images, music, and other outputs from scratch as much as they are re-generating various recombined versions of the data on which they have been trained. In effect, GenAI systems are collaging inputs from multiple sources together to create a new output that’s mostly an amalgamation of those inputs.
The problem is many of those inputs come from sources that are protected by various IP rights. Recent lawsuits from IP owners have put a highlight on the fact that AI systems rely on vast datasets to learn patterns from the data in order to generate new content. However, these datasets often include copyrighted, trademarked, and otherwise protected materials, many of which have been used without explicit consent of their IP owners. This is raising concerns about the legality of using such data without explicit permissions.
The process of indiscriminate scraping the web for images, text, and other content to train AI models often incorporates copyrighted works, leading to potential IP infringements. Not surprisingly, many of the largest foundation models and large language models (LLMs) do not disclose the source of their training data for these reasons.
The main challenge with this regenerative, collaging style use of data is that the outputs can often be very similar to those training data inputs. Sometimes, these GenAI outputs even include watermarks or other telltale signs of the origins of the training data. And clever use of prompting can even produce the exact training data inputs as output. When these sorts of outputs happen that closely mimic or incorporate existing copyrighted works, what might otherwise be considered original creative works blur the lines between original creation and reproduction.
However, more often are GenAI outputs that mimic the style or specific elements of protected works. In non-AI situations, simply mimicking a style or using other works as inspiration or influence isn’t’ something that infringes on the works of others. Indeed, most works of art, music, and writing are often inspired by or mimics other works. However, when those outputs come from a machine that is purposefully built to generate new outputs based on patterns from specific inputs, there’s no surprise that legal disputes arise over ownership and the right to distribute such creations. As AI capabilities grow, so does the need for a nuanced approach to IP laws that can address these emerging complexities.
Are GenAI Outputs Protectable?
People are born with innate creative abilities, but it often takes skill and talent to translate that creativity into a real-world output. You have to learn how to write, paint, play an instrument, or otherwise translate your idea into a real-world output. However, AI is now dramatically dropping the barriers to going from idea to output, bypassing the need for skill, training, or talent.
Using AI, not only will artists and writers get their visions generated with the help of AI, but so too will anyone else be able to create those outputs with similar levels of quality. Anything that has a digital, and perhaps even physical, output will be augmented with the help of AI systems trained on all the previous creative outputs of humanity.
While getting AI to augment your skills is a boon to those who can accelerate and enhance their capabilities, this leads to as many problems as solutions it provides. In many ways, people might start to become skeptical about creativity itself. With AI blurring the lines between human ideation and produced output, it becomes harder to say what we should credit humans for, and what we need to credit the pre-existing trained data that was used to generate those outputs. Will people start to doubt human creativity? Will we seek some sort of “proof” of human creation or human-assisted work?
For this reason, the concept of intellectual property itself will become challenged. World intellectual property organizations are pushing back at the idea of providing intellectual property protection over AI generated works. Even if the human was involved in the ideation of the concept and curation of the outputs, IP rights organizations say that there’s not enough human involvement in the generated outputs. This poses a real problem for the industry.
GenAI changing the IP landscape
With today’s crop of AI generation tools, people are looking to gain some IP ownership of these produced works. If they are using AI tools to augment their creative output, they want to feel like they can have some ownership of the output, and not have those outputs unprotected. People want to copyright AI generated text, music, and art, trademark AI generated logos, and even patent AI-generated ideas. Right now, the world IP organizations are pushing back, demanding more human authorship and creation. But in the AI-enabled future, this might be unrealistic.
When AI is used every day, in every activity, as part of every generation step, separating the human from the machine will be a very difficult, if not impossible, task. In the past, copying works was a matter of cut and paste, but that was simply a form of plagiarism. With AI, no one will need to plagiarize, since it will be just easier to have the AI system generate something new from existing works. Will IP itself as a concept become a dinosaur of the past? The idea of the patent is only a few hundred years old, as are the concepts of trademark and copyright. Maybe the concept only mattered in an ecosystem where translating from idea to reality required skill and talent, but that may no longer be the case in a future rife with AI generated outputs.
Taking a More Sophisticated and Nuanced Approach to GenAI IP Issues
Clearly we’re at a crossroads when it comes to intellectual property and the answers aren’t cut and dry. Simply banning IP protection from AI-generated works might not be possible, and preventing AI systems from making use of pre-existing IP works might be a Pandora’s box we can’t close. We need to find new approaches that balance innovation with IP protection. This means a more sophisticated and nuanced approach to clarifying the legal status of data used in AI training and developing mechanisms to ensure that AI-generated outputs respect existing IP rights, while still providing for aspects of human creativity in curation and prompting. Clearly we’re in the early days of the continued evolution of what intellectual property means.
