John B. Quinn is the founder of Quinn Emmanuel Urquhart & Sullivan LLP, the world’s largest law firm dedicated exclusively to corporate litigation.
Would Michelangelo still have been this great if he had been born with a different set of tools at his disposal? How do you separate the art from the artist? What are the limits of the craftsman and to what extent do those tools play a role in the final output?
These questions are not new. But at no point in history have they been more relevant with the recent spate of generative AI products, from the image-generating DALL-E 2 to text-creating ChatGPT, spawning a wave of new legal challenges. The following provides a brief overview of the key intellectual property implications it brings to companies judging what is original and what can be construed as “inspired.”
What is Generative AI?
Generative AI is a subset of artificial intelligence and machine learning where a computer program learns complex algorithms to create new works, such as images, text, and music. Two commonly used techniques are generative hostile networks (GANs) And automatic encoders. GANs use two AI programs; the first generates new work based on a training set and the second checks whether the generated work is newly created or part of the set until the first can be reliably used as a standalone generative AI.
Autoencoders use two AI programs: an encoder, which reduces the work to a small representation called “latent vectors”, and a decoder, which then tries to expand the latent vectors back into the original work; the difference between the two is used to train the system to generate new material and change input material.
The issue of copyright
Generative AI developers often use copyrighted material to train their tools without obtaining a license. However, pending lawsuits is now investigating whether such use violates the copyright owner’s rights and whether they have the legal right to use existing works to train their algorithms, raising concerns about fair and transformative use.
The Legalese of Fair Use
There are certain exceptions within copyright law that allow for creativity and innovation. In the US, the fair use doctrine allows the use of copyrighted work for purposes such as criticism, education, research, and news reporting.
One of the key components of fair use is transformative use, which tests whether you can use a copyrighted work in a new and innovative way that was not originally intended, such as training an AI to create art.
In determining whether a particular use of a copyrighted work is “fair,” one of the central pillars upon which the entire argument rests is transformative useassessing whether the original work has been modified enough to become new and different and is no longer considered an infringement.
Because it is not an accurate calculation, fair use can only be claimed as a defense when accused of infringement. For example, when Google was sued for scanning books and display fragments of texts in search results, their court victory set a precedent for future cases. I predict many generative AI companies will use this precedent to argue that the courts have ruled that including content or code without permission to create new tools is considered a transformative use.
AI developers can also argue that using someone else’s work as an internal database for training their generative AI tools is a transformative goal, recognized by the Supreme Court ruling where copy parts of the Java API creating a new platform for programmers was seen as transformative.
Recently, the US Copyright Office struggled with this issuewho initially Kristina Kashtanova’s “AI-assisted graphic novel” Zarya of the dawn, as registerable but later reversible. Ms. Kashtanova contested the revocation attempt, making generative AI tools analogous to using a camera to create photographic images (accepted as copyrighted since the 19th century).
Implications for business leaders
The emergence of generative AI systems raises several concerns, many of which have not yet been extensively explored. Due to the lack of established legal and case law guidance, it is best to approach these developments with caution and draw on past experience.
For companies looking to explore the potential of integrating generative AI into their core business, one way to experiment with it is on the hackathon level. This will allow you to become somewhat familiar with the technology and its potential benefits. As generative AI becomes more accessible, you can integrate it into your core business.
Another option is to buy generative AI capabilities from startups that specialize in purpose-built tools. This focus on specific-use AI can be a cost-effective way for you to reap the benefits of generative AI without investing in costly in-house development. As technology becomes more democratized, these options are likely to become more widely available and accessible to businesses of all sizes.
It is still too early to predict exactly where this will go and how it will ultimately affect business decisions. Given the prevailing uncertainty around generative AI systems, legal action is being taken against such systems, including the lawsuit against Microsoft, OpenAI and their Copilot code generating systemwill have a significant impact on developers’ approach, and I see the court ruling having a substantial impact on how these models are trained.
There’s no denying that generative AI has opened the floodgates, and we’re only just beginning to see how this new technology will unfold in the legal world. Business leaders will have to watch the courts closely to learn where the line between man and machine will be drawn.
If they ban the use of copyrighted material to train large-scale models, it could be a significant setback for developers. Conversely, if the courts allow the use of data, whether copyrighted or not, it will have significant consequences for those who own the images and other property used to train these systems. For now, I think it’s fair to require developers of generative AI systems to get permission before using data.