ChatGPT Poses Questions of Copyright Ownership, Infringement
It sounds both promising and chilling: artificial intelligence that can answer complex questions, write reports and essays, and even create art. Are we edging closer to a machine takeover of the world? Or is the hype greater than reality? And what are the legal ramifications of this technology?
These are just some of the questions raised by ChatGPT, a chatbot developed by San Francisco-based OpenAI. Introduced on November 30, 2022, ChatGPT had more than 1 million users within days. By January 2023, several major school districts had banned ChatGPT or were considering doing so. Their concerns about cheating and plagiarism were seemingly well-founded. In a survey of 1,000 college students conducted January 18-19, 2023, 30% of respondents said they had used the AI chatbot, with 60% of those using it on more than half of their assignments.
Other applications of chatbots have had mixed results. On February 6, 2023, Google parent company Alphabet announced a ChatGPT rival called Bard that would be integrated into the Google search engine. However, Bard gave an incorrect answer to a question in an ad, causing Alphabet stock to plunge 9% at a cost of $100 billion in market capitalization. Microsoft Bing’s ChatGPT-based chatbot seemed to have trouble controlling its “emotions,” getting testy with some users and remorseful with others.
The AI Copyright Conundrum
Developers of these chatbots said they expect problems in the early stages of the technology as they continue to improve the large language models used to train them to make decisions on new, unseen data. In a sense, the chatbot is creating something new. But can the work of ChatGPT be copyrighted?
The answer depends on the jurisdiction. The UK is one of a handful of countries that grants copyright protection to computer-generated works. Under Section 178 Copyright Designs and Patents Act 1988, work generated exclusively by computer with no human intervention receives protection for 50 years, 20 years less than works created by a human. The “author” of an AI-generated work is the “person by whom the arrangements necessary for the creation of the work are undertaken.”
The U.S. offers no such protection. The U.S. Copyright Office has made it clear that it will “register an original work of authorship, provided that the work was created by a human being.” In a February 14, 2022, decision, the Copyright Office elaborated on the legal precedent requiring that an author be human, and stated that AI-generated work cannot be “work for hire” because a machine cannot enter into a contract.
How ChatGPT Works
If ChatGPT isn’t creating copyrightable work, is it infringing on the copyrights of the work used to train it? Answering that question requires a high-level understanding of how ChatGPT works.
AI systems such as ChatGPT “learn” from very large data sets—the larger the better. Algorithms run iteratively, analyzing the data to identify patterns. After each round of processing, the system or its human trainer tests the results and tweaks the model based upon whether the answer was right or wrong. Because a computer can run through millions of tasks without a break, it can learn a great deal very quickly.
ChatGPT is built on the GPT-3 neural network model, which has 175 billion machine learning parameters. Because it is so large, it performs far better than earlier language models. But while it seems to generate text as well as a human, it’s really just calculating probabilities.
GPT-3 was pre-trained using vast amounts of text scraped from the Internet, from sites such as Wikipedia. As it writes, it’s constantly asking, “Which word should come next?” It chooses the next word based on the probabilities it has learned from analyzing the data—but not always the word that’s highest ranked. Choosing a less likely word makes the writing seem more natural and interesting. It also enables ChatGPT to produce different work from the same prompts and prevents it from repeating existing text verbatim.
Risk of Infringement
That doesn’t mean that AI systems are immune to copyright infringement. On January 13, 2023, several artists filed a class action lawsuit against three makers of AI tools that generate images “for direct and vicarious copyright infringement.” The lawsuit alleges that the companies downloaded and stored the artists’ work without permission, and used the images to “train” the systems to create derivative works “in the style” of the artists. The resulting AI-generated images are for sale on the Internet, depriving the artist of a commission or license fee.
The UK is considering a copyright exemption for text and data mining for any purpose, which would give AI developers immunity from such lawsuits as long as they accessed the material. The U.S. offers no such exemption. AI systems such as ChatGPT must use work in the public domain or claim the “fair use” defense. In order to determine whether the use of a copyrighted work for AI training constitutes fair use, a court must consider the purpose of the use, the nature of the copyrighted work, how much of the work is used, and the effect of the use on the value of the copyrighted work.
While ChatGPT’s use of text scraped from the Internet may constitute fair use, the issue is unsettled. The law evolves more slowly than technology. In the U.S., companies seeking to sell computer-generated work or products that incorporate it must consider whether that work falls into the public domain, and those using copyrighted materials to train AI systems may be liable for copyright infringement.
Find Out More About Current Issues in Law
Purdue Global Law School can help you stay up to date on the most current legal developments in California and the rest of the nation.
Purdue Global Law School offers an online Juris Doctor for those who wish to become an attorney licensed in California, as well as an online Executive Juris Doctor for those who wish to advance their legal education but do not intend to become a practicing attorney.
Single law courses are also available to help you explore a particular area of law without committing to a full degree program. Request more information today.