As 2022 drew to a close, OpenAI released a public iteration of its web-based, machine-learning chatbot program, ChatGPT, and quickly garnered 100 million global sign ups. Designed to interact conversationally with users, ChatGPT takes written inputs and quickly produces an array of literary works, including essays, poems, plays, song lyrics, strings of computer code, and answers to complex questions. It has left users the world over both marvelling at its potential to help their lives, and recoiling in horror at what it may mean for their livelihoods. In this article, we look at some challenging IP questions raised by this exciting technology.
ChatGPT is a variant of the Generative Pretrained Transformer series of models, which use a deep neural network to generate human-like text. The program can be used for a variety of natural language processing tasks, such as answering questions, generating text, and translating languages. It can be integrated into chatbots and conversational interfaces, allowing people to interact with it in a conversational manner, hence the name “ChatGPT.”
ChatGPT is trained on a diverse range of internet text, including web pages, articles, and social media posts. It draws on machine learning algorithms and statistical models to generate responses based on patterns in the training content. It analyses user prompts/inputs and selects the response that has the highest probability of being the correct or relevant based on its available training content (which currently goes up to 2021). It provides the response that the model considers to be the best match for the input, considering context, grammar, and other language patterns.
ChatGPT is owned by OpenAI LLC (OpenAI). The corporation is also the creator of the text-to-image art generator program Dall-E 2, which has upset art communities globally, and which we wrote about in our recent article on who owns AI art under Australian law. OpenAI was founded by Elon Musk and Sam Altman, though Musk has since exited.
Microsoft has invested heavily in OpenAI. It was recently announced they would tip another US$10Bn into the corporation and its technologies. This latest round of investment brings Microsoft’s total equity investment since 2019 to US$13Bn. It is a major sign of its commitment to becoming the global leader in AI, as it attempts to out-manoeuvre major competitors, including Google and Apple, and regain its position in the search engine category. Ahemm, Bing, anyone?
Under basic principles of copyright law, originality is a key requirement for a work to be protected by copyright. A work must be original, originate from an author and not be a mere copy of another work. The work must not be entirely commonplace or lacking in originality, and must contain a sufficient degree of independent thought, skill, or effort by the author.
As we explored in our article on Dall-e 2, under current Australian legal principles a computer program such as ChatGPT cannot be the author of the work, as they must be a “qualified person”.[1] A “qualified person” is a citizen or resident person.[2] An “intelligent” technology platform ChatGPT cannot meet the legal definition of a person, or be recognised as an author or owner.
ChatGPT provides the following answer when asked about the intellectual property ownership of outputs from its program:
Machine learning outputs, such as text generated by a language model like myself, are not considered original works in the traditional sense. The text generated by a machine learning model is based on patterns and relationships learned from a large corpus of training data, rather than being created from the imagination or original thought of a human author.
As a result, the text generated by a machine learning model can be seen as a recombination or interpretation of existing information, rather than a truly original work. While the text generated by a machine learning model may be new and unique in a technical sense, it is not considered to be an original work in the same way that a work created by a human author would be.
Under the current Terms of Use for OpenAI products (including ChatGPT and Dall-e 2), the user interacting with the program owns both their inputs into, and their outputs from, the program. Clause 3(a) of the Terms of Use state:
As between the parties and to the extent permitted by applicable law, you own all Input, and subject to your compliance with these Terms, OpenAI hereby assigns to you all its right, title and interest in and to Output.[3]
Until late October 2022, this was not the case. OpenAI claimed ownership of copyright in its outputs (previously called ‘generations’). We will explore this backflip on IP ownership in more detail in a future article.
With respect to the ownership of the outputs, OpenAI offers this caveat in the Terms of Use to deal with what must be millions of similarly generic requests:
Due to the nature of machine learning, Output may not be unique across users and the Services may generate the same or similar output for OpenAI or a third party. For example, you may provide input to a model such as “What color is the sky?” and receive output such as “The sky is blue.” Other users may also ask similar questions and receive the same response. Responses that are requested by and generated for other users are not considered your Content. (bold added)
The Terms of Use address the rights of underlying rights holders by acknowledging that the assignment of IP by the program is “as between the parties” and “to the extent permitted by applicable law”. In other words, there may be third parties with supervening rights to the inputs or outputs. For example, if the user enters copyrighted song lyrics into the program, OpenAI does not purport to assign the copyrighted lyrics to the user.
It is unclear to what extent the outputs spawned by the program are subject to third party rights, including copyright. Much like the Lensa AI image creation app that captivated social media late last year, ChatGPT utilises billions of pieces of written material mined from the internet. This has raised concerns about the intellectual property rights of those authors, as well as questions over the copyright of what ChatGPT produces. The outputs from the ChatGPT are not sourced or credited by OpenAI, nor have authors generally authorised their work to be utilised in the program.
The production of information by the bot is veiled in secrecy, in a way that is very different to an internet search engine such as Google, which takes the user to the source of the information. With ChatGPT, in a deus ex machina like act, the answer is omnisciently delivered by the program without any kind of supporting information.[4]
If the outputs are created through a process of independent intellectual effort by either the user and/or OpenAI, then under current Australian legal principles a new copyright work can arise (Telstra Corporation Limited v Phone Directories Pty Ltd (2010) 264 ALR 617). In this case, the assignment of copyright by OpenAI in its terms would likely be valid. However, if the outputs produced by OpenAI lack the requisite element of independent intellectual effort, then no new copyright arises and the assignment of the copyright in the output by OpenAI is legally invalid.
From the user’s point of view, they currently have no practical way to know whether the outputs are owned by a third party, or indeed whether that third party is a powerfully resourced and litigious corporation. The Terms of Use make it clear that the use of the outputs commercially lies at the user’s own peril. All liability for infringement of third party rights is disclaimed by OpenAI in clause 7(b) of its terms of use.[5]
Copyright owners who believe their rights have been infringed are advised under the Terms of Use to contact OpenAI.[6]
Based on a document dated January 30, 2023, a Columbian Court is the first to openly admit that ChatGPT was used to help determine a case. Justice Juan Manuel Padilla Garcia of the First Circuit Court in Cartagena used the chatbot to provide legal answers about the case and included those answers in his decision.[7]
School students have been early adopters in using the program to help with homework and assignments. Many Australian schools in NSW, Queensland and Victoria have responded by banning the program’s use due to fears of plagiarism and destruction of the traditional education system. Yet South Australia’s Education Minister, Blair Boyer, has welcomed the technology, stating that AI is “going to have a big impact on all of our futures” and should be embraced as a tool in the classroom.
Whether we are excited or terrified by its potential, the extraordinary capabilities of programs like ChatGPT are here to stay. As increasing capital floods into machine learning technologies from the world’s richest companies, it is inevitable that they will play an increasingly mainstream and important role in society, and industries failing to adapt may do so at their peril of extinction.
The ramifications for IP law will be tremendous, and the future may head in a direction that lies beyond the present thinking of legal scholars and professionals. Who knows, if we interpret the recent Columbian Court judgment as the first sprinkles from the future, perhaps an AI program like ChatGPT will rewrite the law for all of us.
For advice on emerging laws surrounding AI, Web 3.0 and the digital landscape generally, please contact us below. For other articles in our METAVERSE+ series, click here.
[1] IceTV Pty Ltd v Nine Network Australia Pty Ltd (2009) 239 CLR 458; Copyright Act 1968 (Cth) s 32.
[2] Copyright Act 1968 (Cth) s 84.
[3]‘ Terms of Use’ OpenAI (Webpage, 2022) <https://openai.com/terms/>.
[4] Not only does this have the potential to infringe upon an author’s copyrighted work, but it could also cause the spreading of misleading or incorrect information when the sources of fact are not identified.
[5] ‘ Terms of Use’ OpenAI (Webpage, 2022) <https://openai.com/terms/>.
[6] Ibid.
[7] Janus Rose, ‘A Judge Just Used ChatGPT to Make a Court Decision’ Vice (Article, 4 February 2023) < https://www.vice.com/en/article/k7bdmv/judge-used-chatgpt-to-make-court-decision>.
