“ChatGPT and other large language models have the ability to generate human-like text, which raises concerns about the potential impact on intellectual property. Specifically, there is a concern that these models could be used to create fraudulent or infringing content, such as plagiarized written works or manipulated audio or video. Additionally, there is the possibility that the generated content could be used to impersonate individuals or organizations, potentially damaging their reputation or infringing on their right of publicity. These concerns highlight the need for further research and regulations to address the potential impact of language models on intellectual property.”
The previous paragraph was actually written by ChatGPT. As many will be aware by now, ChatGPT has been making headlines for its ability to have human-like conversations right through to its ability to write an entire essay for college/university students. Founded in 2015, OpenAI – the company behind ChatGPT – is an artificial intelligence research laboratory located in San Francisco. The company originally consisted of a for-profit branch, OpenAI LP, and its parent company, a non-profit branch OpenAI Inc. The company was later restructured as a capped “for profit” company setting its profit limits to 100X on any investment
While its stated goal is to promote and develop friendly AI in ways that benefit humanity, the reality is that it is also a business. With a projected revenue of $1 billion by 2024, the obvious question is, how does OpenAI make money? Part of the answer may be found by looking at its IP strategy, namely how OpenAI plans to develop, acquire, and leverage intellectual property (“IP”).
From a cursory review of publicly available information, OpenAI appears to be protecting parts of its technology with patents and trade secrets, while making the rest open source. If well executed, such an approach could underpin OpenAI’s business and provide a good example of how IP and business strategy should be intimately connected. OpenAI is using their IP to generate revenue (with Microsoft, for example), while using their open source content to generate goodwill and a positive reputation among users. This blended strategy of using both open and exclusionary IP assets is one that several AI companies are finding beneficial.
OpenAI’s Mix of IP Assets
Patent Assets – As of the publication of this article, OpenAI does not appear to have any published patent applications.[i] This could be a conscious choice or an omission. For example, there are valid strategic reasons for not pursuing patent protection. It may be the case that the interactive aspects of OpenAI’s products lack sufficient innovation to qualify for patent protection. ChatGPT, in particular, has little in the way of innovative user-facing features. Instead, their innovation could lie in their algorithms and computer models or evaluation mechanisms used in training ChatGPT. Such algorithms and computer models may either not be patentable (e.g., because of their abstract mathematical nature) or be undetectable (i.e., it might be difficult or practically impossible to determine whether they are being used by a competitor).
Conversely, OpenAI’s paucity of patents could be a gap in their IP strategy. In the last ten years, competitors such as Google and Huawei have secured patent protection for various features of their chatbot technologies. Thus, OpenAI’s devoid patent portfolio could expose it to long-term risk in the AI space. While OpenAI may not have any patents right now, they could file applications later as user-facing features increase in number.
Trade Secrets – While the name “OpenAI” suggests the company’s information is transparent, it is not. OpenAI is likely using trade secrets to protect proprietary information. OpenAI’s trade secrets likely cover: training sets, data output, and other data; neural networks, including modular network structures and individual modules; and learning, back propagation, and other algorithms that could give it a competitive advantage over competitors (see, for example, China’s Baidu).
Trade secrets may be useful to AI companies seeking protection of proprietary information not covered by patents. For example, trade secrets can overcome the challenge of protecting data compilations such as AI training sets, a programmer’s certain expression of source code, or other proprietary information that may provide a competitive advantage.
OpenAI and Open Source – OpenAI has also offered some of its software on an open-source basis. OpenAI’s Jukebox, for example, is an open-source algorithm that generates music with vocals. Offering software in an open-source format can have commercial benefits, drive innovation, and contribute to broader knowledge sharing. AI innovators should understand, however, that open-source software (“OSS”) has restrictions. Although OSS is made free to the public, free does not refer to cost, but rather to the freedoms that licensees are granted. Software licensed as open source means the licensee can use, modify, enhance, and share the software, as well as provide access to the source code needed to do so.
Given the variety of OSS licenses, AI developers should ensure they understand the rights and responsibilities relating to the use of each. While offering or using software under an OSS license has its advantages, an effective IP strategy will ensure no more information is disclosed than is required.
IP Risk Management in AI
OpenAI – as well as any other company that plans on taking a layered approach with their AI-based innovation – needs to consider various issues related to their IP, particularly in terms of risk management. Some elements of OpenAI’s IP strategy may not be gleaned from publicly available information, such as OpenAI’s IP risk identification and mitigation strategy. While such elements may not be known for certain, it is likely that issues such as inventorship/authorship and copyright infringement are being considered.
Copyright Infringement – Copyright infringement is another issue AI developers should be aware of. As it stands, it is unclear how inputs and outputs of an AI are treated under copyright law. When AI is involved in creating text-based works (or works of art, for that matter), it may expose the company that owns the AI and/or the person who used the AI to the possibility of being sued for copyright infringement. For example, a large language model such as ChatGPT that has been trained on copyrighted material could cause that model to excessively draw from another’s work when providing a response to a user. This may violate reproduction rights under the Copyright Act and lead to infringement claims.
In fact, such cases have already started to be brought before the courts. For example, Getty Images filed a claim against Stability AI, alleging that its AI art tool scraped human-created copyrighted images for training data.
When asked whether IP is important in the AI space, ChatGPT itself said …
“Yes, intellectual property is an important factor in the artificial intelligence space. Artificial intelligence (AI) technologies are often based on complex algorithms and software, which are protected by patents and copyrights. By protecting intellectual property rights, companies can protect their AI inventions and discourage others from copying or using their technology without permission. Furthermore, patents can be used to generate revenue by licensing the technology to others.”
Not bad, ChatGPT, but what about trade secrets, open source, and IP risk? One should not overlook the fact that AI-based companies that leverage the use of open-source software cannot reveal all of their secrets if they want to be profitable. As such, an IP strategy that uses a layered approach, where the company decides which innovations to patent and which to keep as a trade secret, can have significant advantages.
Moreover, companies engaging in AI innovations such as text and art generators will want to ensure they have a robust and well thought out IP risk mitigation plan. As the law evolves to address advances in AI, businesses should routinely review their IP strategy. The reality is IP is everywhere and is becoming increasingly more valuable in today’s knowledge economy. Protecting AI innovations is no exception. For more information, or if you have questions about your IP strategy and AI-based innovation, please feel free to contact our artificial intelligence practice group.
Denis Keseris is a partner in Bereskin & Parr LLP’s Montreal office, which provides a full range of IP services relating to technology, branding, and marketing.
Ray Kovarik is an associate with Bereskin & Parr LLP and member of the Electrical & Computer Technology practice group.
Nicole LaBerge is an articling student with Bereskin & Parr LLP.