Key legal considerations for generative AI

Mass media attention following the release of ChatGPT by Open AI has pushed the subject of artificial intelligence (AI) back into the limelight. The discussion is particularly focused on the use of generative AI. It’s a term used to describe AI programs that have the ability to create entirely new content rather than simply analyzing existing ones. As ChatGPT shows, the content that can be created by today’s generative AI can be difficult to distinguish from human-generated content, and given simple instructions, AI programs can create digital images, video, audio, and text. , or can generate code in relatively high quality. , in just a few seconds. If you don’t believe me, ask ChatGPT.

The reality is that generative AI is on the way to becoming not only faster and cheaper, but in some cases even better than what humans can create by hand. From social media to games, advertising to architecture, coding to graphic design, product design to law, marketing to sales, all industries that require humans to create original work have been reshaped to varying degrees. may be subject to

This three-part series explores how key areas of law influence the development of generative AI and how it’s used in business. Part 1 focuses on copyright law and key issues regarding ownership and authorship of AI-generated content, while Part 2 discusses privacy law considerations based on Canada’s proposals. artificial intelligence and data law, and Part 3 examines the issue of liability for the creation and use of AI-generated content. In other words, who is responsible for AI-generated content and when?

Part 1: Generative AI and Copyright Law

Canadian law does not determine when and how copyright law should apply to AI-generated content.In light of the current gap copyright law There are two government publications containing recommendations that address some aspects of the AI-copyright interface, as opposed to AI in general. Following his five-year review as required by law, copyright law¹in 2021, the Government of Canada’s Innovation, Science and Economic Development Agency (ISED) will issue a consultation document. copyright law In light of current AI capabilities²The government has not yet submitted any amendments in light of the recommendations contained in the INDU report and ISED document. Despite the uncertainties, we believe that major commercial concerns and legal I fully understand that there will be problems. A.I.

1. Input Issues: Training Data Is Risky

Generative AI can rely on deep learning, a subset of machine learning, to continuously improve its output. Therefore, the quality, usefulness, and breadth of applications for generative AI depend on the training set that the AI ​​learns from. To create this training set, a machine learning engineer needs to source a large amount of reliable and relevant data. For example, if an AI is programmed to recognize a particular item in an image, it will need a number of images with and without that item for the AI ​​to learn and accurately recognize that item. You need a training set consisting of I am asked. One way to obtain these large data sets is through data mining and web scraping. This includes extracting data in various formats from various sources, including the Internet.

The main copyright liability issue that can arise from data entry is the use of data consisting of or contained in copyrighted works to train AI programs, or the use of data by AI programs to It may arise from copying or duplicating in the process of creating a “new” work. AI program. Developers of AI programs should be mindful of the data collection practices they employ to mitigate this risk, but regardless of their due diligence efforts, it is important to efficiently collect the large amounts of data needed to train AI. can be difficult to do. Input is copyrighted.

The INDU report states (in Recommendation 23) that Canada copyright law “to facilitate the use of” [copyright-protected materials] For information analysis purposes. The UK government has indicated that it plans to introduce a new exception into law to allow copyrighted works to be mined and used as part of training data sets for AI programs. After consultation on the proposal, the UK Intellectual Property Office said the new statutory exception would allow the use of text and data mining for any purpose, including commercial, without the need to obtain permission or pay a fee to the copyright owner. published a paper stating that it would be permissible. This is a significant departure from existing rules and text and data mining practices should not be used for research and non-commercial use of copyrighted data. only allowed to be disposed of.³However, UK guidance explains that rights owners still have safeguards in place to protect their content. The main safeguard is the requirement for lawful access. In other words, copyright owners should be able to choose the platforms that make their work available for text and data mining, and charge data collectors for access through such platforms.^Four.

2. Output issue

Many of the copyright issues related to AI-generated content output remain unresolved by Canadian legislators. copyright law It just doesn’t address AI in clear terms. However, two main issues relate to the interpretation of “authorship” and “ownership” of AI-generated works.

Who is the author of a work that is entirely generated by artificial intelligence? This is the fundamental challenge of AI-generated content. For something to be copyrightable in Canada, it must have been created by an “author.” on the other hand, copyright law does not define the term “author”, but most commentators agree that authors must be human, given the text of the law and the axiomatic assumptions underpinning case law. This clearly poses challenges for works created by AI programs. Whether or not to treat AI programs as “authors” inevitably becomes an issue in Canada. In December 2021, the Canadian Intellectual Property Office (CIPO) registered the copyright of a painting co-authored by both a human and an AI painting app for the first time in Canadian history.^FiveWhile the registration’s precedential value is questionable (since CIPO does not review registration applications for substantive accuracy or compliance), the registration is intended to allow copyright owners to It shows that you take the position that works can and should be protected by copyright. The phenomenon is unlikely to abate.

This issue has only been addressed in a limited number of other jurisdictions. For example, in the United Kingdom, the author of a “computer-generated” work is “considered to be the person who made the necessary arrangements for the making of the work”.⁶ Although the meaning of this phrase is ambiguous, it seems inevitable that the author is the programmer or user of the computer in question. Due to the lack of case law interpreting the provisions, the issue remains open, but given the power of modern generative AI, it is unlikely to continue for long. In Canada, the ISED paper explores, without recommending, several potential approaches for determining authorship of AI-generated works. This includes taking those used in the UK and considering them “copyright free” and therefore not subject to copyright protection at all. Or give them a modified and more limited set of rights. For example, the UK approach to “computer-generated” works has a short term of protection of only 50 years (compared to the “duration + 70 years” term that applies to human-generated works). The UK approach also denies moral rights protection to computer-generated works. This should also be addressed by Canadian law.

Thus, in Canada, a solution remains to resolve the copyright issue, and the degree of human involvement in the generation process itself (programming, providing ‘prompts’, etc.) depends on the final resolution. may play a role. problem. In a situation where AI becomes fully autonomous and creates content of its own volition, policy makers will be forced to curb how to assign authorship status, or to consider AI-generated works authorless. You have to decide if it makes sense to establish an out.

2. possession

Following the copyright authorship issue, there is the copyright ownership issue. Ownership issues usually arise only as a function of authorship, as there is a default rule that the author is the first copyright owner (unless there is an employment relationship). Therefore, any solution to the problem of authorship must necessarily contest the problem of ownership. It means that there is no “owner” for the purpose either. It is also necessary to respond to The INDU report emphasizes the importance of this issue and states (in Recommendation 14) that ” copyright law Or to introduce other laws to clarify ownership of computer-generated works. ”⁷ Once the initial ownership answer is identified, most ownership issues can usually be managed by contract. Owners are free to assign or license their rights.

3. practice tips

to copyright law Once modified to address the above issues, the primary concern for users of generative AI is what they can do with the output produced by the AI. be used? Can I use an image created by an AI program on a t-shirt, book cover, or movie? Can I use text created by an AI program on a brochure or novel?

The answers to these questions will ultimately reveal who the content creators and owners are and what uses are permitted. In the meantime, users of AI-generated content should at least check the terms and conditions that accompany AI programs to see if they address those issues. The service provider provides some explanation (ideally a license or ownership transfer). Users of AI-generated content are also aware that use of such content may be viewed by the copyright owners of the content captured by AI as a violation of their rights in such content. must be kept in mind. Lack of proper representations and indemnification from the operators of AI programs may expose you to infringement claims.

4. Conclusion

The following was generated by ChatGPT with minor edits.

In conclusion, Open AI’s release of ChatGPT has drawn attention to the use of generative AI in business, especially the issue of copyright law and how to attribute copyright and ownership to AI-generated content. CIPO has not published guidance on how to address these issues, although other jurisdictions such as the United Kingdom have done so, providing valuable insight into the varying consequences of specific regulations, and the Canadian could serve as a potential benchmark for

Apparently a Canadian copyright law It should be revised to address these concerns and uncertainties. Significant commercial and legal issues arise in the context of both the inputs used to train AI systems and the outputs generated by AI. As the use of generative AI continues to expand, it will be important for businesses to be aware of legal considerations regarding AI-generated content and data input to ensure compliance and mitigate risk.

Part 2 of the series will be released on April 10th and Part 3 on April 17th.