WIPO releases patent navigation "Generative Artificial Intelligence"

Submitted by song on
来源:
页之码IP

In February 2024, WIPO released the patent navigation of "Generative AI: Navigating Intellectual Property" with the aim of helping all parties understand the intellectual property risks of using generative artificial intelligence tools and the possible protection measures that can be taken.

What is Generative AI?

  • Generative AI, abbreviated as: Generative AI, referred to as: GenAI, is a method of briefly describing the desired product and creating new content such as text, code, images, music, sound and video. Currently, GenAI's tools include ChatGPT, Midjourney, Copilot and Firefly.
  • GenAI uses machine learning methods and uses a large amount of data, usually including billions of pages of text or images, for training. The training data may be freely available, unrestricted data, protected data, or a mixture of the two. The trained AI tool will trigger a series of complex calculations based on human input prompts to produce results, and the results of the output are usually impossible to predict, and it is impossible to determine whether the training data will affect the output.

GenAI and Intellectual Property

GenAI has many intellectual property risks and uncertainties. Although these IP risks cannot be completely resolved, companies and related institutions can consider possible protection by taking the following precautions.

I. Confidential Information

  • Confidential information refers to non-public information, not limited to information with commercial value, such as trade secrets. Trade secrets are confidential information with potential economic value or a competitive advantage due to their confidentiality. If this information is used for training or AI prompts, companies and institutions using GenAI tools may inadvertently disclose trade secrets or abandon the confidentiality of trade secrets. Companies should consider taking necessary technical, legal and practical protection measures to prevent this from happening.

II. Intellectual Property Infringement

  • Many GenAI tools have training data from works protected by IP, and legal disputes have arisen over this, alleging that the use of IP-protected works to train AI, and that the trained AI models and their outputs constitute IP infringement. These disputes are mainly concentrated in the fields of copyright and trademark, but also involve industrial design, databases, and invention patents. There is currently a lot of legal uncertainty as to whether the training, use, and output of AI tools constitute IP infringement, and it may vary by jurisdiction. Companies should seek compensation, review data sets, and examine technical and practical measures to reduce the possibility of infringement by using compliant IP tools.

III. Open Source Obligations

  • Code generated by AI may be subject to open source obligations. When a piece of software is derived from open source code, users have the right to freely use, modify, and disseminate the software, but these rights and freedoms are accompanied by obligations that users must comply with, such as clearly indicating the source, etc. These obligations may vary depending on the specific open source license that governs the software. Enterprises should consider whether their code is subject to the above risks, investigate potential compensation protections, and develop technical and practical measures.

IV. Deep Fakes: Image and Voice Rights

  • In many countries, portraits and voices may be protected, but the protection systems vary. The forms of protection include intellectual property rights, unfair competition, human rights, constitutions, and the right of public expression. GenAI may imitate the portrait or voice of a specific subject, and may even be designed for this purpose. This is also a related risk that companies need to consider.

V. Intellectual Property and Ownership of GenAI

  • The intellectual property attributes and ownership of GenAI's generated products, such as text, images or other creative works, are currently unclear, and companies should clearly stipulate ownership in the contract.

Users can go to the official website of Yezhima to download the full text of WIPO Generative AI Patent Navigation.