Initial Guidance for Researchers:  Generative AI in Research 

Generative AI tools, such as chatbots (ChatGPT, Bing Chat) and art systems (Midjourney, DALL-E), have great potential for university research.  Brown investigators contemplating using generative AI tools in their own research should be cognizant of the various intellectual property (“IP”) issues related to its use.  

Technology typically outpaces policy and law, and generative AI (“GenAI”) is a particularly dynamic area of technology development.  The guidance on the intersection of generative AI and IP is an evolving area.  Any questions regarding the use of generative AI in your specific research program related to IP may be directed to  [email protected]

Issues Arising from the Use of Generative AI

A. Public Disclosure/Confidentiality.

  1. Public Disclosure. Patentability.  Any given GenAI may not provide sufficient protections to ensure data privacy or confidentiality.  Make sure the GenAI has terms of use stating that any data you provide and your results will not be made public, bearing in mind that any cloud-based GenAI can be vulnerable to public disclosure.  Public disclosure of an invention can preclude Brown’s ability to secure patent rights for that invention. 

  2. Misappropriation of Research Results.  Make sure the GenAI terms of use prevent the GenAI company from using your research results for its own commercial purposes. Brown researcher’s data and results are typically considered confidential and proprietary until a decision is made to publish a manuscript. 

B. IP Ownership.  

Ownership issues pertaining to three distinct types of properties are addressed:  researcher developed generative AI models (the software algorithms designed to “learn” from training sets and generate outputs); researcher-curated training sets (the large data sets “learned” by GenAI systems); GenAI output results.  

  1. Ownership of generative AI models.  Brown researchers may be involved in developing novel GenAI software and algorithms.  As a reminder, software source code is protectable by copyright and algorithms may be protectable by patents. 

  2. Ownership of Brown Researcher-Curated Training Sets.  Often, the value of a GenAI tool is derived from its training sets that yield superior results. BTI works with researchers to protect proprietary rights in commercially valuable training sets, while preserving the ability to make it available for nonprofit research. 

  3. Ownership of GenAI Results.  Make sure the GenAI tool has terms of use that state that the user will own output results.  Not all GenAI tools have terms of use that are clear or sufficient to ensure ownership. 

  4. Infringement Risks in GenAI Results.  Safe practice is to only use GenAI systems that disclose the sources of its training sets.  Many GenAI tools do not disclose their training set sources and may contain material scraped from the internet without permissions, which amounts to copyright infringement. The output results of such GenAI tools may contain or reflect copyright infringement.