Use of Large Language Model Generative Artificial Intelligence Tools in Research Activities

The emergence of ChatGPT and other large language model artificial intelligence tools since late 2022 has led to rapid developments utilizing these tools in many research disciplines. This is likely to continue to evolve rapidly over the coming years.  This policy is written to provide at least a minimum level of guidance to users of these tools in research activities.

Research sponsors of all types are evolving in their approach to the use of generative AI tools in proposal, review, or report preparation. Publishers are also adjusting their policies to reflect their approach to the use of generative AI tools in publication preparation or review. Whenever generative AI tools are used, the author is always responsible for the final content. It is the University’s policy for Michigan Tech faculty, staff, and students to comply with funding sponsor or publisher requirements for the use of generative AI tools in funding applications or reporting of research results.

All of these commercially available tools, whether free or for purchase, are accompanied by a licensing agreement.  Most of these agreements do not protect data privacy or security. In fact, most explicitly grant the tool provider the explicit right to capture and utilize any information entered into the tool, meaning that users have no proprietary interest either in their input into the tool or in the content generated by the tool. The University does not have any license with any provider of large language model generative AI tools that provides protection of data privacy and security.

Allowable Use

Unallowable Use

  • Any information whose release is prohibited including, but not limited to, HIPAA, FERPA, or Graham-Leach-Bliley protected information, or confidential, identifiable human subjects data, or human subjects data in violation of an approved IRB protocol.
  • Any category of information identified as Controlled Unclassified Information by the United States National Archives, or any information explicitly identified by a sponsor as CUI.
  • Any propriety or other restricted information covered by an executed Non-Disclosure Agreement, grant or contract agreement, or other agreement with an external party.
  • Any information with a copyright owned by an external party. Note that while information posted on the Internet may be perceived as being in the public domain, it is often subject to copyright protection. Entry of copyrighted information into a generative AI tool may constitute infringement of the owner's copyright.
  • Generation of output that would be non-public. This includes, but is not limited to, legal advice, tax advice, personnel evaluations, proprietary research, research with publication restrictions or sponsor review prior to publication, and other non-public work documents.
  • Invention disclosures, patent applications, license agreements, non-disclosure agreements, material transfer agreements, and other intellectual property-related documents and agreements.
  • Any information that can only be accessed behind the University firewall using an ISO password, including non-public information obtained by request from Information Technology, Institutional Research, or others with access to protected data.
  • The providers of generative AI tools may prohibit certain uses of their products. For example, see the Usage Policy for Chat GPT. It is the user’s responsibility to understand such restrictions for any given tool they choose to use in their research, and to comply with those policies.