Introduction
Please review the Generative Artificial Intelligence: Policy Intersections, Considerations, and Recommendations published by Minnesota State. The below information underscores key notes from the current Minnesota State guidance.
Procurement Process for AI Services
The Minnesota State guidance on generative AI services requires adherence to standard procurement processes, including legal, security, and contract reviews, even for zero-cost services. Authorized services with data security contracts include Adobe Express & Creative Cloud, Microsoft 365 Copilot Chat & Microsoft 365 Copilot, and Zoom AI Companion. Highly restricted or restricted data, like personal health information or student grades, must not be used with non-contracted services like ChatGPT, Google Gemini, or Anthropic Claude.
Data Use with AI Services
Providing or using highly restricted or restricted data in any third-party application or service, including generative AI services, requires a contractual agreement with the third party that ensures adherence to data security and data sharing protocols.
- Highly Restricted – Social Security Numbers, personal health/medical information, banking, or credit card information, etc.
- Restricted – Student grades, transcripts, class schedule, employee personal contact information, individual demographics including age, race, ethnicity, gender, etc.
- Low – Data that by law is available to the public upon request.
No highly restricted or restricted data should be used with applications or services without a contract (examples: ChatGPT, Google Gemini, Anthropic Claude).
Content Accuracy
Always verify the accuracy of content generated by AI to ensure its reliability and correctness. Pay close attention to omitted data, inaccurate data, or unexpectedly added data.
Generative AI tools are not infallible, and their accuracy is subject to a variety of factors, including:
- Prone to filling in replies with incorrect data if there is not enough information available on a subject.
- Lack of the ability to understand the context of a particular situation, which can result in inaccurate outputs.
- The extensive data sets used to train these models may contain biases, which can influence the outputs generated by the AI.
- Algorithms used by these technologies can, and do, replicate, and produce biased, racist, sexist, etc. outputs, along with incorrect and/or misleading information.