Artificial Intelligence

Data Types:

• Public Data: Use AI for general information and academic topics within the bounds of Michigan Virtual policies. 
• Confidential/Private Data: Do not enter sensitive student data, confidential information, or information protected by privacy laws.
• Research Data: With appropriate training, prioritize anonymization and legal compliance when using AI for research purposes in education.
• Intellectual Property: Secure written authorization before using a third-party’s intellectual property.  Use caution and when in doubt, seek supervisor approval when using Michigan Virtual IP.

Risks and Limitations:
Beyond considering requirements around different types of data inputs, users of generative AI should be aware of other risks and limitations related to the output generated by these products. 

• Misinformation and Inaccuracies: Actively verify the accuracy of AI-generated content and timeliness, especially for subjects prone to rapid changes.
• Bias: Regularly evaluate AI outputs to identify and mitigate biases, ensuring fairness in information dissemination.
• Inappropriate Content: Monitor AI tools for potential generation of or response to offensive or unsuitable content.

Current Recommended Practices:
When employees use generative AI tools that are not governed by a formal agreement, we recommend the following practices: 

• Consider Impact: Assess the potential consequences of using AI, especially in terms of data privacy and integrity.
• Critical Evaluation: Encourage cross-referencing AI-generated content with reliable sources and consult with technology leadership.
• Stay Informed: Regularly update your knowledge on AI advancements and adapt your use accordingly.

Disclosures:
It is important for Michigan Virtual to set exemplary standards in the use of AI technology. Our employees are expected to:

• Clearly Disclose AI Involvement: Whenever AI tools contribute to the development of higher profile educational materials, research, or work-related artifacts (eg: AI-assisted Michigan Virtual in the creation of this resource).
• Educate Through Example: In demonstrating our use of AI, we also aim to educate. When appropriate, descriptions of how AI was used should be informative, helping the audience understand AI’s role and capabilities.
• Continuously Update Citation Practices: Currently, APA has a recognized citation for ChatGPT. We commit to continuously updating citation practices based on new information as it is made available. 
  • Example: OpenAI. (2023). ChatGPT (June 16 version) [Large language model]. https://chat.openai.com/chat

When to Cite/Attribute: Use of AI in text generation should be cited if the first draft of written content was generated by AI or if AI contributed significantly to the revision of human-drafted text (edits over more than 50% of the text). AI does not need to be cited if it is used as a tool for research, idea generation, or refining existing user-generated text (if edited less than 50% of content). 

Where to Provide Citation: In general, citation or attribution should be provided on the page where it is used. If used consistently and repeatedly over many lessons, attribution in the “course attributions” is also acceptable. When attribution at the point of use is not possible or practical (e.g. within an assessment item), then attribution in the course attributions is also acceptable. 

How to Cite: Exact wording can vary, but generative AI attributions for text should include the following three components.

1. The tool used (e.g. ChatGPT, Claude, etc.).
2. The nature of the use (e.g. generated original text, adjusted reading level, etc.).
3. The nature of the human component (e.g. reviewed and edited by Michigan Virtual). 

Examples: 

• Original text copyright Michigan Virtual, adjusted for reading level by ChatGPT-3.5. 
• Text generated by Claude AI, reviewed and edited by Michigan Virtual.  
• Question text generated by Questionwell, reviewed and revised by Michigan Virtual. 

AI-Generated Images or Media: 

When to Cite: Use of an AI tool in the generation of images or other media (e.g. music or video) should be cited if the original image was generated by an AI tool. AI does not need to be cited if it is used as a tool for editing or modifying existing images (e.g. “magic eraser” or auto-fill tools). 

How to Cite: Exact wording can vary, but generative AI attributions for media should include the following components.

1. The tool used (e.g. Dall-E, Ideogram, etc.). 
2. The nature of the human component, if applicable.

Examples: 

• Image generated by Ideogram. 
• Image generated by Canva, edited by Michigan Virtual. 

Additional Disclosure Requirements: 

• External partners or internal stakeholders should be informed in writing of any AI use that constitutes a need for citation as described above. 

Note: Including the prompt text is not required for citation/attribution. However, when possible, it may be advisable to include prompt text in scripting documents for reference. 

• All AI-generated content must be reviewed, in its entirety by Michigan Virtual staff or contracted staff. 
• Avoid generating content that specifically identifies a unique author or artist when generating content for courses. For example, avoid generating images “in the style of Salvador Dali” or text “in the style of Ernest Hemingway”. Exceptions could be made when those authors/artists are the subject of instruction, but the general idea is to avoid generating content that could be deemed by courts as plagiarized at a later date. 
• Avoid entering any MV IP or data into accounts that are not paid/approved for such use.
• The nature and use of AI tools is constantly changing. If you have questions about whether and how to attribute generative AI tools, please ask. 

• Generative Artificial Intelligence
• Style Guide: PLS Audience
• Style Guide: Language Usage
• Style Guide: Readability – PLS
• Style Guide: Icons & Images
• Style Guide: Image Attribution