AI at Middlebury

AI (Artificial Intelligence) in education often refers to generative AI (GenAI) tools that can create text, images, code, and other content based on user prompts. These tools use large language models trained on vast amounts of data to generate responses.

However, the term AI could be referring to any of the below items. These items fit together like nesting dolls, from the broadest concept to the most specific component:

• Artificial Intelligence (AI) is the overall field of creating tools that can perform tasks that typically require human intelligence.
• Machine Learning (ML) is a subset of AI. Instead of being explicitly programmed, a machine identifies patterns and makes predictions from data. A good example of this is the spam filter in your email that “learns” to identify junk mail.
• Deep Learning (DL) is a specialized type of Machine Learning that uses complex, multi-layered systems to find intricate patterns in massive datasets. Think of the search function in your phone’s photo app. You can type “dog,” “beach,” or “birthday cake,” and it seeks every relevant picture you have, even though you never tagged them.
• Large Language Models (LLMs) are a prime example of Deep Learning in action. An LLM is a giant neural network trained on a vast amount of text from the internet, giving it a sophisticated ability to understand and generate human-like language. LLMs are the core technology behind tools like ChatGPT.

How GenAI Works (Simply Put)

GenAI tools like ChatGPT, Claude, Microsoft CoPilot, Google Gemini, and others predict what text should come next based on patterns in their training data. They don’t “think” or “understand” in the human sense—they’re sophisticated pattern-matching systems that can produce remarkably human-like outputs.

Important Limitations of GenAI

AI can produce inaccurate information (called “hallucinations”), reflect biases from training data, lack true understanding or reasoning, struggle with very recent information, and cannot replace human judgment, creativity, or critical thinking. These limitations make it crucial to verify AI outputs and guide students to become critical users.

Middlebury faculty are empowered to choose their approach to teaching with generative AI (and any other technology). We recommend that faculty explicitly state their position on AI usage in their course syllabi. That can be done at a macro level for all assignments or set by individual assignment.

It is important to keep all these Middlebury AI policies in mind when crafting your classroom AI approach.

Middlebury provides single-sign-on access to a variety of online AI tools, detailed below. Additional domain-specific AI options may be found in particular software (like ESRI or Adobe’s tools). AI tools generally fall into these categories:

• General Chatbots
• Media Generation Tools
• Discipline-specific Tools

Why Use Institutional Tools?

When recommending AI tools to students, we recommend you prioritize those available through Middlebury credentials to ensure data privacy and equitable access. If students use external tools, remind them not to input sensitive personal information or proprietary institutional data. Institutional tools provide:

• Enhanced privacy protections for student and institutional data
• Compliance with educational technology standards
• Technical support from Middlebury ITS and DLINQ
• Often offer more powerful features than free versions
• Integration with other Middlebury systems

General Chatbots

Chatbots are the most familiar generative AI tool. Middlebury provides access to the free educational tier for two chatbots. This tier of access limits certain functionality and restricts the total number of prompts that can be run in a day. Our educational licensing provides additional privacy protections by default.  

• Google’s Gemini chatbot supports multimodal interaction (including speech-to-text, image generation, and image/file upload options) making it a versatile tool for tasks like visual analysis, content creation, and multi-modal exploration. The “deep research” mode provides a sophisticated search integration. Google App integration is not available at this time but some content can be exported directly into Google document formats from the chat interface.
• Microsoft’s Copilot provides similar functionality. A key difference is that Gemini allows code previews without needing to download and open generated code. MS Office integration is not available at this time.
• Google also provides access to Notebook LM. This is a popular tool in education because it enables you to upload content that is prioritized and cited in the chat conversations to allow for verification. It can also generate audio overviews in a variety of languages and concept maps.

Media Generation

In addition to the image generation options available in Gemini and Copilot, Middlebury provides access to Adobe’s Firefly which can generate images, video, and vector graphics. Adobe Express also allows a variety of different media editing and creation options including generating editable templates for documents like posters, cards, and flyers.

Discipline-specific Tools

Many tools have incorporated AI for years. For example, Mathematica and MATLAB offer machine learning toolboxes, and ESRI products include deep learning models for pattern recognition and predictive analysis. These forms of AI focus on structured modeling and analysis and are distinct from the generative AI currently receiving the most attention.

Some platforms, like Mathematica, offer a sophisticated mix of AI options, including the integration of generative AI models through API connections, code assistance, machine learning modules, etc.

We continue to see more generative AI embedded directly in software. It can show up as assistants, automated categorization, or content generators. In some cases, these features can be enabled or disabled by local administrators. In others, generative AI is integrated by default, with limited or no ability to turn it off.

Specialized Generative AI Tools

Middlebury does not currently support any generative AI tools specific to research, audio creation, computer code, data analysis etc. These tools exist, however, and your students may be aware of them.

For Instructors

If instructors have questions about teaching and learning with AI either in their classrooms or their research, they can reach out to:

• ITS Helpdesk can answer questions about tool access and your Middlebury login account.
• DLINQ can help answer questions and provide support related to AI use in teaching and learning, including:
  • designing AI-inclusive or AI-resistant assignments or assessments
  • crafting AI syllabus statements
  • providing AI workshops for classes or groups or faculty
  • designing AI tools for specific teaching or research use
• Middlebury Library can help with questions about academic integrity and research support related to AI.

If you’d like to request that DLINQ staff visit your class and provide a workshop on specific AI tools and/or help curate resources for your students, email dlinq@middlebury.edu with your request or fill out our online consult form and we will be in touch to discuss your request. In your request, please:

• Identify your learning objectives (e.g., “I want students to learn effective prompt engineering”)
• Specify your discipline and course level
• Indicate whether you need support for yourself, your students, or both
• Provide timing preferences and format (in-class visit, separate workshop, online resources)
• Share any specific tools or applications you want to focus on
• Note how many students are in your class.

DLINQ also has designed a short Canvas course titled Intro to GenAI @ Middlebury that you are welcome to use with your classes

For Students

If students have questions about using AI in their learning at Middlebury, here are some key contacts they can reach out to:

• The course instructor should be the first point of contact for students with questions about AI use in a specific class.
• DLINQ interns are Middlebury students who can help answer questions about AI tools, capabilities, and responsible use.
• Middlebury Library can help with questions about academic integrity and research support related to AI.
• ITS Helpdesk can answer questions about tool access and your Middlebury login account.

DLINQ is providing a monthly email update containing discipline-specific AI resources. Each newsletter contains a Faculty Spotlight highlighting how Middlebury faculty are using (or refusing) AI in their teaching and research. Read the updates on our website or subscribe to receive the newsletter in your email inbox.

What should I include in a syllabus statement about AI?

Course AI expectations typically fall into three categories:

1. Prohibited Use: In these courses, using generative AI for any part of an assignment without permission is inappropriate and would be considered a violation of the Honor Code.
2. Limited Use: These courses may permit specific use of AI and indicate appropriate use for tasks like brainstorming ideas, supporting research, checking code or providing feedback on a draft. These courses will often include requirements for the acknowledgment and citation of AI use.
3. Allowed or Supported Use: Some classes will actively encourage or even require use of AI tools as a part of the learning process. In such courses, instructors will often treat learning to collaborate with these tools as an emerging and valuable competency.

In addition to including an AI policy in your syllabus, we recommend providing AI-related guidance for the major assignments in the class. This guidance should outline what AI use, if any, is allowed and why AI is permitted or restricted on the assignment.

If you would like additional help creating a syllabus or assignment AI policy, you are welcome to reach out to DLINQ for a consultation.

We recognize that faculty have different comfort levels and expectations for AI use in teaching and learning, including prohibiting its use. Below are some tips for designing AI-inclusive and/or AI-resistant assignments in your classroom.

AI-Inclusive Assignment Strategies

• Comparative Analysis: Have students compare AI-generated responses with expert sources or their own analysis
• AI for Brainstorming: Students use AI for initial brainstorming or drafting, then critique and substantially revise, documenting AI prompts used, revisions made (and why), and questions or concerns that are raised along the way
• Error Detection: Provide students with your own AI-generated content that includes intentional flaws for students to identify and correct
• Prompt Engineering: Assign tasks where students develop and refine prompts to achieve specific outcomes
• AI Tool Evaluation: Have students test multiple AI tools and compare their effectiveness for specific tasks

We recommend always requiring reflection on the AI collaboration process. Ask students to document what AI helped with, where it fell short, and what unique human contributions they made. This metacognitive process is crucial for deep learning.

AI-Resistant Assignment Strategies

Focus on process, personalization, and synthesis rather than just final products. Require multiple drafts, reflective essays, or connections to specific course discussions.

Use local examples, current events, or personal experiences. Design assignments that benefit from AI collaboration rather than fighting it—for instance, having students critique and improve AI-generated responses.

AI-resistant strategies include incorporating:

• Process-focused assessments
• Personalized and contextual tasks
• Authentic performance tasks
• In-class and synchronous assessments
• Metacognitive and reflective approaches
• Assessment design principles

Don’t assume students know how to use AI tools well just because they have access to them. Topics that can be helpful to cover for students new to AI use include:

• Effective Prompting: how to write clear, specific prompts and iterate based on results
• Critical Evaluation: how to verify facts, check sources, and identify AI limitations
• Ethical Use: how and when to provide attribution, transparency, and appropriate vs. inappropriate use cases
• Tool Selection: how to choose appropriate AI tools for different tasks, and what AI tools are available via their Middlebury credentials
• Privacy Awareness: general data privacy concerns related to AI, including what information should not be shared with AI tools

Other AI resources that can be helpful for students include the following.

• Specific examples of appropriate use for different assignment types
• Step-by-step instructions for using recommended tools
• Common pitfalls and how to avoid them
• Requirements for documenting and citing AI assistance
• Where to get help if they encounter problems

If the AI tool you are using in class is unique or completely new to most students, you might also consider dedicating class time to hands-on practice with the tool, allowing students to experiment in a low-stakes environment with your guidance before they begin work on a higher-stakes or more complex project.

DLINQ can provide resources and deliver workshops for you and your students on any of the above topics.

How do I grade work that intentionally involves AI assistance?

Focus on evaluating critical thinking, reflective documentation, process, and learning progress rather than just final products. Use rubrics that value process, reflection, and synthesis.

Consider having students document their AI use and reflect on its effectiveness and how they integrated AI output into their work. Grade conversations, presentations, and/or revisions to assess genuine comprehension beyond what AI might produce.

Please reach out to DLINQ if you’d like to talk about creating content or AI tools for specific teaching or research applications. DLINQ also offer fellowships supporting your exploration of AI.

When creating course materials, faculty should consider both the benefits and limitations of using AI tools. Faculty should critically review all AI-generated content for accuracy, bias, and alignment with learning objectives before using it with students. Ensure the outputs reflect your teaching philosophy, disciplinary knowledge, and specific course context. Consider whether AI-generated materials might inadvertently model approaches you don’t want students to emulate, and ensure content matches your voice and values.

There are a variety of ways you can use AI in course content creation. AI can be a useful brainstorming partner or efficiency tool.

Traditional content

AI can assist in traditional content creation like drafting syllabi, generating discussion prompts, creating practice problems, developing rubrics, or designing assignment variations. Prompts like the following can play a roll.

• Assignment Variation: “Generate 5 alternative versions of this essay prompt that maintain the same learning objectives but vary the context, examples, or approach to accommodate different student interests and learning styles.”
• Discussion Question Development: “I’m teaching [topic] next week and want to move students from basic comprehension to critical analysis. Suggest 3-4 discussion questions that build on each other, starting with accessible entry points and progressing to more complex synthesis.”
• Assessment Rubric Refinement: “Review this draft rubric for [assignment type]. Identify any unclear criteria, suggest more observable indicators of student performance, and point out potential gaps between what I’m assessing and my stated learning objectives.”

Media creation

AI tools can help faculty create multimedia content that supports Universal Design for Learning (UDL) principles by providing multiple means of representation, engagement, and expression. Tools like Adobe Firefly, Gemini, and Copilot can generate images, diagrams, and other multimedia that complement text-based materials. NotebookLM and AI Studio can create audio overviews in multiple languages. AI can help create visual metaphors for abstract concepts, or produce scaffolded content variations for different skill levels.

• Create a visual metaphor. “Create a photorealistic image of George Washington as a bodybuilder. Have him doing a bicep curl. Make him very muscular, but preserve the George Washington face and hair.” See examples images from Adobe, Gemini, and ChatGPT.
• See a 3 minute video on creating two-person audio conversations in Korean (or the language of your choice) using Google’s AI Studio.
• Transcribe handwritten primary source material to improve accessibility, improve workflow efficiency, and enable digital analysis.

AI-generated multimedia requires careful review for accessibility (alt text, captions, color contrast), accuracy, and cultural sensitivity. Consider whether generated visuals reinforce or challenge stereotypes, whether they’re appropriate for your disciplinary context, and whether they actually clarify concepts rather than add decorative complexity. AI works best when you provide specific parameters about audience, purpose, and accessibility requirements.

Interactive content and tool creation

AI can serve as both a ready-made tool and a tool for creating non-AI tools tailored to your specific needs (pedagogical or otherwise). Platforms like Google’s Gemini allow you to create custom “Gems” with specialized instructions and content, while NotebookLM enables course-specific interactions grounded in your syllabus, readings, or discipline-specific materials. At the same time, you can use AI to help build non-AI tools that accomplish your goals even if you cannot program with tools like Google’s Opal and .

• Custom Gem Design: “Help me design a custom Gemini Gem for my [course name] that acts as a Socratic tutor. It should ask clarifying questions before providing answers, connect student questions to course readings and key concepts we’ve covered, and encourage students to articulate their reasoning process rather than giving direct solutions.”
• Interactive Tool Development: “I want to create a simple web-based tool where students can input [data/parameters] and visualize [concept/relationship]. I have no JavaScript experience. Walk me through building this step-by-step, explaining what each part of the code does in terms I can understand as a [discipline] professor.”
• Course-Specific NotebookLM: Upload relevant course content to a NotebookLM and guide students in healthy use patterns. NotebookLM is popular among both faculty and students due to the ability to restrict responses to uploaded content and verify responses easily using the in-tool footnotes.

See these additional pages for more details, examples, and resources related to each topic:

Crafting Classroom AI Policies

Crafting Assignments to Address Specific AI Concerns

You can use AI to search the internet much like you would Google or another search engine. It becomes more beneficial when you use a very specific prompt to formulate the results and to make sure it provides a link to the source material.

Provide a list of the 5 most reputable Alan Turing sites. Focus on sites associated with education, non-profit, or government domains. For each site, provide a brief summary of its unique aspects and explain why it is reputable. Indicate what kind of primary resources it provides and link directly to the site.

Increasingly, academic database tools like JSTOR, ProQuest and Statista are including AI to simplify search and to provide overviews and interactions with the content they contain. These platforms are changing rapidly and don’t necessarily provide choice in whether the tools are activated or not.

AI platforms can be used directly for immediate tasks. We’ve had success with the following when working with faculty across disciplines.

• OCR: Extract text from images across multiple languages, both handwritten and printed
• Data translation: Translate content between languages or convert data formats
• Function generation: Create formulas for Excel/Google Sheets
• Code troubleshooting: Debug and optimize code

There are times when you want to reuse complex prompts, focus responses on defined content, or do some combination of the two.

• Custom Gems: Build specialized assistants in Google Gemini with specific instructions and behaviors for tasks you do regularly or that you want to share with colleagues
• NotebookLM: Ground AI interactions in your own documents and materials, ensuring responses draw directly from your content

Use AI to help build traditional (non-AI) tools that automate workflows or provide other useful functionality.

Please reach out to DLINQ if you’d like to talk about using AI in your research or to make tools. We can provide examples, best practices, and provide overall guidance to help you get to your desired destination.