AI Guidelines
AI Resources

An AI tool refers to any software that uses machine learning, natural language processing, neural networks, or other computational techniques to perform tasks that traditionally require human intelligence. The range of tasks that AI tools can do include, but are not limited to:

• Generate text, images, tables, or other content based on user prompts (e.g., ChatGPT, DALLE, CLAUDE, Perplexity).
• Analysis of extensive data sets to provide insights and predictions.
• Assistance in repetitive tasks like email replies or meeting notes (e.g., Microsoft Copilot, Google Gemini).
• Enhance written content with AI-powered spelling and grammar review tools (e.g., Grammarly, Microsoft Word Editor, Copilot).
• Assistance in conducting scientific or academic literature reviews (e.g., NotebookLM, scite, Research Rabbit, Semantic Scholar, Elicit, Consensus).
• Assistance in conceptualizing and executing creative projects (e.g., Adobe Creative Suites, RunwayML, MidJourney, Stable Diffusion)

These tools vary in complexity, from simple spell checks to advanced generation of functional computer code. These guidelines recognize all types of AI that can generate or transform content.

At FLC, we acknowledge that AI will continue to evolve, and AI-powered tools are hard to avoid. We are committed to fostering an environment that prepares students for the workforce and supports staff and faculty to use tools that improve their productivity. Here are guiding principles adapted from TeachAI, an initiative by leaders in education and technology that creates practical tools to assist in using and monitoring AI in education settings (see Resources section for a link to the source document).

We reaffirm adherence to existing policies.

AI is one of many technologies used at FLC, and its use will align with existing regulations to protect student privacy, maintain academic integrity, ensure accessibility to those with disabilities, and protect against harmful content. Faculty are strongly encouraged to specify their AI policy in the course syllabus. Without a clear policy, unintended AI use that contradicts course expectations may occur.

We reaffirm adherence to existing regulations.

We will not share personally identifiable information with consumer-based AI systems. We will thoroughly evaluate existing and future technologies and address any gaps in compliance that might arise. We educate our faculty, students, and staff about AI. Promoting AI literacy among faculty, students, and staff is central to addressing the risks of AI use and teaches critical skills for students’ futures. All members of the FLC community will have access to resources to develop their AI literacy, including initial training and best practices in using AI. Faculty interested in adapting instruction to use AI tools can access training and resources through the Center for Teaching and Learning (CTL).

We explore the opportunities of AI and address the risks.

In continuing to guide our community, we will work to realize the benefits of AI in education, address risks associated with using AI, and evaluate if and when to use AI tools, paying special attention to misinformation and bias.

We support using AI to help all students achieve their educational goals.

We will use AI to help us reach our community’s goals, including improving student learning, teaching effectiveness, and college operations.

We aim to make AI resources accessible, focusing primarily on bridging the digital divide among students and staff.

We are committed to evaluating AI tools for biases and ethical concerns, ensuring they effectively serve our diverse educational community. The FLC AI Resources Committee maintains a list of AI tools available to students, faculty, and staff.

We maintain student and faculty agency when using AI tools.

AI tools can provide recommendations or enhance decision-making, but faculty and students will serve as “critical consumers” of AI and lead academic decisions about its use. Faculty will have shared governance in decisions on how and where AI tools are used.

We follow academic integrity policies related to the use of AI.

All campus community members must adhere to academic integrity policies regarding AI use. The FLC Integrity Procedures on GenAI and any AI-related policies outlined in course syllabi take precedence. It is the responsibility of individuals to review and comply with these relevant campus guidelines.

We commit to auditing, monitoring, and evaluating FLC’s use of AI.

Understanding that AI and technologies are evolving rapidly, we regularly review and update our policies, procedures, and practices frequently and regularly. 

The totality of the risks of using AI tools is not fully understood. FLC is committed to monitoring risk, communicating broadly when new risks emerge, and keeping reference documents updated with known risks. 

All output generated by or edited by AI tools must be reviewed for inaccuracies. The primary types of errors include:

• Hallucinations: Information that is completely fabricated and often presented in an authoritative tone.
• Algorithmic bias: Any bias in the data used to train the model will be replicated in the output. For example, an AI tool for face recognition can be biased by the race and ethnicity of the images it was trained on. See the Ethics section below for related information.
• Unnatural language: The output can sound robotic and use fragmented, unnatural wording. Newer LLMs and AI tools explicitly designed to address this issue, called AI text humanizers, have addressed this issue.
• Lack of depth: Results can be accurate but be over-simplistic, lacking the depth of a critical evaluation of a topic.

A lack of accountability and transparency in the source data used to train AI models and how user data is stored and potentially used can create ethical dilemmas and privacy risks when using AI tools. A few common examples are:

• Improper use of intellectual property: Often, AI is trained on copyrighted material, and the tool may not provide the user with the information needed to create a proper citation. •
• Opaque settings: Due to the risks associated with copyrighted material, it is often not transparent what uncopyrighted material may be used to train an AI model, including any information the user volunteers when using a tool.
• Limited oversight: Industry and government oversight of AI is under development and will need to evolve as quickly as the technology.
• Terms and conditions of use: The training data used to develop AI models is a significant part of an AI company's intellectual property. It is not uncommon for the terms and conditions statement a user must agree to before using a tool to give the company permission to add any information a user adds to their training data. In addition, digital content publishers, like academic journals, can also have a similar statement in the publishing agreement. This introduces a privacy and/or intellectual property risk.
• Environmental impact: AI technologies, particularly large-scale generative AI models, can have significant environmental implications due to the high energy consumption required for training and deploying these systems. Users should be mindful of the carbon footprint associated with AI usage and prioritize tools or practices that minimize environmental harm when possible.

As with any technology, AI tools are vulnerable to being outdated, cumbersome, and not adequately protected.

• Outdated source data: If an AI model is not regularly updated, it can be pulled from obsolete materials, and the user cannot unquestioningly trust that the output is the most current.
• Security risk: The AI models built and maintained by well-established entities, like OpenAI, Google, Amazon, and Microsoft, are more likely to be secure from hacking than those built by small unknown entities.
• Limitations of a free version: Using a free version of any AI tool can come with notable tradeoffs compared to a paid version. The output may be of lower quality and/or include more biased results due to the free version using a less powerful or outdated model. The security protocols of free versions can be weaker, and user data can be harvested and monetized.

AI tools can introduce legal risks, particularly in areas where intellectual property, liability, and regulatory compliance intersect. Fort Lewis College is committed to proactively understanding these risks, regularly reviewing the legal frameworks governing AI, and providing users with best practices for mitigating legal exposure.

• Copyright infringement: AI tools may generate content replicating copyrighted material without permission. Users and institutions could face legal liability if this content is distributed or used without proper attribution or licensing.
• Data protection laws: AI tools often collect and process user data, which may lead to breaches of data protection regulations, such as GDPR or FERPA. For example, sharing personally identifiable information with an AI tool could unintentionally violate privacy laws.
• Contractual obligations: Many AI tools have terms of service agreements that include clauses on data ownership, liability, and permissible uses. Users must ensure they are not violating contractual obligations or inadvertently exposing the institution to legal disputes by agreeing to these terms without fully understanding.
• Defamation and misinformation: If AI-generated content includes inaccurate or defamatory information, the institution or user responsible for distributing this content could face legal repercussions.
• Product liability: Determining liability can be complex in cases where AI outputs lead to harm or damages (e.g., incorrect medical advice or biased hiring decisions). Users and institutions may need to demonstrate due diligence in reviewing AI outputs and adhering to ethical use standards.

Before using any AI tool, it is recommended that you consider the following topics:

Has this tool been approved for use at FLC?

Consult the list of approved AI tools before trying to use a new one you heard about from a friend or came up with from an internet search. The FLC AI Resource Committee has considered the critical privacy, safety, and functionality issues for you.  

Is this the right tool for the task?

Not all AI tools are created equal, and each has been optimized with a specific outcome in mind, such as generating citations, summarizing text, or processing large numerical data sets. For example, ChatGPT could be great at summarizing meetings or lecture notes, but it is not the right tool for help with a scientific literature review.  

Is there any training available?

Many new AI tools are designed to be easy to use, but it can be easy to use an AI tool and not get the best performance out of it. Attending training can improve the results you get from an AI tool and reduce privacy risks. At FLC, training is offered by the Center for Teaching and Learning, Reed Library, the AI Resources Group, and the AI Institute at the FLC Center for Innovation.

Can I accomplish this without adding protected information to the AI tool?

Most AI tools reserve the right to use any information you type in as a prompt or upload to the tool to train their models. Protected information includes personal data (e.g., mailing address, social security number, student ID), FLC data (e.g., grades, budget details), and any proprietary data (e.g., results from an unpublished research project). Complying with federal regulations such as the Family Educational Rights and Privacy Act (FERPA) and the Health Insurance Portability and Accountability Act (HIPAA) when handling sensitive data is crucial. These laws protect student education records and personal health information, respectively, and require careful consideration to ensure AI tools do not inadvertently share or process such data.

Are the results accurate?

It is essential to review results carefully for accuracy and clarity. In the same way you check the results of an internet search for quality links you can trust, text generated by AI needs a human to review the results for accuracy. One way to do this is to ask the AI tool for references that you can check yourself.

Should I cite the use of the AI tool in my work?

Yes! Properly citing the use of Generative AI (GenAI) tools in creating written materials, visual images, or other outputs is essential for maintaining transparency, academic integrity, and accountability. Examples of how to cite AI can be found in the Resources section of this document.

Help with citations

American Psychological Association (APA)

In APA 7th Edition, you would cite it as a personal communication. Since personal communications are not recoverable, they are only cited in the text and do not appear in the reference list.

In-text citation: (OpenAI, personal communication, October 21, 2024)

If citing a published version of the model (like an article or documentation): OpenAI. (2024). ChatGPT: Language model. OpenAI. https://www.openai.com/chatgpt

Chicago Manual of Style (CMOS)

In Chicago style, you would handle it similarly to software or personal communication.

In-text citation: (OpenAI, accessed October 21, 2024)

Bibliography: OpenAI. ChatGPT. Accessed October 21, 2024. https://www.openai.com/chatgpt.

Modern Language Association of America (MLA)

In MLA style, you can cite the LLM as a software or tool.

In-text citation: (OpenAI)

Works Cited: OpenAI. ChatGPT. Version 4, OpenAI, 2024, www.openai.com/chatgpt.

Guidelines from trusted external organizations

• AI and Education: Guidance for policymakers: A comprehensive United Nations Educational, Scientific and Cultural Organization (UNESCO) report.
• AI Preparedness Checklist: 1EDTECH guidance on policy, literacy, and pedagogy.
• Columbia University AI Governance: Policy on the use of generative AI.
• Teach AI: A non-profit focused on developing tools to assist educators in understanding and using AI.

AI text humanizer: A class of AI tools explicitly designed to edit content from generative AI to sound more like natural language. These are often marketed to individuals looking to pass AI detection tools.  Algorithmic bias is a systematic and repeatable error in the results of a computer system, like AI, that results in unfair outcomes. These errors often reflect and amplify societal biases in the data used to create and train the system. Since the details of the data used and the algorithms applied are the intellectual property of the system's developers, it can be challenging to prescreen for these errors or biases.

Confidential information: Sensitive data is restricted to authorized individuals or organizations to prevent unauthorized access, disclosure, or harm. This can include, but is not limited to, health data, academic records, and financial information.

Generative AI (Gen AI): A type of AI that can create new content, such as text, images, or music, by learning patterns from existing data.

Large Language Models: Advanced systems that train on vast amounts of text data to understand, generate, and manipulate human language.

Machine learning: A type of AI that enables systems to learn from and make data-based decisions without being explicitly programmed for each specific task. It includes different approaches, such as supervised learning (learning from labeled data), unsupervised learning (finding hidden patterns in unlabeled data), and reinforcement learning (learning through trial and error with rewards).

Personal information: Like confidential information, it is any data related to an identifiable individual, including names, contact details, and personal identifiers like social security numbers.

These guidelines will be reviewed at least annually and edited more frequently if needed if they meaningfully impact the types of AI tools available, the regulatory landscape, or changes in related FLC policies. An interdisciplinary team will review edits to this document's content before implementation, and the document will be versioned and tracked in the Document History table at the end of this document.