Information Services & Technology

 WCU's Acceptable Use Policy (AUP) emphasizes that University network and resources are intended to support the academic mission, facilitate information-sharing, and manage the University’s administrative and service operations. When using generative AI tools, it is important to adhere to the established principles of academic integrity and professional responsibility. This includes using these tools ethically and responsibly to support academic and administrative activities without compromising the integrity of the University's systems and data.

 The university’s data classification policy provides direction regarding the privacy, security, and integrity of WCU data and outlines the responsibilities of institutional units and individuals for such data. To protect the confidentiality and integrity of sensitive data and prevent unauthorized disclosure or use, do not input any sensitive data into generative AI tools.

To uphold and maintain the ethical standards outlined in the Undergraduate Student Academic Integrity Policy  and Graduate School Academic Integrity Policy, it is important to use generative AI tools responsibly and ethically. Avoid violating the following academic integrity standards from each policy: 

1. Plagiarism: Do not present someone else's work as your own. Output for generative AI tools is based on the existing works of others. This includes words, ideas, data, visual representations, audio, video, and digital materials. In keeping with recommendation #5 verify the accuracy of AI outputs, you should confirm that any AI output you use is not a direct copy of other’s work. This is based on the University of Florida’s recommendation regarding plagiarism and generative AI.
2. Fabrication: Do not present fabricated or unverified information, results, information, citations, or other findings.
3. Cheating: Do not misrepresent your mastery of information or skills for assessments. This includes, but is not limited to, using or attempting to use unauthorized materials, information, or study aids in any academic exercise. 

Generative AI tools can be used ethically and responsibly to assist or enhance content development without compromising or violating academic integrity.

• The Family Educational Rights and Privacy Act (FERPA) protects the privacy of students by maintaining the confidentiality of their education records. When using generative AI tools, do not input any student records, grades, class rosters, or other educational information protected by FERPA.
• In accordance with the Use of Human Subjects in Research Policy, faculty, students, and staff conducting research should ensure that they do not input into generative AI tools any information collected for specific limited purposes, such as research data or information where a person’s identity is or could easily be connected to the data. 

Confidential data requires the highest level of protection to prevent unauthorized disclosure or use. This includes data the University must keep private under federal, state, or local laws. Examples include personally identifiable information (PII) such as social security numbers and protected health information (PHI) such as health records.

Entering any personally identifiable information (PII) such as names, email addresses, birthdates or ID numbers, into an AI platform could cause the tool to repeat this information as part of interactions with other users at a later time.

Sensitive data is generally private to West Chester University and access is limited to the West Chester community members on a need-to-know basis. These data are not typically available to external parties. Examples include certain research records, library and archive circulation and order transactions, and university partner or sponsor information.

Any sensitive information, personally identifiable information (PII), or text/information that is not already public knowledge is at risk of being exposed if it is entered into an AI tool. Extra caution should be taken with information regarding hiring, admissions, or evaluation of students or employees as it is often combined with other detailed personal and confidential information.

Public data has no legal or other restrictions on access or usage and is generally open to the WCU community and external parties. Examples include information already available to the general public on WCU’s website, approved official meeting minutes, published research, policies, and posted press releases.

Generative AI tools are based on mathematical models that predict likely outputs for a given prompt, or input. These predictions are based on mathematical patterns in the underlying training data the AI tool is based on. In some cases, these patterns lead the model to give factually accurate answers to questions. For example, if asked for the capital city of Germany, most current models will accurately answer “Berlin,” because the model has noticed patterns linking words like “capital” and “country” and “Germany” and “Berlin.” 

However, models may also provide inaccurate information. This may occur for two reasons:

1.  In some cases the model’s ability to predict words may lead it to make statements that make grammatical “sense” but do not reflect any underlying reality. This may happen especially when the model is asked to provide unusual information that was not frequently repeated in its training data. For example, asked for an overview of the Alaskan salmon harvest, ChatGPT will provide a generally accurate summary. However, asked for the number of salmon harvested in Alaska for any particular year, ChatGPT will generate a plausible, but incorrect, number. This failure mode is sometimes called hallucination. All known generative AI tools are prone to this kind of failure, to varying degrees. 
2.  In other cases, the model may faithfully report an inaccurate piece of information in an underlying source its been asked to summarize or work with. This is particularly acute in models that use what’s called Retrieval Augmented Generation (RAG) as a way to limit hallucination by tying a model to human authored sources. In one famous example, Google’s AI search summary told users to add glue to pizza sauce based on a joke post to the Q&A site Quora that turned up in its search results. 

Because of these potential sources of inaccuracy, we suggest that users of AI tools do the following:

1. Do NOT rely on AI for answers in fields you do not know well. You are less likely to be able to catch factual errors in new information.
2. Verify any factual statements in AI output. Be aware that the fact that an AI tool is using confident sounding language does not mean it is being accurate. AI can introduce inaccuracies with a great deal of confidence.  Always check any factual statement in any final document to ensure its accuracy. Do not take the AI’s word for it.
3. Do NOT use AI as a kind of information search tool (like Google) if the AI tool you are using does not include a list of underlying sources. It will summarize for you by default. Some tools will provide a list of “sources” if asked for one, but this source list may not be accurate. In some cases, the sources “cited” may not even exist! Find and evaluate underlying sources.

In addition to introducing inaccuracy , users should be aware that AI tools can introduce bias. This bias is a result of underlying biases in the information the AI tools were trained on. Since we rarely know the full scope of training data or how it was processed and selected, it can be difficult to say for sure what kinds of bias might be produced by any particular tool. 

However, we know from prior research that AI tools will often reproduce biased associations based on gender, race, or similar markers. For example, multiple studies have shown that tools like ChatGPT tend to associate the role of “doctor” with men and “nurse” with women. Image producing tools may also reproduce similar forms of bias, depicting people in positions of power as white and male more frequently. Cultural stereotypes are more likely to appear in AI generated content. An example, image generators asked to show "poor" people often conflate that with homelessness.  

Because of these potential sources of bias, we recommend that users of AI tools doing the following:

1. Confirm all claims with known, verified sources. Be especially skeptical of AI claims about identity-based categories like race, gender, etc.
2. Carefully read any copy or images produced by AI for potential bias, keeping in mind your own implicit bias as you do so
3. Use caution when prompting AI, since accidental or implicit bias in prompts may result in biased results and add to bias to the language learning models.

The bottom line is that you are accountable for all output based on the prompts you enter. As the user, you (and not the tool) will be held responsible for any inaccuracies or biases in the content created. Consider the following scenarios:

• A student uses Generative AI to write an essay in which the facts are wrong and/or the sources are made up. This presents a lack of rigor that the student is accountable for.
• A faculty member drafts an email with Generative AI, containing false information, and sends it without review. The faculty member is responsible for sending out false information. As a matter of good practice, it is advisable to review information that a third party has written before you hit send.
• A staff member uses Generative AI to generate an image for an upcoming career fair. Some peers and students find the image offensive. The staff member is accountable for the content of their work.
• A staff member uses Generative AI to generate an image for an upcoming career fair. The image recreates the signature style of a well-known photographer. The staff would face risk of being accused of appropriating someone else’s intellectual property. If AI-generated art is needed, the staff member should provide their own creative direction rather than mimicking an existing artist.

Being transparent about your use of AI is important because it

• develops and maintains trust and credibility for all stakeholders
• maintains the academic tradition of showing your work through citing sources of ideas, providing credit to those who created artistic elements, and conveying information in ways that can be replicated by others
• allows for the protection of individual’s privacy and copyrighted work
• allows others to engage an item knowing that potential AI mistakes or biases could be present

In addition to being transparent about AI use, this framework recommends those in positions of authority (faculty in a classroom, administrators) to be transparent in their expectations for AI use in settings for which they are responsible to prevent mis-understanding or improper use. For example, faculty should be transparent about their expectations for AI use in the classroom to prevent students from getting into academic trouble if they use AI inappropriately. See sample generative AI syllabus statements for examples.

Here is what AI transparency might look like for:

• Students
  • Students communicate the ways they engage with and use AI tools in support of their learning process and any products they create.
  • Students are aware of how and for what purpose AI might be used with their information that is maintained by the institution. For example, advising records and transcripts.
• Faculty
  • Teaching
    • Transparency of the justification for AI use in a course. A clear, easy to understand syllabus statement on if and/or when students may use AI in their coursework.
    • Transparent in how they are using AI to support the work of teaching. Do faculty indicate where AI was used to generate questions, assist with grading and feedback, or influence teaching decisions?
  • Research
    • Clear statement by the IRB on how AI may be used and/or disclosed when conducting research.
    • Clarity on how AI may be used or disclosed in the promotion and publication of research.
    • Faculty are open about their use of AI in support of research and manuscript development.
  • Service
    • Transparency on how AI may or may not be used in completing tasks.
• Staff
  • transparency on how staff use AI tools and functionality to complete the work of the institution. For example, AI use in relation to employee data, for communication, for monitoring, or for managing support requests.
  • Transparency on how administrators will use AI tools and functionality at the institutional level to track, monitor, and analyze data/services

It is your decision (except where it is a requirement for a course or job) to disclose use of AI. If you do decide to do so, we recommend the following:

• Share the name of the tool or tools used, when the AI was used, and what the AI was used to do. If you can link to the record of your use or provide the outputs as an appendix or addendum, do so. See examples of how to communicate AI usage transparently.
• If required, cite your use of AI. Examples of Formal Citation Standards for AI.

We recommend that faculty using AI do the following:

• Communicate expectations and the reasons behind decisions for transparency. For example:
  • A faculty member explains to students that they will use AI to create content for class while prohibiting students from using AI in class work using a justification that they have credentialed expertise in the field whereas students are still working towards that goal.
  • A faculty member might create classroom rules about how and when the technology can be used. Those rules are communicated frequently and clearly. There is also an opportunity to co-create these rules with students.
  • A faculty member may prohibit AI in the classroom completely. The instructor will not use it, and neither with the students. Faculty may articulate that the environmental concerns justify no one using the tools.
• Communicate, where appropriate, encouragement and patience as students have received mixed messages about AI use. Furthermore, since a solid and agreed-upon K-12 AI literacy foundation has yet to be established, these messages might come from parents, former teachers, society, friends, the tech industry etc. etc. Our students did not ask to be a college student during this transition.
• Avoid blanket statements regarding intent in terms of AI use such as “AI is cheating” or “AI is great and transformative” or “AI is bad.” The world is facing mixed messages about AI. Blanket statements risk shutting down conversations when we need to keep the lines of communication between students, faculty and staff wide open. Many aspects of AI are incredibly nuanced and deserve thoughtful consideration.
• Avoid making comparison about AI use in other disciplines to justify AI polices or usage. For example, statements such as “AI might be a good fit for this department/discipline but not for us,” muddies the waters during an already complex time. Instead, we encourage faculty to select and feel confident in their strategy and communicate that strategy loudly and proudly. Giving students more chances to encounter differences in decision-making and strategy makes us stronger. Yes, we will have to “experience” increased student complaints and frustration…but we can do it.