Important Note: At UMW, each instructor sets their own policy on the use of AI. Make sure to check the syllabus or ask your instructor before using AI tools for classwork.
What is Generative AI? #
Generative Artificial Intelligence (AI) is a type of AI that can create new content, whether it’s writing a story, composing music, or generating images. Generative AI tools are trained to recognize patterns and styles in large amounts of existing data, and then produce original creations that use similar patterns and styles.
It’s important to understand that generative AI tools don’t actually “think” – in fact, they don’t “know” anything at all. They simply attempt to generate content that follows the patterns identified in their training data. For example, when given a prompt, a text AI (like ChatGPT) searches its training data and generates words that are grouped together in ways similar to articles, web searches, and other texts related to the prompt’s subject.
Think of generative AI like a person who doesn’t know Latin using a phonetic guide to read aloud from a Latin text. The guide allows the person to speak the words correctly (for the most part), but the speaker has no idea what the words actually mean, and so they have no way to verify that the information they are speaking is accurate.
Similarly, AI tools can usually generate responses that sound (and often are) correct. But because the tools don’t actually think, they can easily generate responses that are false, outdated, or nonsensical. These are called “hallucinations,” and they are one of the many reasons that AI results must be fact-checked before use.
What AI Tools are Available? #
There are countless AI tools on the market, with new ones being launched (and existing tools getting retired) almost every day. With the environment changing so rapidly, it is difficult to make recommendations that aren’t immediately outdated. If you are thinking about signing up for an unfamiliar AI tool, consider the following:
- Research the tool thoroughly by searching for what users, critics, and reputable publications have to say about it
- Be very wary of new tools, and any that require a subscription or paid “tokens” to use
- Be aware that things you create on AI platforms may not last forever. Consider backing up anything you want to keep outside of the AI platform
- Never share personal or sensitive information with an AI tool
Ithaka S+R’s AI Product Tracker is a good place to start if you are searching for an AI tool for a specific purpose.
The Simpson Library also has a Artificial Intelligence (AI) for Library Research guide that can help you understand the tools available to you for research.
Microsoft Copilot #
Microsoft Copilot (formerly Bing Copilot or Bing Chatbot) uses Open AI’s GPT-3, GPT-4, and DALL-E 3 (the same AI models used in ChatGPT) to generate text and images from written prompts. It is available for free to all UMW students, faculty, and staff using your UMW Microsoft 365 account. More information on accessing and using Copilot can be found in our Microsoft Copilot guide.
Copilot is a popular and (relatively) established tool that is available to all UMW users, so it is a great place to get started experimenting with AI.
What can I do with AI tools? #
There are many AI tools that can do a variety of different things, from text generation to image and even video creation. This guide focuses primarily on the “AI Assistant” suite of AI tools. With tools like this, you can do things like:
- Brainstorm topics for a research paper
- Ask AI to rephrase an explanation of a class concept that you didn’t understand during a lecture
- Get critique on your draft of a writing assignment or oral presentation
With more advanced tools, you might be able to do things like:
- Train an AI tool to answer questions using a specific body of work (e.g. answer questions on UMW policy using the official policy handbook)
Be sure to check your course’s AI policies before using AI for assignments.
How do I use AI Ethically? #
Cheating with AI #
AI tools can be a great way to brainstorm, speed up tedious processes, and overcome “blank page syndrome” to get started on projects, but they can also be used in ways that shortcut the learning process. These methods might generate a correct answer, but leave you without the skills necessary to accomplish tasks unassisted in the future. This includes things like:
- Copying and pasting an AI response for a writing assignment that was designed to help you practice your writing skills
- Submitting AI generated code for a computer science project that was intended to help you understand how the code works
- Generating an AI image for a digital art project that was meant to help you learn Photoshop
Keep in mind that each course at UMW can have different AI policies depending on the learning goals of the class. The instructor and the course syllabus are the best place to go to understand what tools and uses are acceptable in any given course. Inappropriate use of AI may be considered a violation of the UMW honor code.
The Simpson Library has created a guide on how to cite AI generated content that may aide you when documenting how you used AI to assist in your course work.
Other Ethical Considerations #
Since AI tools are new and changing rapidly, there are many ethical concerns that are highly debated. Below are a few to consider.
- Intellectual property
- AI models are trained using creative works (writing, visual art, music, spoken word, etc.) that were often gathered without the creator’s consent.
- It is unclear who owns the copyright to works created wholly or partially by AI tools.
- Privacy
- AI companies can harvest the prompts and other data that users enter into their AI tools.
- Bias
- AI models can, unintentionally or intentionally, be trained on materials that have innate bias. For example:
- Some versions of ChatGPT assumed doctors were male by default.
- When Google Gemini was first launched, it was trained to prioritize people of color. When asked to generate images of the US founding fathers, it refused to show them as white men, instead depicting them as people of color.
- AI models can, unintentionally or intentionally, be trained on materials that have innate bias. For example:
- Environmental Impact
- AI tools take a tremendous amount of computing power to operate, leading to an increase in power consumption as well as carbon emissions.
- Jobs Impact
- Many jobs can be accomplished with fewer people, or eliminated entirely, when tasks are outsourced to AI.
- Shortcutting learning
- AI can generate answers quickly without requiring a user to understand how to get to that answer themselves.
- Misinformation/Disinformation
- Generative AI makes it very easy to generate false information, both intentionally and unintentionally.
How do I Even AI? #
Prompt engineering – writing effective requests that AI tools can use to generate the responses you want – can take a bit of practice. Each tool has its own quirks, but there are some techniques that are effective across most of the current AI platforms.
Be specific in your prompts #
Specific prompts get better results than general ones. It can also help to avoid language that is subjective or can be misinterpreted.
Weak Prompt: “Describe UMW”
Stronger Prompt: “Give me three pros and three cons to attending the University of Mary Washington”
Weak Prompt: “What are some good things to write about in a history class”
Stronger Prompt: “What are some common research topics about nurses in World War II”
Iterate on the results #
The first response from an AI tool is just a starting point. Often you will need to refine your prompt further to get the results you want. Giving background or contextual information can help tailor the response to your needs.
Prompt 1: “Give me three pros and three cons to attending the University of Mary Washington”
Prompt 2: “Give me three pros and three cons to attending the University of Mary Washington as a biology major. Compare the pros and cons to Christopher Newport University.”
Fact-check responses #
AI does not know whether it is giving you accurate information or not. Often responses will sound correct, but have no factual basis. Verify every response before sharing or using it in an assignment.
Additional Resources #
Prompt Engineering Guide by DAIR.AI
Ithaka S+R’s AI Product Tracker
University of Sydney’s AI in Education Canvas course