LibGuides: How Tos: Using AI

What is generative AI?

Overview:

The purpose of this guide is to:

provide content on the development, funding, and function of generative AI tools
highlight major ethical issues that surround the creation and use of generative AI tools
help users think critically and creatively about using these tools responsibly in their academic and professional work as well as within society in general
provide resources for learning more about the highlighted aspects of AI tools
provide some frameworks for using chatbots like ChatGPT effectively and responsibly

What do we mean when we say Generative AI?

Generative AI tools are a new set of AI tools that generate text based on a provided prompt or command. They "learn" from vast amounts of training data pulled from across the internet in order to provide text, image, video, code, and/or other type of medium as a "response" to an inquiry. Often in the form of a chatbot, the tools can "interact" with a user as the user asks them one or more questions, but there are many types of generative AI tools that provide different types of content for different purposes. The idea of the tools "learning" is that they should improve their text/media prediction abilities the more they "learn."

What are some examples of generative AI tools? Which companies created which ones?

ChatGPT, DALL-E, owned by OpenAI
Copilot, owned by Microsoft
Gemini, owned by Google
Claude, owned by Anthropic
And many more

Definitions:

Artificial Intelligence (AI) - The theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages

Generative AI - “A machine learning model that is trained to create new data, rather than making a prediction about a specific dataset. A generative AI system is one that learns to generate more objects that look like the data it was trained on.” from Explained: Generative AI

Large Language Model (LLM) - “A category of foundation models, trained on immense amounts of data making them capable of understanding and generating natural language and other types of content to perform a wide range of tasks.” from What are Large Language Models

Machine Learning - The use and development of computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze and draw inferences from patterns in the data.

Natural Language Processor (NLP) - “A subfield of computer science and artificial intelligence that uses machine learning to enable computers to understand and communicate with human language.” from What is NLP (Natural Language Processing)

Algorithm - A process or set of rules to be followed in calculations or other problem-solving operations especially by a computer.

Neural Network - A computer system modeled on the human brain and nervous system.

Learn more:

Ethical Issues Surrounding AI

Overview

Understanding ethical issues surrounding AI is essential, as it is for any technology, in order to understand how it can be used responsibly and for the benefit of our communities.

Ethical Issues Surrounding AI

May Be Inaccurate

Generative AI tools might provide different responses to the same prompt, and one or more of those responses might not be accurate.
Training data used by the tools includes data that is wrong or made up, and the tools can reproduce this data in their responses.
Generative AI tools don’t have the capacity yet to fully differentiate between data that is correct and true and data that is incorrect or false.
While progress is being made, generative AI tools can still produce hallucinated sources, or sources that seem real but are not.

“Hallucinations: Why AI Makes Stuff Up, and What’s Being Done About it” - CNet With all the progress made in making AI tools accurate and able to respond to any prompt it gets, there are still times when the tools provide inaccurate or made-up information. The reason this occurs is because the tools cannot distinguish between factually true and not factually true data in their training models. So much data is incorporated into the training models, and that includes data that is wrong, misleading, or false. Tech companies are aware of this issue and are working to improve the tools, but it’s still occurring.
“Lawyers Blame ChatGPT for Tricking them into Citing Bogus Case Law” - Associated Press In a strong example of the consequences of using AI tools without verifying the information provided, two lawyers faced punishment for filing court documents that contained hallucinated court cases. The lawyers used ChatGPT to create their legal brief, which included references to fictitious court cases.

Bias

AI tools include existing biased information and so information produced may also include this bias.
Since AI tools are created, managed, and trained by humans living and participating in systems that are built on bias and oppression, those biases get incorporated into the way those tools work and learn.
Being aware of the ways in which bias is incorporated into the output of generative AI tools can help users better identify biases in the responses gathered from AI tools and use the tools in a more intentional way to reduce the prevalence of bias in gen AI tools

How AI Could Perpetuate Racism, Sexism, and Other Biases in Society - NPR Machine learning, which is one way AI tools like ChatGPT “learn,” uses a large dataset, and uses algorithms and statistical models to identify patterns. When the algorithms and models are created by humans who have biases (explicit or implicit), those biases can show up in the AI tool in different ways.
Bias in Algorithms: Artificial Intelligence and Discrimination - European Union Agency for Fundamental Rights Data and parameters used to train AI may be biased and result in biased output. Users’ prompts affect AI output; users entering biased prompts may result in biased output. Censorship may also exist within AI tools.
Yale Study Finds Racial Bias in ChatGPT's Radiology Reports - New Hampshire Public Radio Systemic bias prevalent in our medical system means that the use of AI tools in healthcare could lead to biased medical data and recommendations.
When AI Gets it Wrong: Addressing AI Hallucinations and Bias - MIT Sloan Technology Services Teaching & Learning A 2020 study of 5,000 images created with the gen AI tool Stable Diffusion found that the tool amplified both gender and racial stereotypes, sometimes simultaneously.

Privacy and Data Collection

Training generative AI tools like ChatGPT includes gathering large amounts of data from the available web, and sometimes data from parts of the web that might include sensitive personal information.
Because the tools are proprietary, their methods of data storage and management are not known beyond what companies share - leaving users in the dark about how protected that data is.
Users should be aware that any data they input into a tool like ChatGPT is probably being stored and used by the tools, and could be reproduced by the tools given the right prompts.

Privacy Considerations for Generative AI - University of Illinois Urbana-Champaign Sharing high-risk, sensitive or internal data (intentionally or unintentionally) with generative AI tools can have unintended consequences. The tools operate on the data they receive, and since we’re not aware of how the tools are storing data and how protected that data is, users should avoid sharing any private information with any AI tools.
Your Personal Information Is Probably Being Used to Train Generative AI Models - Scientific American In the push to produce larger and more responsive generative AI tools, developers are integrating training data from much of the searchable internet. The methods of this process include using web crawlers, some of which are able to get behind internet paywalls and other internet sources that are not necessarily meant for wide-spread public use, and probably pulling your personal information in the process. This is important because what the tools incorporate, they also regurgitate, so there is potential for your personal information to be reproduced by a generative AI tool.

Context, Critical Thinking, and Creativity

Generative AI are not information tools, but rather word completion tools.
They will provide text, images, code, or other media based on exactly what you input, so if your prompts lack detail and depth, so will the responses from chatbot tools.
Be aware of the shortcomings of the tools as you use them, so that you can mitigate those shortcomings or use information sources that will provide you with better information based on your needs.

"AI Isn't Dumb, But it Might be Dumber Than You Think" - Washington Post The release of flashy new AI tools came with promises of new ways to interact with information, new ways to assist you in mundane tasks, but since many of the tool rollouts, the actual experiences of interacting with the tools is a bit lackluster in meeting those promises. Keep in mind that the tools are still adapting and changing, and keep your expectations muted for what tools like ChatGPT can actually do for you.
“AI Can’t Think Critically, but People Must” - Economist Education AI tools can’t “think” - they can only produce based on the prompts or commands we provide them, whereas true thinking is a process that includes elements unique to humans, like purpose. We think based on our goals, desires and needs, and AI tools do not yet have that capability. There’s no AI tool that can critically engage with the data it learns.

Transparency

Because major companies own the AI tools outright, there is no obligation to share information about how they manage, store, or use the data they collect from their AI tools.
Corporations have the goals of their shareholders as the bottom line as opposed to the public good, so decisions made about updates and changes to the tools or the data the tools are trained on may affect users in ways that are unexpected or unwelcome.

It’s High Time for More AI Transparency - Technology Review The most powerful AI tools on the market (e.g. ChatGPT) are closely guarded by the companies that own them. More transparency in the inner workings of AI Tools that are more and more embedded into our technology tools is important moving forward to build trust in tools’ uses and applications, as well as in the tools’ ability to mitigate future issues that could have big impacts on society.
“A Call for AI Data Transparency” - University of Washington Electrical and Computer Engineering News Advocating for greater transparency in AI tools is imperative as we integrate these tools into more and more of the technology we use in our day to day lives. Two tech entrepreneurs and researchers created an advocacy group, Transparency Coalition that calls on tech companies to be more transparent in sharing data about the development of AI tools, the training data they use, and more.

Environmental Impact

Water usage for cooling large data centers used by companies like Google and Microsoft rose with the implementation of their generative AI tools. In areas where water usage is verging on crisis, that high usage could be a major issue.
The same data centers also significantly contribute to the CO₂ emissions in the atmosphere. According to researchers, the electricity usage of these data centers will only grow over the next years.

The Growing Environmental Footprint Of Generative AI - Undark The companies that manage AI tools like Gemini and Copilot are using more water and producing more emissions with the implementation of their generative AI tools.
AI’s Climate Impact Goes beyond Its Emissions - Scientific American High CO2 production and water usage AI companies sell to companies who use it to increase their negative impact on the environment

Copyright

Uses others’ intellectual property without attribution or permission
May not accurately reproduce others’ intellectual property
Several major cases will decide if generative AI tools infringe on the copyright of creative works; these will have implications for creators and users of AI tools alike.

Generative AI Could Leave Users Holding the Bag for Copyright Violations - The Conversation The way AI tools are trained on existing data, as well as how they use that data to ‘learn’ does not result in violations of copyright law, but the ways in which their users interact with the tools might lead to copyright issues. For example, users can ask prompts that lead to the creation of a response that is strikingly similar to a copyrighted work. It is important to consider what guidelines might alleviate issues with copyright and ownership of data produced by generative AI tools.
Generative AI Has an Intellectual Property Problem - Harvard Business Review AI tools that produce generated text and/or images from their training data pulled from the internet may be infringing on the copyright protections of established creative works. Currently, the courts are working out several major issues related to copyright and ownership of the content being produced by generative AI tools. Several cases will have major implications for copyright as it relates to the content created by AI tools.

Labor Practices

Labor used to train and categorize training data used by AI tools is often forgotten, since the tools are typically viewed as machinated; this leads to an underappreciated work force that is elemental in the AI tools’ development
People doing the work of categorizing and organizing training data may not be paid appropriately for the work they’re doing, and the conditions that they work under are draining and exploitative.

Behind the AI Boom, an Army of Overseas Workers in 'Digital Sweatshops' - The Washington Post While AI is often thought of as human-free machine learning, the technology actually relies on the labor-intensive efforts of a workforce spread across much of the Global South and often subject to exploitation.
‘It’s Destroyed Me Completely’: Kenyan Moderators Decry Toll of Training of AI Models - The Guardian Employees describe the psychological trauma of reading and viewing graphic content, low pay and abrupt dismissals.

How to Use AI

Prompt engineering

The prompts that you enter affect responses you receive from generative AI. Use the following guidelines to help make the AI output closer to meeting your goals.

Role: Describe its role or your role in the assignment. “Act as a job applicant.” “I am applying for a position as a bookkeeper.”
Task: Summarize the function you want the it to complete. “Create a cover letter.”
Requirements: Include desired elements in the output. “Highlight my three years previous experience as a bookkeeper.”
Context or Constraints: Explain the intended audience or define elements that should not be included. “I am applying for this position at a charitable organization.”
Goal: Establish the why of the output, meaning what the reader should be able to do or understand at the end. “Highlight my technical skills with QuickBooks and Excel.”
Format of Output: Provide a word or page limit or formatting specifications. “Make it 1 page long.” (Mowreader, 2023)
Neutral Tone: Make your questions neutral and avoid leading questions since your questions influence the response. Avoid questions like, “Why is X the most effective treatment for X?” Ask more open-ended questions like,” What effective treatments for X?”

Privacy Considerations

Data that you enter into generative AI is usually used to train the AI and may be used in future generative AI output for other users. Take this into consideration when entering personal information into generative AI tools.

AI Policies

You are responsible for know the AI policies of the organizations that you are a part of.

Education: University policy: Cite AI if used; Program Policy; Course Policy: Check syllabi. Never enter student information into an AI unless it has been approved as being FERPA compliant by your institution.
Scholarly publishing: COPE Authorship and AI Tools: AI cannot be listed as an author; AI use must be included in methods.
Clinic/Work: Follow your organization’s policies to ensure data security and privacy. Never enter patient information unless AI has been approved as being HIPAA compliant by your organization.

Evaluate AI Tools

Who created/owns it? Is it for profit? Open source?
What is its purpose? What does it do?
How does it work? What information is it using? How is it using this information to provide its output?
How accurate is it?
What possible bias does it operate under?
How does it affect people and the environment? What ethical issues does it bring with it?
Based upon these things, how can it be used responsibly and ethically in my current context?

Recommend finding peer-reviewed product review articles of AI tools you're considering using.

Evaluate AI Output

What are the sources of information?
Did the AI accurately represent the information in these sources?
How current, relevant, and credible are the sources it used?
How does the AI output compare to the body of literature on the topic?
What is missing from the AI output? What perspectives are missing?
Is there bias apparent in the AI output?
How in-depth is the AI output for your information need?

Creative Uses of Generative AI

Consider creative ways that you can use AI ethically and responsibly
Take part in conversations with your peers and colleagues about creative uses and ethical issues in your discipline
Keep learning about AI as it evolves
Test and evaluate new uses

AI Tools to Try and Evaluate

Pacific University AI Tools FAQ: Learn about the AI tools available to the Pacific community, including Gemini, Box AI, and Zoom AI Companion.

Note that some of these tools will ask you to create an account, even when using a free version of the tool.

Goblin Tools: This tool helps you break down tasks into smaller, easier-to-accomplish segments. For students looking to get help planning how to accomplish big research papers or even just day-to-day tasks, this tool might help break down large, overwhelming projects into smaller, easier to manage chunks.
Teaching Tools: Specifically made for teachers, or those creating teaching materials, this tool helps brainstorm lesson plan ideas, classroom activities and more.
Chat GPT
Copilot
Gemini
Perplexity: An alternative option for anyone looking for a chatbot similar to ChatGPT and Copilot.

Tools for Searching and Summarizing

We currently do not recommend using non-library AI tools such as Elicit or Connected Papers for searching for your research. The library search box and databases are currently the most effective tools to find relevant, quality sources for your research. Some of these databases, such as PubMed, do have AI integrated effectively into their search features. Learn how to search databases effectively.
We also do not recommend using AI tools to summarize articles since AI tools make mistakes in their summaries. Instead, read the abstracts or descriptions provided by the authors and see Read the Sources You Found to learn how to read efficiently.

Learn More

How Large Language Models Work
Ideas for Experimenting with Generative AI: Use Cases and Things to Keep in Mind
Generative AI Prompt Engineering Tutorial: Learn how to create effective AI prompts.

How to Cite AI Generated Content

How to Cite AI-Generated Content - Purdue University
Learn how to cite AI generated content in APA, MLA, and Chicago styles.

Consider ethical, responsible, and creative uses for artificial intelligence (AI).

Academics: According to Pacific University policy, you need to cite AI if you use it. Additionally, make sure to check any AI policies for college, program, or syllabi.
Scholarly publishing: COPE Authorship and AI Tools: AI cannot be listed as an author and you must include it in your methods section for any research studies.
Clinic/Work: Follow your organization’s policies to ensure data security and privacy. For healthcare, never enter patient information unless AI has been approved as being HIPAA compliant by your organization.