Whether you are a general user of this Guide or part of the UB Community, please practice the policies your institution has set to maintain academic integrity when using Artificial Tools.

If you are affiliated with the University at Buffalo, please familiarize yourself with the policies set in place as to how to use the tool appropriately as a student and as an instructor. More information for instructors can be found on Artificial Intelligence Guidance, or for students at Artificial Intelligence from the Office of Academic Integrity.

Bias is defined as "prejudice in favor of or against one thing, person, or group compared with another, usually in a way considered to be unfair."

Artificial Intelligence can have possible biases that it may produce in the instance of only providing information that is the most recent, has the most peer reviews, or provides information specific to only one demographic. 

Controlling for biases is difficult to manage unless you know what type of material you are gathering and appropriately include your target population within your research. Additional bias can accidentally be produced if not enough research is done to thoroughly review the evidence and not find counterclaims to the question at hand.

Generative AI is known to produce bias and users should be aware that not all data is presented equally.

Further Readings

• Bias in AI - Chapman University
• There’s More to AI Bias Than Biased Data - National Institute of Standards and Technology
• About Ethics, Bias, and Transparency for People and Machines - National Institutes of Health
• Machines and Society - New York University Libraries

It is difficult to have an AI tool to operate with 100% accuracy. With some AI models, it is found that they can produce false responses (hallucinations), which can then lead to disinformation, skewed research results, and creating answers that can potentially harm users.

It is important to read forums produced by the AI model company that openly recognize that their model will generate inaccurate results or produce bias and how to avoid these issues. An additional course of action is to locate the generated sources in the journal in which they are published or through a website search.

At UB, you may be able to locate many of these sources through the University at Buffalo Libraries or Google Scholar.

Further Readings

• University at Buffalo Information Technology (UBIT) Guidance: Using Generative Artificial Intelligence (AI) Technology at UB
• How to Check the Reliability of Artificial Intelligence Solutions—Ensuring Client Expectations are Met - Dr. John Patrick 
• Contextualizing End-User Needs: How to Measure the Trustworthiness of an AI System - Carnegie Mellon University
• Generative AI Reliability and Authority - University of South Florida
• What is hallucination - University of Arizona Libraries
• Hallucinations (Artificial Intelligence) 

Most, if not all, AI models require an account to use their site even if it is free. Some sites allow you to use preexisting accounts you may already have such as Google, Microsoft, and Apple accounts. The sites you allow access to your preexisting accounts could potentially fail if exposed to cyber security attacks harvesting your personal information. Using a burner account or alias does not provide a barrier to your identity when performing academic research. Using a VPN provides a buffer of security that allows you to stay safe while on the web and downloading resources and is highly recommended for all computers. 

To access a free VPN download through UBIT click here (you must be an affiliate UB to download software from UBIT). 

U.S. Copyright law balances the interests of authors who have created works with the interests of users of those works. The purpose of copyright is "to promote the progress of science and the useful arts," while also serving as a form of protection provided to authors of "original works of authorship."

This balance is maintained by the provision of certain rights to authors – such as the right to reproduce, distribute, or prepare derivatives of the work – and the provision of some exceptions and limitations to those entitlements.

The U.S. Copyright Act can be found at Title 17 of the United States Code. The U.S. Copyright Office also provides information on copyright.

Learn More at the Library Guide below.

Microsoft Co-Pilot

Microsoft Copilot, an AI-powered chatbot grounded in up-to-date information, can help you find answers and create content. Copilot features the power of ChatGPT-4 with commercial data protection from Microsoft.

Artificial Intelligence Library Guide

The Guide provides a comprehensive introduction to AI, including its history, current state, and future possibilities. The Guide includes a variety of resources, including books, articles, videos, and websites, to help you understand the fundamentals of AI and its impact on society.

• Generative AI chatbots are designed to be plausible rather than credible. They often hallucinate, which means they make up information, including citations to non-existent scholarly sources.
• Generative AI chatbots do not have access to and have not been trained on the vast majority of scholarly materials which are behind a paywall (available via the Libraries subscription databases). Although this is beginning to change with some scholarly publishers selling access to journal contents to train large language models,the majority of scholarly information is not available to train models.
• Because generative AI chatbots create original responses to prompts, the information they provide is not reproducible. In addition, you may find that multiple users supplying the same prompt to a chatbot at the same time will produce different responses.

The content in this Guide was adapted from content available from the University of Arizona Libraries and the University of Texas Libraries.