Skip to Main Content

Artificial Intelligence : AI Tools

This guide provides an introduction to artificial intelligence including definitions, copyright issues, and generative AI tools.

Evaluating AI tools and Output

Before and when using an artificial intelligence tool it is important to evaluate the tool and the tool's output. Below are some questions to think about: 

  • Consider the biases present in the training data of the AI tool. AI models are prone to groupthink bias, caused by a general consensus of views found in the training data rather than presenting a diversity of viewpoints. 
  • What is the purpose of the tool?
  • Are there any ethical concerns? 
  • Are there copyright concerns you should be aware of? Is there a way for you to opt out of the tool using your data in its' training data? 
  • If there is any evidence cited, is it real or just a "hallucination"? 

What can AI be used for in academic research?

AI can ASSIST you with your research but should not be doing ALL the research/work for you. AI tools can generate ideas, offer feedback, provide formatting assistance. You still need to verify sources by using lateral reading. See the B.E.E libguide for more information on this. 

Benefits

  • Brainstorm- AI tools can help brainstorm ideas for research topics, keywords for database searching, or help tackle writer's block
  • Break down concepts- AI tools may be able to help break down complex concepts or topics

Limitations for AI in Research

  • AI tools hallucinate often, by making up information such as a citation to a scholarly source that doesn't exist
  • Many AI tools don't have access to scholarly work that is behind a paywall. These resources, however, are available through the library databases. 

 

AI Tools & Features

ChatGPT

  • Basic version is free. That version uses only training data and is trained up to Oct. 2023
  • ChatGPT Plus (GPT4) which is subscription based, uses web search results along with their data. 
  • Don't enter any personal or private data. The tool may use your input data to improve the LLM. In some AI tools you can change the settings to limit their access to your information. 

Microsoft Copilot

  • Basic version is free with a Microsoft account. 
  • Copilot generates text, code, and images using ChatGPT and DALL-E, which is a series of AI models developed by Open AI.  
  • Data collected from Copilot is made anonymous so that an individual can't be identified, however, still be incredibility careful with what you are inputting into the tool. 

Google Gemini

  • Basic version is free but requires a Google account. 
  • Gemini generates text, code and images and includes links to some web sources. 
  • Gemini uses location and past conversation data to provide responses. Data will be saved and shared with other Google products if linked. 

Notebook LM

  • Basic version is free but requires a Google account
  • Uses Gemini to generate summaries and explanations of the content that was uploaded by the user
  • Can transform the uploaded documents into a podcast 
  • Uses only the information from the material that you upload. The files and are not visible to other users and data is not used to train Notebook LM. However, make sure that you aren't uploading documents that you don't have the rights to. 

 

Prompting

Prompting is where you question or give a task to an AI system to evoke a response or output. The prompt you input affects the output the AI tool gives you. 

Tips for writing effective prompts: 

- Give the tool context or a role to play. For example, "Act as an expert in ....

- Give detailed instructions 

- Ask it to revise the answer 

OpenAI has created a guide to assist users of their products with prompt engineering

Best practices for prompt engineering with the OpenAI API

Frameworks are being developed to specifically help us form prompts that get the best possible result. 

PROMPT Design Framework by Sarah Hartman-Caverly, Librarian Penn State Berks

  • Persona - assign a role
  • Requirements - define parameters for output 
  • Organization - describe the structure of the output
  • Medium - describe the format of the output
  • Purpose - identify the rhetorical purpose and intended audience
  • Tone - specify the tone of output (ex: academic)

 

Ethical Concerns & Bias

The AI output that is generated is reliant on the data that the system was trained on. AI uses a enormous amount of data from the internet to train. Using this information means generative AI can replicate the biases, stereotypes, and hate speech found online. As with any research article you find, it is important to look for any bias within the work. 

AI Fact-Checking

University of Maryland Libraries