University of Colorado Anschutz Medical Campus

AI Literacy & Resources

There are a lot of Artificial Intelligence (AI) tools available to try, and using AI comes with its own learning curve just as using any other new software or application. AI literacy empowers individuals with the knowledge and skills to understand, use and evaluate AI technologies responsibly and effectively.

Learning a new tool or application, in this case AI, means needing to learn new terminology for using that tool, how it works, and the capabilities and limitations of that tool. Below are some key concepts to help you get started on your AI journey.

Understanding AI

Basic AI Terminology:

AI Term	Definition	Use
Artificial Intelligence (AI)	Technology that enables machines to perform tasks that typically require human intelligence, like understanding language or making decisions.	AI has been in use for years. It is the chatbot on websites asking how they can help, it's the GPS on your phone, Siri and Alexa are common household AI names. AI has taken a recent leap forward with the development of Generative AI.
Prompts	Prompts are entered into the AI application to provide instruction or direction on what the user would like the AI to do. Crafting your prompt with specific directions will improve the results created by the AI to more accurately match what you are needing.	The more details and information you can provide the AI in your prompt the better the result will be. Details include size, shape, length, format, specific focuses, etc.
Algorithm	A set of rules used to organize or analyze data.	Algorithms are used by AI to review data sets, organize and analyze the information to pull what is most relevant to your prompt base on what you entered.
Application Programming Interface (API)	API is a set of protocols that determine how two or more applications will interact.	For AI, this is how the AI application (such as Zoom AI or Microsoft Copilot 365) will interact with other applications such as Zoom, Outlook, Teams, etc. to provide meeting transcriptions, calendar reminders, meeting summaries, and more.
Chatbot	A type of AI application that imitates human conversation through text or voice commands.	Many help desks and websites utilize chatbots to answer questions wit pre-programmed solutions or identify a user's problem and sort them to the best department for assistance.
Machine Learning	The process of a computer system learning how to do something by identifying patterns in data that is repeatedly run through algorithms and making predictions based on the patterns it has identified in those data sets.	Machine learning is used to train AI applications to produce the results you are looking for when you enter a prompt.
Large Language Models (LLMs)	Large Language Models (LLMs) are a type of AI software. Not all AI applications are LLMs, but examples of LLMs include Chat GPT and Microsoft Copilot 365/Copilot Chat. LLMs are trained using machine learning to process language through a significant amount of text to learn patterns of human communication so that they can mimic it.	Large Language Models (LLMs) will use the data entered into it to improve upon its machine learning. This is why it is crucial that only a university-approved AI application is used with university data, as many LLMs will hold onto the data entered into it. Microsoft Copilot 365 and Copilot Chat do NOT retain the data entered into it and are secure to use.
Generative AI	Generative AI is a subtype of AI that uses LLMs to create new content. This is different from other types of AI because it doesn't regurgitate existing information, like the Google search AI, but takes what it learns from the LLM and creates something new and unique based on the prompts you enter.	Generative AI can be used to create images, text, documents, presentations, data sets, and more. Make sure you are using it securely and ONLY use university-approved AI applications.
Hallucinations	Generative AI systems can't tell the difference between real and fake information, so they can give inaccurate answers. These inaccurate responses are called "hallucinations." Hallucinations can show up in inaccurate statements or predictions.	It is best practice to always double check the results of generative AI to ensure accuracy and that there are no hallucinations. Developers and users can improve accuracy and reduce hallucinations by catching them and uploading additional information from a trusted source.
Multimodal models	Multimodal model AIs can work with different types or mediums of data simultaneously (images, text, audio, etc.) It will combine these multiple modes of information together to give a result.	Microsoft Copilot 365 and Copilot Chat are both multimodal. They can respond to various forms of input. If you upload a document, picture, and video all at once and ask it to provide a summary of the content, it will be able to do so.
Natural Language Processing (NLP)	NLP is a type of AI that understands the way humans write and speak.	NLP allows for speech recognition and is used by Zoom AI and Microsoft Copilot 365 for meeting transcripts and summaries.
Plugins	Plugins are additional add-on features that can fill specific needs with the AI without modifying the original AI model.	Plugins allow AI to interact with other software and applications such as Outlook, your calendar, and Teams.

Learn More:

How AI Works

Each type of AI application works a little differently as you may have learned from the AI Terminology section.

Type of AI	How it Works	Use
Machine Learning	The computer will review massive data sets and analyze it for patterns. The more data it interacts with, the more it can analyze and "learn". It then uses these patterns to develop predictions based on its pattern recognition.	Most AI tools use machine learning to improve and generate accurate results to prompts. This includes your GPS that can re-route you when you make a wrong turn, and generative AI when you ask it to summarize a data set.
Natural Language Processing (NLP)	The computer will analyze human speech and writing to understand and mimic the ways humans communicate.	NLP is used by AI to provide transcripts of meetings or videos, understand voice commands, and provide search results based on key words in a query you enter into a search engine.
Generative AI	The computer will review existing data and through machine learning create something new such as images, documents and presentations.	Generative AI is used to create new content. You can ask it to create test questions based off of learning outcomes, a PowerPoint based off of a document you upload, or generate an image based on a description you enter into the prompt.
AI Agents	The computer uses integrated data (data you feed it) and advanced analytics for complex reasoning that allows it to make informed decisions. They can be autonomous decisions or ones with human-guided oversight.	AI agents can automate tasks for greater efficiency, provide diagnostics based on a set of parameters you enter in, and perform repetitive tasks.

Learn More: SAS 5 AI Technologies You Need to Know

University-Approved AI Capabilities

AI is still a developing technology. While it has many features and capabilities, it also has limitations in what it can do.

University-Approved AI Tool

Capabilities

Limitations

Microsoft 365 Copilot Chat

(Formerly known as Bing Chat)

An AI powered web chat that can generate content, answer questions and improve productivity.

Can be used with confidential and highly confidential data.

Available and protected through your university Microsoft account. Be sure you see the "protected shield" messaging at the top of the webpage to confirm that you are signed in.

Copilot Chat has limitations on how many images it can generate in a day, a prompt limit per day, and document size limit. These limitations vary based on Microsoft updates, so the exact number is not listed.

While it can help with drafting emails or content, it cannot send that email or execute the actual task. You will still need to do that manually.

You can only enter up to around 2,000 characters per prompt, which is plenty for a typical prompt. You can also add to a previous prompt in a follow-up prompt if you want to add more detail.

Learn more from Microsoft.

Zoom AI Companion

A generative-AI digital assistant currently available for meetings, whiteboard and events. Visit Zoom support for more information.

Available through your university Zoom account.

Only the meeting host can start or stop the Zoom AI Companion during a meeting.

You should always stop the AI Companion when discussing HIPAA or highly confidential information.

Zoom AI Companion is not available in Breakout Rooms.

Need version 5.16.6 or higher for questions and responses in languages other than English.

Learn more from Zoom.

Microsoft 365 Copilot

An AI assistant for use across the Microsoft 365 suite of applications.

Can be used with confidential and highly confidential data.

Available with a license. Visit the webpage for purchasing information.

Microsoft Copilot 365 has application-specific limitations (can only work with Tables in Excel, page limits in Word for summaries, cannot connect to shared mail boxes in Teams and Outlook, etc.)

While it can help with drafting emails or content, it cannot send that email or execute the actual task. You will still need to do that manually.

You can only enter up to around 2,000 characters per prompt, which is plenty for a typical prompt. You can also add to a previous prompt in a follow-up prompt if you want to add more detail.

Limitations of Generative AI

While certain applications may have feature limitations, the technology itself is still evolving.

Limitation	Definition	Work Around
Non-Determinism	Generative AI will produce different results every time even if the exact same prompt is used. While fun an exciting for creative activities, this can be problematic for tasks requiring predictability and consistency.	Always fact-check the results of any prompt to ensure accuracy.
Hallucinations	As mentioned above, hallucinations are when a generative AI produces a clearly false statement or result but presents it as fact. Hallucinations happen because LLMs are trained by massive data sets, including information and data found on the internet. Not all data sets have been fact checked because of the enormity of their size. It is well known that not all the information found on the internet is accurate, but the AI has no way of differentiating fact from fallacy or misinformation.	Always fact-check the results of any prompt to ensure accuracy. If you discover a hallucination, correct the AI and either upload or reference a reliable source with the correct information. Then, try your prompt again. If you get another hallucination, revise your prompt to include specific and well-defined boundaries and constraints.
Limited Scope	Generative AI is great for administrative tasks and narrow prompts. However, it is limited in its scope and currently unable to process broad challenges like developing a strategy, decision making, or navigating ethical dilemmas. It does not currently have the capability to understand context or complex situations. It focuses mainly on causality (cause and effect) and pattern recognition.	Use it to support your work in these areas to make the administrative part easier. It can also help with brainstorming and idea generation, but you will need to piece it all together.

Learn More: Forbes Understanding the Limitations of Generative AI

How to Write AI Prompts

Learning how to write a good AI prompt will increase the accuracy of the result so that you are more likely to get what you are looking for.

Here are some prompt writing best practices.

Provide a clear direction or ask. This will ensure Microsoft Copilot 365 or Copilot Chat know what you are asking it to do.
Provide details and context to your direction. Specifics will allow it to give you a more tailored result.
Structure your prompts by putting the most important details or ask last. Microsoft Copilot 365 and Copilot Chat are more likely to emphasize the information toward the end of the prompt than the beginning.
Tell it what to do. Focus on what you want to get, and do not include what you do NOT want as it may cause confusion and a mixed result.
Iterate and ask again. If you don't get the result you're looking for on the first try (or second, or third) iterate on your prompt, change the phrasing and provide more or clearer details. This will change the result and hopefully get you closer to the outcome you are looking for.

More resource on writing prompts:

Document Classification Quick Guide

Public Data	Confidential Data	Highly Confidential Data
Directory data Public policies Publicly published business documents	Faculty/staff personnel records, benefits, salaries, performance evaluations, employment applications Internal memos and email Purchase requisitions Level 2 and 3 of student data All internal, un-published, or non-public university work product.	Protected health data Social security numbers Payment card numbers Financial account numbers Driver’s license numbers Level 4 and 5 of student data