Getting Started with Generative AI
Generative AI is a form of artificial intelligence that can be used for a wide variety of tasks. If you decide to integrate this technology into your work practices, use the following three steps to help ensure that you’re following institutional guidance about data protection and getting the most out of your AI explorations.
What do you need to know before you start using an AI tool at Northwestern?
Data protection and use of AI tools are deeply intertwined. Begin by answering these two questions:
- Which level(s) is my data?
- Based on the data level, which AI tool can I use?
Know Your Data
Data and information are classified into four different levels at Northwestern. This table is a summary of the levels, types of data for each, and examples. For more details, see the Data Classification Policy.
| Data Level | Impact Level | Data Type (Not exhaustive) | Examples (Not exhaustive) |
|---|---|---|---|
| 1 | None or low on the University and affiliates | Non-confidential Data and Publicly Published Data | Information on Northwestern’s website |
| 2 | Publicly noticeable impact on the University and affiliates | Personal Private Data, Business Sensitive Data | Employee hiring information, internal accounting information, unpublished research data |
| 3 | Serious or severe impact on the University – risk of civil or criminal penalties | Business Sensitive Data, Personal Private, HIPPA PHI, Contractually/Legally Restricted Data (Potential for civil or criminal penalties | Restricted financial information, social security numbers, medical diagnoses |
| 4 | Severe or catastrophic impact on the University or national security - inherent risk of significant fines or penalties, regulatory action, or civil or criminal violations | Classified Data | Government classified data, export-controlled data |
Data levels matter when you are using generative AI because, in most cases, the data you share as part of your queries in generative AI tools will be accessible by others using the same tools. This is because generative AI learns by collecting, analyzing, and storing user-provided information. Therefore, University faculty, staff, students, and affiliates should not enter institutional data into any generative AI tools that have not been validated by the University for appropriate use and have explicit permission of the data provider. View Northwestern Guidance on the Use of Generative AI.
Quick Tip Video
Do you know how to determine which level of data you’re working with? View this video to learn more about data categories and what they mean.
Which Tool Can I Use with My Data?
Now that you’ve identified the levels of data you’re working with, you need to find the AI tool that is appropriate for that data.
Start with Copilot Chat
Copilot Chat has been validated by the University for appropriate use. When you sign in with your Northwestern NetID, your data is protected. Microsoft does not use your prompts or Copilot’s responses to train or improve AI models. Find out how to access Copilot chat.
Data Levels and Appropriate AI Tools
| Data Level | Microsoft Copilot, when signed in with Northwestern account | Northwestern Azure with Open AI | Publicly available tools (e.g. ChatGPT, Gemini, Claude, etc,) |
|---|---|---|---|
| 1 (Public Data) | Yes | Yes, with appropriate security and access controls | Yes |
| 2 (Sensitive/Regulated Data) | Yes | Yes, with security and access controls that meet or exceed regulatory or data protection requirements | No |
| 3 (Sensitive/Regulated Data) | Yes, with some exceptions. You are responsible for checking contractual or legal obligations. | Yes, with security and access controls that meet or exceed regulatory or data protection requirements | No |
| 4 (Sensitive/Regulated Data) | Yes, with some exceptions. You are responsible for checking contractual or legal obligations. | Yes, with security and access controls that meet or exceed regulatory or data protection requirements | No |
If you have additional questions about whether your data type is appropriate for Copilot, please contact the Information Security Office (security@northwestern.edu).
Additional Information
- Quick Tips Video [LINK NEEDED]
- What is Copilot and why should it be your default AI tool at Northwestern?
- Microsoft 365 Copilot Add-On (Subscription based)
- Microsoft 365 Copilot add-on is a yearly subscription-based product that integrates AI into Teams, Word, Outlook, PowerPoint, Excel, and other Microsoft 365 apps. This is currently only available to faculty and staff. How to purchase 365 Copilot [LINK NEEDED]
For most of us, working with an AI tool will mean a conversation in a chat window. Your part of the conversation is called a “prompt” and building on that initial prompt with corrections, changes, and updates, is called “iterating.” There are many approaches to this process, but some general tips include
- Give the AI tool a character or role to play. This helps define its expertise, perspective, and can significantly improve the specificity of interaction.
- Give the AI tool examples of what you want the answer to look like and what you don’t want it to include.
Examples
Here are some examples to try, with ideas for further iteration and improvement. Copy the prompt text into a tool like Copilot, edit the text to match your interests, and begin your conversation. Use the Iterations column for ideas to keep exploring.
Information Overload
Your office provides a large amount of information to students and their families around the admissions process. Sometimes it is difficult to find the exact document or URL for reference for the inquiries you receive. You’d like to generate an index or other documentation that will serve as an internal finding aid for you and your colleagues.
| Prompt | Iterations |
|---|---|
| Create an internal reference guide for our [e.g., admissions] office using the documents stored in [enter link to SharePoint site]. Please generate a categorized index that includes document titles, brief summaries, and direct links. Organize the index by topic [e.g., Financial Aid, Application Deadlines, Campus Visits, Transfer Students, International Applicants]. Make it easy to search and update as new documents are added. |
|
Engaging Interviews
You are interviewing a candidate for a leadership position in your area. Generate a list of questions tailored to your subject matter expertise that are important for a leader to answer. In addition, generate a short list of bullet points that the candidate should mention when answering these questions.
| Prompt | Iterations |
|---|---|
| You are a leadership search specialist. I am assisting with interviews for potential director of my department, [e.g., facilities management]. Help me think of questions to ask during the interview that speak to what an effective leader will need to succeed in this area. |
|
Quick Tips Video
Use AI to teach you how to build a great prompt!
What do you do with what you get out of an AI tool?
Generative AI tools can create new things for us but they are fundamentally different in how they go about that process compared to a human. Because of errors, inconsistencies, inappropriate tone, and other reasons, your work with an AI tool isn’t complete until you, the human, review the output and decide how best to use it. Your own review process should reflect the policies, best practices, and values of your role, department, and unit.
Some review practices could include:
- Review for accuracy: Are the dates, calculations, names, and citations verifiable and accurate? Check links and sources cited by the AI tool. Verify that a process suggested by the tool works in real life.
- Review for authenticity: Where does the AI tool not fully reflect your voice, perspective, or personality? Consider when you should add or edit to bring your voice to the fore.
- Review for expertise: Where does the AI tool provide a starting point but not the full picture? Your knowledge and experience should augment output created by an AI tool.
- • Review for impact: When is human creation non-negotiable? Are there situations or contexts in your work in which having AI-produced output could have a negative impact on the work you are doing?
Quick Tip Video
Creating your own quality check process