Lesson 9: Model Cards - Exploration
45 minutes
Overview
In this lesson, students are introduced to Model Cards and Foresight in AI charts as tools used to plan for how AI products are used. Students begin by reflecting on what materials are provided to users when we buy a new product, then explore how a Model Card serves a similar purpose for chatbots to clarify and caution how a particular chatbot product should and shouldn’t be used. Then, students explore an example of a Library chatbot and use a Foresight in AI table to examine the intended and unintended users and use-cases for this chatbot. After this example, students create their own model card for a fictitious chatbot, and then analyze each other’s model cards with a Foresight in AI chart.
Agenda
Objectives
Students will be able to:
- Analyze and create model cards for chatbots, including potential intended and unintended uses
- Explain the importance of providing safety, transparency, and usability information for chatbots through model cards
Preparation
- Print copies of the activity guide for each student
- Run through the slide deck and videos ahead of time
- (Optional) Print out the slide deck with speaker notes as a resource during class. Click here for steps on how to do this.
Links
Heads Up! Please make a copy of any documents you plan to share with students.
For the teachers
- LEADERS Framework - Resource
- Model Cards - Exploration - Slides
For the students
- Foresight in AI Chart - Activity Guide
- Model Cards - Activity Guide
Teaching Guide
Before The Lesson
Preparing to Lead Discussions
LEADERS Framework: This lesson includes opportunities for students to discuss the impacts of generative AI on society, both the intended and unintended consequences. Navigating these conversations so they elevate and honor the experiences of your students, while also keeping the conversation grounded and connected to the lesson content, requires careful consideration and intentionality as the discussions are happening.
To help prepare for these moments, we recommend reviewing the LEADERS Framework for facilitating lessons and discussions involving the societal effects of AI, including bias and ethical considerations. Not all of the tenets are applicable to every lesson, but many of the teaching tips and discussion goals in this lesson follow the principles outlined in this framework.
Warm Up (5 minutes)
Slide | Lesson Guide |
Discuss: When you buy a new appliance or tool or new device or even a children’s toy, what different kinds of material does it come with? List as many things as you can think of. For example: an instructions manual
Have students discuss in small groups, then ask folks to share out in a popcorn style. This can be a quick moment before continuing to the next slide. Discussion Goal: Try to generate a wide variety of items, such as: - Instructions manual - Safety sheet - Contact information - Warranty card - Similar products card |
|
Discuss: Why do you think it’s important for companies to supply all of this stuff with a new product?
Have students discuss in small groups, before continuing to a class discussion. Discussion Goal: Students may list a variety of reasons, such as: - It’s important to give the user guidance on how to use the new thing they bought - It’s important to warn the user about any safety features so they don’t misuse it or get hurt - There may be laws or regulations that require certain information to be provided - Companies may be trying to sell more things As more examples are generated, try to tie this to a larger theme around the “social contract” between things we use and the companies that make them. Sometimes companies are required to provide information through regulation, but other times it’s just part of our own expectations around how we should start using a new product. |
|
Say: When we buy new products, there’s a whole host of information provided to us so we can better understand and use their products. Today we’ll investigate a similar idea with chatbots and what information we should be providing to people who want to use the chatbots we create.
Question of the Day: What information should we be providing to potential users of our chatbots? |
Activity (35 minutes)
Slide | Lesson Guide |
Discuss: If we were “shipping” our chatbot to a new user, what kind of information do you think should be included? For example: what’s an equivalent of a “contact card” or “warranty sheet” or “instruction manual”. Try to phrase things as “We should tell the user…” or “The user should really know about…”
Have students write down some of their responses first before discussing with a neighbor, then have a full-class discussion. Discussion Goal: This is a very open-ended prompt meant to foreshadow the rest of the lesson. Some answers might include “The user should really know who made the chatbot”, or “We should tell the user about any hallucinations” or “The user should really know why we made the chatbot and how it works”. Focus on responses that can be generalized across chatbots rather than something that might be specific to certain kinds of chatbots. In general, responses to this prompt should generally fall under Safety information (ie: what to do and not do), Transparency information (ie: how the chatbot was made and why), and Usability information (ie: how to get started). |
|
Say: One way to add this kind of extra information to a chatbot is to give it a special nametag, called a Model Card.
![]() Say: Model cards contain additional information about the chatbot for the user to review. These can include safety features which help the user know how to use the chatbot responsibly. Or transparency information, so the user knows how the chatbot was created or designed. And even usability information, which helps the user know where to get started. |
|
Say: Another way to think about model cards is how nutrition labels work on your food.
![]() Say: They both provide information about how the product was made ![]() Say: any warnings you should be aware of before using ![]() Say: and sometimes there are even examples of what you can use the chatbot for. Today, we’re going to explore an example model card and see how it can be useful in preparing for how chatbots can be used in society. |
Teaching Tip: Model Cards in the Real World - Model Cards are not a concept we invented - they were first coined in the paper Model Cards for Model Reporting via a project led by Dr Timnit Gebru. To date, model cards are used for many AI models and are an industry best-practice. To reinforce this point with students, you could decide to show them the Model Card Template on HuggingFace or show them an example of a model card like the one for DistilGPT2.
Say: Let's look at a model card for LibBot, a chatbot that specializes in library recommendations
Distribute: Activity Guide - Model Cards to each student |
|
Do This: Read over the model card. Be ready to discuss:
- What’s a piece of information related to safety? - What’s a piece of information related to transparency? - What’s a piece of information related to usability? Circulate: Have students read the article. Consider using a “marking the text” strategy, like highlighting key words or annotating in the margins. |
|
Discuss: What’s a piece of information related to safety? What’s a piece of information related to transparency? What’s a piece of information related to usability?
Have students share in a small group before sharing as a class. Discussion Goal: Try to generate 1-2 answers for each topic. Examples might include: - Safety: the model card says not to enter personal information - Transparency: the model has been tested with staff at the Des Moines. The model isn’t always accurate with obscure recommendations or recent releases. - Usability: the intended uses section describes how to use the model, and there are example prompts and topics. Ultimately, this is an opportunity for students to get familiar with the structure of a model card and how it communicates information to a potential user. |
|
Do This: Read the scenario aloud: The Des Moines library has decided they want to use this chatbot in their library system. But - they’ve also read stories of chatbots being misused in the past, so they don’t want any surprises. Using their model card, can we help them imagine how this chatbot will impact their community? | |
Say: One thing that can help us with this analysis is a Foresight in AI chart. It’s a way to imagine how chatbots or other AI products will get used in the real-world, imagining both the intended and unintended situations that can occur. Let's go through an example together.
|
Ethics Opportunity: Real-World Connection - The Foresight in AI chart was initially developed by Dr Margaret Mitchell, who is also a co-creator of Model Cards. You can read more about Model Cards and the Foresight in AI chart in the article The Pillars of a Rights-Based Approach to AI Development. Consider surfacing for students that this chart is a real tool used by practicing AI professionals and auditors.
Say: For example, here’s how we could use this chart with an Image Generation model.
![]() Say: The intended users might be artists, and the intended use-case might be to creative artistic images. The result is: artists make cool new art! Awesome! ![]() Say: And then there’s the intended audience, or the folks who are affected by this. For example, different art collectors who want to buy and curate new art. Also awesome! ![]() Say: But there may also be some unintended consequences. For example, all the artists whose images were used to train the model are also affected by this. They were never paid for their work, and now it’s easy to generate similar artwork based on their own style. How unfair is that? ![]() Say: There also may be some unintended users trying to use the model in unintended ways. For example, someone interested in misinformation could use this tool to make realistic fake images of political candidates to create confusion during an election. ![]() Say: These are just some possible answers - there are lots of different users and use-cases we could imagine! And as we make the chart, not all boxes get filled - some of them are left empty if the different situations don’t match up neatly. And that’s fine! Someone else could’ve imagined a different set of users or use-cases and ended up with a different chart! |
Teaching Tip: Crediting this Example - The example on this slide is based on an example presented by Dr Margaret Mitchell on X (formerly twitter).
Say: Let’s return to the Library chatbot for Des Moines and imagine how we could start filling this out.
Discuss: Who do you think is an intended user for this chatbot? Someone this chatbot was made for? Who do you think is an unintended user of this chatbot? Someone who might use this chatbot, but isn’t the target audience? Have students share with a neighbor before generating a few answers as a class. Discussion Goal: Students should use empathy to imagine themselves in situations where this chatbot could be used by them or someone they know. There are many acceptable answers to this prompt, such as: - Intended users: a student taking a class, an adult looking for a hobby, an avid reader wanting to save money by using the library instead of buying books - Unintended users: someone outside of Des Moines who might accidentally discover the chatbot, a malicious hacker looking to disrupt services As students share responses, consider writing them down on poster paper or a whiteboard. |
|
Discuss: Who do you think is an intended use for this chatbot? What is a task someone might try to complete? Who do you think is an unintended use for this chatbot? What is a task someone might try to use this for, but they really shouldn’t?
Have students share with a neighbor before generating a few answers as a class. Discussion Goal: Students should imagine reasons someone might use this chatbot. They should rely on the model card, especially the intended uses section of the model card. There are many acceptable answers to this prompt, such as: - Intended uses: looking for a recommendation, checking the status of a book, looking up the library hours. - Unintended uses: asking off-topic questions, using it as a search engine for specific catalog information As students share responses, consider writing them down on poster paper or a whiteboard. |
|
Discuss: Based on the last two discussions, can we fill in one of these boxes?
Circulate: Students should attempt to match the combinations of users and use-cases brainstormed in the last few slides. Some of these combinations are easier than others - for example, the intended user matching with the intended use-case usually looks like the ideal scenario you would imagine, and the unintended user matched with an unintended uses case can also be easier to imagine. Encourage students to think of the more nuanced pairings, like users who are affected by the intended use cases. Don’t regroup for a class discussion - instead, as students discuss, begin passing out the handout on the next slide. |
|
Say: Let's take a look at an example Foresight in AI chart that another student filled out for this scenario and see if they thought of any scenarios we hadn't considered.
Distribute: Foresight in AI chart to each student |
Teaching Tip: Another Example, not a "Correct" Example - The first page of this chart is one possible way to complete the Foresight in AI chart, identical to the discussion the class just had. This chart doesn't invalidate any of the points the class came up with and shouldn't be viewed as "the correct answer" to this exercise. Instead, this chart can highlight that there are multiple ways to view intended and unintended impacts that we may not have thought of, and it is useful for students to compare their ideas with the ones in this chart. This chart can also act as an exemplar for the next activity where students make one of their own.
Discuss: Another student filled out this chart for the Des Moines Library LibBot. What’s an idea that felt similar to your thoughts in the last discussion? What’s one idea that’s a surprise and you didn’t consider?
Have students read over the chart, then discuss with a neighbor before regrouping. Discussion Goal: This can be a quick discussion, but it should also cement students’ understanding of how this chart can be used to represent potential uses (and misuses) of chatbots and AI before continuing to the next activity. Student responses will vary depending on how the discussion has gone so far, but listen for ways in which students are identifying unique users and use-cases, and imagining both the intended and unintended consequences of this chatbot. |
Ethics Opportunity: Real-World Connection - This scenario investigates the potential uses and misuses of a chatbot in the context of library book recommendations. This is very loosely inspired by real-world events where school districts relied on chatbots like ChatGPT to make decisions about books to comply with state laws. In an effort to comply with these new laws, several school districts hastily resorted to using AI to make recommendations on certain books to remove from circulation. This is an example of how an intended user - libraries and schools - were using chatbots for a surprising and unintended use case with a wide variety of affected uses (also both intended and unintended). You can read more about this situation in the following articles:
Do This: Now it’s your turn! Flip over the Model Card handout to the back side! Imagine a chatbot you would like to make, or think of a chatbot you made earlier in this class. Create a model card for this chatbot!
Circulate: Monitor students as they start to make their own model card for a chatbot they want to create. This is an aspirational activity, where students are imagining how this chatbot would look and function before they actually create it. Students may struggle to fill in the "testing and evaluation" setting since they haven't actually made the chatbot yet - in this case, have them imagine how they would want to test and evaluate the chatbot and write that in this box. |
|
Do This: Now Trade Your Model Card with Someone Else! | |
Do This: Flip over the Foresight in AI handout to the back side! You’re going to fill out the chart for your partner’s model card.
- Try to identify the set of intended users and use-cases that this model is aiming for. - Try to identify some unintended users and use-cases they may not have considered! Circulate: Monitor students as they fill in the Foresight in AI chart. If students seem stuck, have them consider the model card to fill in some of the information such as the intended user and use-case. It’s also okay for not every box to be filled in. |
|
Do This: Now Trade Your Model Card with Someone Else! |
Wrap Up (5 minutes)
Slide | Lesson Guide |
Discuss: What’s something you expected in the Foresight in AI chart that your partner completed? What’s something unexpected that you hadn’t considered?
Have students complete these prompts as an exit ticket or in a journal. Consider collecting these along with the Model Card and Foresight in AI handouts. |
Assessment Opportunity: Model Cards & User Experiences - The Model Card sheet and the Foresight in AI charts can be collected and used to assess how well students are able to articulate the uses and misuses of chatbots, and how well students can imagine the intended and unintended impacts of chatbots on society.
Question of the Day: What information should we be providing to potential users of our chatbots? |
This work is available under a Creative Commons License (CC BY-NC-SA 4.0).
If you are interested in licensing Code.org materials for commercial purposes contact us.