Machine Learning Unplugged

Who says you need a computer to learn about artificial intelligence? In this unplugged activity, take the role of a wildlife conservationist who uses data to refine and test a machine learning algorithm.

Preparation

1. Print the Data Set Cards  (1 set per group).
   • Note: Set aside the Part 2 cards for distribution later in the activity.
2. Plan to have learners work in groups of 3-4.
3. Try the activity with other educators or kids you know. This will give you practice with the materials to be able to anticipate student questions.
   • Note: Review the Answer Key if needed. 

Lesson Directions

Part 1: Machine Learning Scenario

Introduction (5 min)

1. Provide students with a basic introduction to artificial intelligence (AI) and machine learning (ML). Use the Background Information above and discussion questions below:
   • What comes to mind when you think of AI?
   • When do you see AI being used?
   • What do you think of when you hear the words "machine learning”?
2. Introduce the scenario.

3. Divide students into groups of 3-4. Let them know that you will give them:

• A set of images (testing data) of animals they are monitoring.
• An algorithm (set of rules) created by the AI to categorize the animal.

These data will help test how well the AI model can identify the animal.

• Their job as humans will be to test the algorithm and sort the images into two categories, A or B.

Test the AI Program (10 min) 

1. Have learners test the algorithm that the AI program created.
2. Distribute the Data Set Cards for Part 1 to each group.
3. Give them 5 minutes to sort the cards for Part 1 based on the rules from the Algorithm (see below).
   • Define any words that students may not understand or provide them with synonyms.
   • Remember that those are the only rules that the AI program understands, so students cannot use their existing knowledge of the species.
   • The algorithm may be imperfect or lead to errors. Remind students that they should notice them rather than correct them.

4. When time is up, have groups share the images they added to each category.

• See if they can identify the two species which are being categorized.
• Refer to the Answer Key for more information. Remind students that their goal was to test the AI model, not themselves.

Examples of sorted cards: 

Debrief (10 min) 

1. Lead a discussion which considers some of the following questions:
   • How did you decide if the image belonged to Category A or B?
   • How many images were categorized correctly based on the AI algorithm?
   • How helpful was the AI’s algorithm? If you could modify the algorithm, what would you change or add?
   • What is the difference between how people see and how AI/machines “see”?

Part 2: Revise and Improve

Iterate (10 min) 

1. Reassure the students that it would have been unlikely that they would have been able to categorize all of the animals correctly based on the algorithm provided.
2. Introduce them to the new scenario:
   
   

   Design Scenario

   The FWS has a new set of data to test with the AI. However, before testing they have asked that you try to improve the algorithm so that it will be more accurate. Your job is to develop a revised set of rules (algorithm) which will lead to more accurate and reliable results. Then, test that the AI can correctly identify a new set of data.

3. Students can base their revisions on the algorithm and images provided in Part 1.
4. They should check first that their new algorithm helps the AI correctly categorize the data set from Part 1 before testing it on a new set of images.

Test the AI Program (10 min) 

1. Let the groups know that they will now be provided with a new testing data set (Part 2) to test the revised algorithm.
2. Try one of the following options:
   • Teacher Testing: Model the testing process with an algorithm that is likely to make mistakes. Demonstrate how to follow the algorithm during testing despite this. Give students a few minutes to adjust their algorithms again before having them test out their algorithm with the new data.
   • Peer-to-Peer Testing: Have groups exchange algorithms and test each other’s rules. Encourage them to share their results with each other and note the mistakes that the AI systems made.
   • Special Guests: Invite special guests to visit the class and help test out the algorithms. Guests could include people from the school (another class, teachers, or principal), volunteers from a local business or even professionals who work with AI. Pair guests up with one or more groups to test the algorithms.
3. Once students understand the process, distribute the Data Set Cards for both Parts 1 and 2.
4. Give them about 5 minutes to sort the Data Set Cards for Part 2 based on the rules from the algorithm.
   • Remind groups that those are the only rules that the AI program understands.
   • Note: It is recommended that the images be categorized with the closest match to the written algorithms even if it’s a mistake. This provides an opportunity for students to see the types of mistakes AI systems can make based on algorithms.
   • Remind students that their goal is to test the AI, not themselves.
5. When time is up, have groups share the images they added to each category.
   • Have them discuss how they considered the new species which were introduced.
   • Refer to the Answer Key for more information.

Examples of sorted cards: 

Debrief (5 min) 

1. Lead a short debrief of the activity and the concepts. Help students make connections between the skills they used in this activity, the work of computer scientists, and real world applications of AI and ML.
   • Remind students computer scientists need to find the data they want to use to train a ML model and then test that model.
     • Example: Security cameras that use AI to recognize objects have large data sets that computer scientists provide the model so that they can test and “see” if it’s a person, animal, or another object.
   • During the activity, students practiced this themselves by using training and testing data and revising an algorithm based on previous data.
2. Possible Debrief Questions include:
   • How did your algorithm determine if the image was under Category A or B?
   • How many images were categorized correctly based on the new algorithm?
   • Compare and contrast the algorithms you used for both data sets.
   • How detailed does an algorithm (set of rules) need to be to increase accuracy?
   • When training AI models, what is important to keep in mind?

Standards Connections

Vocabulary