Lesson 7: How AI Sees - Image Recognition Basics

Learning Objectives

• Explain how AI systems process images as numerical data (pixels) rather than visual scenes, understanding the fundamental difference between human and computer vision.
• Identify the four basic steps in the image recognition process: image input, feature extraction, pattern matching, and classification output.
• Demonstrate understanding of how AI learns to recognize objects through training with diverse example images, and explain why training data quality matters.
• Compare the strengths and limitations of human vision versus computer vision, including accuracy, speed, consistency, and common failure cases.
• Create a simple image classification system using available online tools, documenting what works well and what challenges arise during training and testing.

Standards Alignment

• CSTA 2-AP-10: Use flowcharts and/or pseudocode to address complex problems as algorithms. Students model the image recognition process as a series of logical steps.
• CSTA 2-DA-08: Collect data using computational tools and transform the data to make it more useful and reliable. Students gather and organize training images to teach AI systems.
• CSTA 2-IC-20: Compare tradeoffs associated with computing technologies that affect people's everyday activities and career options. Students explore privacy implications and bias concerns in facial recognition systems.
• ISTE 1.5.c: Break problems into component parts, extract key information, and develop descriptive models to understand complex systems or facilitate problem-solving.
• ISTE 1.6.b: Create original works or responsibly repurpose or remix digital resources into new creations through hands-on creation of image classification models.
• NGSS MS-ETS1-2: Evaluate competing design solutions using a systematic process to determine how well they meet the criteria and constraints of the problem, applied to evaluating AI accuracy and effectiveness.

Materials Needed

• Computer or tablet with internet access (one device per pair or small group of 2-3 students)
• Access to Google Teachable Machine at teachablemachine.withgoogle.com (free, no login required, works best in Chrome or Edge browser with webcam access)
• Optional: Access to Google Quick, Draw! at quickdraw.withgoogle.com for warm-up activity
• Printed image cards for unplugged sorting activity (30-40 cards total, included in downloadable materials—print on cardstock if possible)
• Student handout: "How AI Sees Images" worksheet with diagram and reflection questions (included in downloadable materials, one per student)
• Chart paper or whiteboard for class brainstorming and vocabulary building
• Markers or dry-erase markers in multiple colors for visual diagrams
• Common classroom objects for live demonstrations (books, pencils, water bottles, scissors, etc.—bring 6-8 distinct items)
• Optional: Webcam for teacher demonstration if not built into computer
• Optional: Smartphone with Google Lens or similar app for real-world demonstration
• Optional: Prepared screenshots or video recordings of image recognition tools as backup if internet connectivity is unreliable
• Exit ticket slips or digital form for closing reflection (template included in downloadable materials)

Lesson Procedure

1. Hook and Engagement: AI Mystery Images Challenge (8 minutes)

   Begin class with an engaging "Can you beat the AI?" challenge that immediately captures student attention and generates curiosity about how computers interpret images.

   Display the AI Challenge:

   • Show 4-5 images on the screen or board. Include a mix: some that AI clearly identified correctly (a simple dog photo labeled "dog"), and some hilarious AI failures (a muffin labeled as "chihuahua," or a person in weird lighting labeled as "furniture")
   • Ask students to vote or predict: "Which of these images do you think artificial intelligence identified correctly?" Have them write their guesses on paper or use hand signals
   • Reveal the answers one by one, allowing time for surprise and laughter at AI mistakes
   • Show one more challenging example—perhaps an optical illusion that tricks both humans and AI

   Generate Curiosity:

   • Ask: "Why do you think AI got some of these wrong? What's going on inside the computer when it looks at a picture?"
   • Allow 1-2 students to share initial ideas without correction—this helps you assess prior knowledge
   • Pose the essential question: "How does AI actually 'see' these images? Spoiler alert: It doesn't see them the way you and I do!"
   • Write the lesson's big question on the board: "How does AI see images differently than humans?"

   Optional Quick Activity: If time allows and you have device access, let students try Google Quick, Draw! for 2-3 minutes to experience AI trying to guess their drawings in real-time. This primes them for the concepts ahead.

2. Direct Instruction: Understanding How AI Processes Images (12 minutes)

   Provide clear, concrete explanations of computer vision concepts using visual aids, analogies, and step-by-step demonstrations. This section builds the foundational knowledge students need for hands-on activities.

   Explain the Pixel Concept:

   • Show a high-resolution image on screen, then zoom in dramatically to reveal individual pixels (colored squares)
   • Explain: "AI doesn't see a picture of a dog. It sees thousands of tiny colored dots called pixels—like a digital mosaic"
   • Demonstrate that each pixel has numbers: Red, Green, Blue values (RGB). For example: pure red = (255, 0, 0)
   • Use analogy: "Imagine describing a painting to someone over the phone using only numbers. That's what the computer works with!"

   Introduce Pattern Recognition:

   • Explain: "AI finds patterns in all those pixel numbers. For example, dog photos often have patterns of brown/black pixels arranged in certain ways, especially around the nose and ears"
   • Draw a simple diagram on the board showing: Image → Pixels (numbers) → Pattern Detection → Classification
   • Give concrete example: "To recognize a stop sign, AI learns the pattern: eight-sided red shape + white letters = stop sign"
   • Emphasize: AI is following mathematical rules and patterns, not "seeing" or "understanding" like humans do

   Explain Training Data:

   • Introduce key concept: "AI learns by example. We show it thousands of labeled images so it can learn patterns"
   • Use relatable analogy: "It's like learning to identify birds. The first time you saw a robin, someone probably said 'That's a robin.' After seeing many robins, you learned what makes a robin look like a robin. AI does the same thing, but with millions of examples!"
   • Explain quality matters: "If we only show AI photos of golden retrievers, it might not recognize a chihuahua as a dog. Diversity in training data is crucial"

   Demonstrate the Process:

   • Show a simple flowchart or diagram of the image recognition process with four clear steps: Input (camera/photo) → Processing (break into pixels, find patterns) → Classification (match to learned categories) → Output (label: "dog" 95% confident)
   • Reference one of the earlier AI mistakes and walk through why it might have happened: "The muffin has similar colors and round shape patterns to some dog noses, so the AI got confused"
   • Write key vocabulary on board: Pixels, Pattern Recognition, Training Data, Classification, Computer Vision

   Real-World Connection: Briefly mention 2-3 places students encounter this technology: "Your phone's face unlock, photo organization apps, and Instagram filters all use image recognition AI!"

3. Hands-On Exploration: Train Your Own AI (16 minutes)

   Students work in pairs to train a simple image classification model using Google Teachable Machine, experiencing firsthand how AI learns from examples and observing what works well and what doesn't.

   Setup and Introduction (3 minutes):

   • Divide students into pairs and assign one device per pair
   • Direct them to teachablemachine.withgoogle.com and click "Get Started" then "Image Project"
   • Explain the interface briefly: "You'll create 2-3 classes (categories) and teach the AI to tell them apart using your webcam"
   • Demonstrate the interface: "Class 1 might be 'Pencil,' Class 2 might be 'Book,' Class 3 might be 'Hand'"
   • Emphasize: "You need to allow webcam access when prompted—this is how we'll take training photos"

   Training Phase (8 minutes):

   • Have pairs choose 2-3 simple objects they can easily hold up to the camera (books, pencils, water bottles, hands, scissors, etc.)
   • Students should rename their classes with clear labels
   • For each class, take 30-50 photos: "Hold the object different ways—turn it, move it closer/farther, change the angle. This helps AI learn better!"
   • Encourage variety: "Take some with your hand partly covering it, some in shadow, some tilted. Real-world images aren't perfect!"
   • After collecting images for all classes, click the "Train Model" button and wait 30-60 seconds
   • Circulate and assist with webcam issues, lighting problems, or technical questions

   Testing and Observation (5 minutes):

   • Once trained, students test their models: "Hold up each object. Does the AI recognize it correctly? What confidence percentage does it show?"
   • Encourage experimentation: "Try holding two objects together. Try showing it from a weird angle. Try to trick it! What happens?"
   • Have students document on their worksheet: What worked well? What confused the AI? What was surprising?
   • Ask students to notice the confidence percentages: "Why do you think it's 87% sure this is a pencil, not 100%?"

   Quick Share-Out: Have 2-3 pairs quickly share one interesting discovery or funny mistake their AI made. This keeps energy high and reinforces learning through peer examples.

   Teacher Tips: Have a working example ready to troubleshoot issues. Some common problems: webcam not working (try different browser), poor lighting (move near window), or too-similar objects (choose more distinct items).

4. Unplugged Activity: Human Image Classifier Game (14 minutes)

   This hands-on activity helps students understand image recognition limitations by having them act as "AI" following strict classification rules. It's engaging, requires no technology, and deepens conceptual understanding through kinesthetic learning.

   Setup and Rules (3 minutes):

   • Divide class into groups of 3-4 students. Give each group a set of image cards (face-down initially)
   • Explain: "You're going to be human AI systems! You'll sort these cards based only on specific rules I give you, without looking at the actual images first"
   • Write the sorting rules on the board. Example rules: "Group A: Mostly warm colors (red, orange, yellow). Group B: Mostly cool colors (blue, green, purple). Group C: Mostly black, white, or gray"
   • Alternative rules: "Sort by: Has circles vs. has straight edges" or "Has living things vs. has only objects"
   • Emphasize: "Just like AI, you must follow the rules exactly. You can't use your human judgment about what the picture really shows!"

   First Sorting Round (4 minutes):

   • Groups flip cards one at a time and sort them into categories based strictly on the rules
   • Students may not discuss what the image actually shows—only which category it fits based on the features
   • If a card seems ambiguous, they put it in a "Not Sure" pile
   • Circulate to ensure groups are following rules strictly, like an algorithm would

   Reveal and Discuss (4 minutes):

   • After sorting, groups look at what they classified. Ask: "Did any cards end up in surprising categories?"
   • Show example: "This picture of a sunset over the ocean ended up in 'warm colors' even though it's a nature scene. An AI using color-based rules would do the same thing!"
   • Introduce "trick cards"—ambiguous images intentionally included. For example: a black and white photo of a colorful parrot, or an object with both curves and straight lines
   • Discuss: "Why were these hard to classify? What would AI need to handle these better?"

   Reflection and Connection (3 minutes):

   • Ask whole class: "How did this activity feel? What was frustrating about following strict rules?"
   • Make the connection explicit: "This is exactly what rule-based AI faces! It can't use common sense or context like you can"
   • Discuss: "How is our earlier Teachable Machine activity different from this? That AI learned patterns from examples instead of following rigid rules we gave it. This is called machine learning!"
   • Emphasize: Even machine learning AI has limitations—it might learn the wrong patterns if training data is biased or limited

   Extension: If time permits, let groups try creating their own sorting rules and see if other groups can correctly classify using those rules.

5. Real-World Applications and Ethical Reflection (10 minutes)

   Connect classroom activities to real-world uses of image recognition technology while introducing important ethical considerations about privacy, bias, and societal impact. This helps students see themselves as informed users and future creators of AI systems.

   Brainstorm Where They See Image Recognition (3 minutes):

   • Ask: "Where do YOU encounter image recognition in your daily life?" Write all student responses on the board
   • Prompt if needed: "Think about your phone, social media, cars, stores, hospitals..."
   • Create categories as you list them: Personal Devices, Social Media, Transportation, Security, Medicine, Entertainment
   • Add any important ones students missed: Face unlock on phones, photo organization (Google Photos), Instagram filters, self-driving car vision, medical imaging (X-rays, MRIs), wildlife cameras, store checkout systems, sports replay analysis, accessibility tools for visually impaired

   Explore Real-World Examples (4 minutes):

   • Choose 2-3 examples to explore briefly:
     • Medical Imaging: "AI can spot tiny cancer cells in X-rays that humans might miss. It's helping doctors save lives!"
     • Accessibility: "Apps like Microsoft's Seeing AI describe the world to people who are blind or visually impaired. They point their phone at something and AI tells them what it sees"
     • Wildlife Conservation: "Camera traps in forests use AI to identify endangered animals, helping scientists track populations without disturbing them"
   • Show brief video clip or images if available (30-60 seconds total)
   • Emphasize positive applications while acknowledging this technology is powerful

   Discuss Privacy and Ethical Concerns (3 minutes):

   • Pose questions for brief discussion: "If AI can recognize faces, who should be allowed to use that technology? Should stores track your face? Should schools? Should police?"
   • Introduce bias concerns: "Remember our training data conversation? If an AI is trained mostly on photos of light-skinned faces, it might not work well for people with darker skin. This has actually happened with some commercial systems!"
   • Discuss: "Is it okay to post photos of your friends online without asking? AI systems learn from publicly posted photos. Does that matter?"
   • Briefly mention deepfakes: "AI can now create fake images that look real. How might this be a problem?"
   • End on empowering note: "As you grow up using and maybe creating AI, you'll help decide how these tools should be used. What rules would you want?"

   Student Voice: Allow 1-2 students to share concerns, ideas, or questions. Validate their thinking and acknowledge that these are questions society is still working to answer.

Assessment Strategies