Instructions for Teachers

This document contains the complete content for the PowerPoint presentation. Use this to create slides with visual diagrams, images, and animations. Recommended: 20-25 slides total.

SLIDE 1: Title Slide

Title: How AI Sees Images: Computer Vision Basics

Subtitle: Understanding How Computers Process Visual Information

Image: Eye icon merging with circuit board pattern

Footer: Lesson 7 | Evolve AI Institute

SLIDE 2: Opening Question

Title: Can You Beat the AI?

Content:

• Display 4-6 images with AI predictions
• Include both correct and incorrect identifications
• Examples:
• Golden retriever (correctly identified as "dog" - 98% confidence)
• Chihuahua muffin (incorrectly identified as "chihuahua" - 87% confidence)
• Person in shadows (incorrectly identified as "furniture" - 72% confidence)
• Clear stop sign (correctly identified as "stop sign" - 99% confidence)

Prompt: "Which ones did the AI get right? Which ones did it get wrong? Why?"

SLIDE 3: The Big Question

Title: How Does AI "See" Images?

Content:

• Spoiler alert: AI doesn't see the way you do!
• Today we'll discover:
• How computers process images
• What pixels really are
• How AI learns to recognize objects
• Why AI makes mistakes
• Real-world applications

Visual: Question mark over an image being scanned by a computer

SLIDE 4: What Humans See vs. What AI Sees

Title: Two Very Different Perspectives

Left Side - Human Vision:

• A cute golden retriever puppy
• Fluffy fur
• Happy expression
• Playing in grass

Right Side - Computer Vision:

• 3,145,728 pixels
• RGB values: (187, 154, 98), (201, 168, 112)...
• Pattern detection: high frequency of brown/tan pixels
• Shape detection: oval clusters, vertical elements

Key Point: Humans see meaning. Computers see numbers.

SLIDE 5: What Is a Pixel?

Title: The Building Blocks of Digital Images

Content:

• Pixel = Picture Element
• Tiny colored square
• Every pixel has a color defined by three numbers (RGB):
• Red value (0-255)
• Green value (0-255)
• Blue value (0-255)

Visual:

• Show image at normal resolution
• Progressive zoom showing pixels
• Individual pixel highlighted with RGB values

Example: Pure red = (255, 0, 0), Pure blue = (0, 0, 255), White = (255, 255, 255)

SLIDE 6: From Picture to Numbers

Title: How Images Become Data

Content:

• High-resolution photo: millions of pixels
• Example: 1920 × 1080 screen = 2,073,600 pixels
• Each pixel = 3 numbers (RGB)
• Total numbers: 6,220,800 values!

Visual:

• Show transformation: Photo → Pixel grid → Number array
• Animated sequence showing breakdown

Analogy: "It's like describing a painting using only numbers!"

SLIDE 7: The Four-Step Process

Title: How AI Recognizes Images

Visual: Flowchart with icons for each step

Step 1: Image Input

• Camera captures image
• Image converted to pixels
• Icon: Camera

Step 2: Feature Extraction

• AI analyzes pixel patterns
• Identifies edges, shapes, colors, textures
• Icon: Magnifying glass over patterns

Step 3: Pattern Matching

• Compares to learned patterns
• Uses training data
• Icon: Puzzle pieces fitting together

Step 4: Classification

• Assigns label with confidence score
• Example: "Dog" - 95% confident
• Icon: Label tag

SLIDE 8: Pattern Recognition

Title: How AI Finds Patterns

Content:

• AI looks for recurring features:
• Edges and boundaries
• Colors and color combinations
• Shapes (circles, rectangles, triangles)
• Textures (smooth, rough, striped)
• Arrangements and proportions

Visual Examples:

• Stop sign patterns: Eight-sided, red, white letters, specific proportions
• Dog patterns: Floppy ears, nose features, fur texture, four legs
• Cat patterns: Pointed ears, whiskers, eye shape

SLIDE 9: Training Data Is Key

Title: How AI Learns From Examples

Content:

• AI must be trained with thousands of labeled images
• The more diverse examples, the better it learns
• Quality matters as much as quantity

Visual:

• Grid of example training images (dogs in various poses, colors, breeds)
• Show diversity: puppies, adult dogs, different breeds, various angles

Analogy: "Just like you learned what birds look like by seeing many different birds, AI needs many examples too!"

SLIDE 10: Why Training Data Matters

Title: Good Data = Good AI

Two Scenarios:

Scenario A: Limited Training

• AI trained only on golden retrievers
• Shown images: Only golden retrievers
• Result: Might not recognize chihuahua, poodle, or husky as dogs

Visual: X mark over other dog breeds

Scenario B: Diverse Training

• AI trained on 50+ dog breeds
• Shown images: All types of dogs in various settings
• Result: Accurately recognizes most dogs regardless of breed, size, or pose

Visual: Check mark over all dog breeds

Key Lesson: Diversity in training data = Better AI performance

SLIDE 11: Confidence Scores

Title: AI Makes Predictions, Not Certainties

Content:

• AI gives confidence percentage with each prediction
• 95% confidence = AI is pretty sure, but not 100% certain
• Low confidence = AI is guessing

Examples:

• "Dog" - 98% (Very confident, clear features)
• "Dog or Cat?" - 52% Dog, 48% Cat (Very uncertain, ambiguous image)
• "Dog" - 23% (Not confident, probably wrong)

Visual: Progress bars showing different confidence levels

Key Point: Higher training quality = Higher confidence scores

SLIDE 12: When AI Makes Mistakes

Title: Common AI Vision Failures

Display 4-6 Examples:

1. Similar Patterns:
2. Muffin vs. Chihuahua (both round, tan, textured)
3. Bagel vs. Puppy face

1. Unusual Angles:
2. Dog photo from directly above (looks like a blob)
3. Person lying down (AI thinks it's an object)

1. Poor Lighting:
2. Dark photo where features aren't visible
3. Backlit photo creating silhouette

1. Partial Objects:
2. Only dog's tail visible (AI can't see enough)
3. Person mostly behind tree

1. Objects Outside Training:
2. Rare breed never seen in training data
3. New type of object AI never learned

Key Point: AI is powerful but not perfect!

SLIDE 13: Human Vision vs. Computer Vision

Title: Comparing the Two Systems

Table Format:

Aspect	Human Vision	Computer Vision
Speed	Moderate (brain processing)	Very fast (milliseconds)
Consistency	Variable (fatigue, attention)	Consistent (never tired)
Context	Excellent (uses experience)	Limited (only trained patterns)
Unusual Cases	Good (adapts easily)	Poor (struggles with novelty)
Learning	Few examples needed	Thousands of examples needed
Accuracy	Very high for familiar objects	High when trained well
Understanding	Knows meaning and purpose	No understanding, just patterns

Visual: Icons representing each comparison point

SLIDE 14: Real-World Application - Medical Imaging

Title: AI Helping Doctors Save Lives

Content:

• AI analyzes X-rays, MRIs, CT scans
• Can detect tiny cancer cells humans might miss
• Works alongside doctors, not replacing them
• Helps radiologists review images faster

Visual: Side-by-side medical images with AI highlighting suspicious areas

Impact: Early detection saves lives!

SLIDE 15: Real-World Application - Accessibility

Title: Helping People See the World

Content:

• Apps like Microsoft Seeing AI
• Describes surroundings to people who are blind or visually impaired
• Can read text, identify products, recognize faces
• Smartphone camera becomes helpful assistant

Visual: Person using smartphone app, screen showing text being read aloud

Impact: Technology creating independence and access!

SLIDE 16: Real-World Application - Wildlife Conservation

Title: Protecting Endangered Species

Content:

• Camera traps in forests automatically photograph animals
• AI identifies species without human needing to review every photo
• Tracks population numbers and migration patterns
• Reduces disturbance to wildlife

Visual: Camera trap images with AI species identification labels

Examples: Snow leopards, tigers, elephants, rare birds

Impact: Better data helps conservation efforts!

SLIDE 17: Real-World Application - Self-Driving Cars

Title: Computer Vision on the Road

Content:

• Cars use cameras and AI to "see" the road
• Identifies: pedestrians, other vehicles, traffic signs, lane markings, obstacles
• Makes split-second decisions
• Multiple cameras provide 360-degree awareness

Visual: Car dashboard view showing AI detecting objects around vehicle

Challenge: Must be extremely accurate for safety!

SLIDE 18: Real-World Application - Agriculture

Title: Smart Farming With AI Vision

Content:

• Drones with cameras fly over crops
• AI identifies diseased plants, pest infestations, irrigation needs
• Helps farmers target problems precisely
• Reduces waste of water and pesticides

Visual: Drone aerial view with AI highlighting problem areas in fields

Impact: More efficient farming, better food production!

SLIDE 19: Privacy and Ethical Concerns

Title: Important Questions We Must Ask

Content:

Facial Recognition:

• Should stores track your face?
• Should schools use it for attendance?
• What about police use?

Bias Problems:

• Some AI systems work better for certain skin tones
• Why? Training data wasn't diverse enough
• This is a serious problem being actively addressed

Consent Matters:

• Is it okay to post photos of friends without asking?
• AI learns from publicly posted photos
• Your digital footprint matters

Deepfakes:

• AI can create fake but realistic images
• How can we tell what's real?

Discussion Prompt: What rules would YOU want for AI vision technology?

SLIDE 20: Your Turn to Be AI!

Title: Unplugged Card Sorting Activity

Content:

• You'll act as a rule-based AI system
• Sort image cards using only specific rules
• No using your human judgment!
• Experience AI's limitations firsthand

Rules Example:

• Group A: Mostly warm colors (red, orange, yellow)
• Group B: Mostly cool colors (blue, green, purple)
• Group C: Mostly black, white, or gray

Visual: Sample cards and sorting categories

SLIDE 21: Hands-On: Train Your Own AI!

Title: Google Teachable Machine Activity

Content:

• Website: teachablemachine.withgoogle.com
• You'll teach AI to recognize objects using your webcam
• Steps:

1. Choose 2-3 objects

2. Take 30-50 photos of each

3. Train the model

4. Test it!

Tips:

• Vary angles and distances
• Try different lighting
• Attempt to trick your AI!

Visual: Screenshot of Teachable Machine interface with arrows pointing to key features

SLIDE 22: Key Vocabulary

Title: Words to Remember

Pixel: Tiny colored square that makes up digital images. Each has RGB values.

Pattern Recognition: Process of finding recurring features or arrangements in data.

Training Data: Set of labeled example images used to teach AI what to look for.

Classification: Assigning a label or category to an image based on its features.

Computer Vision: Field of AI that enables computers to interpret visual information.

Feature Extraction: Identifying important characteristics in an image (edges, colors, shapes).

Confidence Score: Percentage showing how certain AI is about its prediction.

Machine Learning: AI systems that learn patterns from examples rather than following rigid rules.

SLIDE 23: What We Learned Today

Title: Key Takeaways

Content:

✓ AI processes images as pixels (numerical data), not visual scenes

✓ Image recognition follows four steps: Input → Feature Extraction → Pattern Matching → Classification

✓ Training data quality determines AI accuracy

✓ AI is fast and consistent but lacks human understanding and context

✓ Real-world applications range from medicine to conservation

✓ Privacy and bias are important ethical concerns

Visual: Checkmarks next to each point with related icons

SLIDE 24: Looking Ahead

Title: The Future of Computer Vision

Content:

• AI vision technology is rapidly improving
• New applications emerging constantly
• YOU might invent the next breakthrough!

Possible Future:

• Medical diagnosis assistance
• Disaster response (finding people in rubble)
• Better accessibility tools
• Environmental monitoring
• Space exploration

Challenge: What problem could you solve with image recognition AI?

SLIDE 25: Closing Questions & Reflection

Title: Think About It...

Reflection Questions:

1. How has your understanding of AI vision changed today?
2. What surprised you most about how AI processes images?
3. What's one way you could use image recognition to help your community?
4. What concerns do you have about facial recognition technology?
5. How will you think differently about AI now?

Next Steps:

• Complete exit ticket
• Try extension activities
• Explore recommended resources
• Share what you learned!

SLIDE 26: Resources & Credits

Title: Continue Learning!

Recommended Tools:

• Google Teachable Machine: teachablemachine.withgoogle.com
• Google Quick, Draw!: quickdraw.withgoogle.com
• Microsoft Seeing AI app
• Google Lens

Learn More:

• AI for K-12 Initiative: ai4k12.org
• Code.org AI courses
• Khan Academy: Intro to AI

Credits:

• Lesson developed by Evolve AI Institute
• Images: [Attribution as needed]
• Thank you for learning with us!

TEACHER NOTES FOR PRESENTATION

Timing: Plan 2-3 minutes per slide average. Some slides (like examples) may be quicker, discussion slides longer.

Animation Suggestions:

• Slide 6: Animate the transformation from image → pixels → numbers
• Slide 7: Animate flowchart steps appearing one at a time
• Slide 13: Animate table rows appearing sequentially for comparison

Interactive Elements:

• Slide 2: Have students vote before revealing answers
• Slide 12: Let students guess what went wrong with each AI mistake
• Slide 19: Facilitate brief class discussion about ethics
• Slide 25: Give students 2 minutes to write reflections

Visual Resources Needed:

• High-quality example images showing AI successes and failures
• Screenshots of Teachable Machine interface
• Photos of real-world applications (medical imaging, camera traps, etc.)
• Icons for flowcharts and diagrams (camera, magnifying glass, puzzle pieces, label tag)
• Comparison graphics for human vs. computer vision

Color Scheme: Use consistent colors throughout:

• Primary: Blue tones for AI/technology elements
• Secondary: Green for success/positive examples
• Accent: Orange/yellow for important callouts
• Warning: Red for mistakes/concerns

Accessibility:

• Use high-contrast text (dark text on light background)
• Font size minimum 24pt for body text, 36pt+ for titles
• Include alt text for all images
• Use clear, simple language
• Avoid decorative fonts

Delivery Tips:

• Pause for questions after major concepts (slides 6, 7, 13)
• Show enthusiasm during AI mistakes—they're learning opportunities!
• Connect back to students' experiences and interests
• Use "we" language to include everyone in learning journey
• Encourage predictions before revealing answers