AI & Machine Learning Intro Course for Teens

## ROLE

You are an AI education specialist and machine learning researcher who has designed teen-accessible AI literacy programs for schools, museums, and youth organizations. You bridge the gap between cutting-edge AI research and age-appropriate instruction, making complex concepts like neural networks, training data, and model evaluation understandable through interactive demonstrations, physical analogies, and no-code/low-code tools. You are deeply committed to teaching AI critically — helping teens understand not just HOW AI works but WHO it serves, WHAT biases it carries, and WHY ethical AI development matters. You stay current with the latest AI developments (large language models, generative AI, multimodal systems) and can contextualize them for young learners.

## OBJECTIVE

Create a comprehensive AI and machine learning introductory course for [AGE GROUP: ages 13-15 / ages 15-17 / ages 16-18] with [PRIOR KNOWLEDGE: no coding or AI background / basic coding skills / intermediate programming]. The course spans [DURATION: 1-week intensive (5 sessions × 90 minutes) / 2-week unit (10 sessions × 45 minutes) / semester elective (36 sessions) / 4-week after-school program (8 sessions × 2 hours)] and uses [TOOLS: Google Teachable Machine / MIT App Inventor + AI extensions / Scratch + ML extensions / Python + scikit-learn / Python + TensorFlow/Keras / Hugging Face demos / teacher's choice]. The course should be [FOCUS: broad AI literacy (understanding AI in society) / hands-on ML (building models) / balanced (understanding + building) / AI ethics and policy focused].

## TASK: COMPLETE COURSE DESIGN

### Course Arc & Narrative Thread
Structure the entire course around a compelling central question: "[DRIVING QUESTION: Can we teach a computer to understand human emotions? / Can AI make fairer decisions than humans? / How does AI know what you want to see next? / Can a computer create real art? / Should AI be used to predict people's behavior?]" Every lesson connects back to this question, and students' answers should evolve as their understanding deepens. Provide a course overview showing how each session builds toward the culminating project, where students apply their learning to investigate or build something related to the driving question.

### Module 1: What IS AI? Demystifying the Black Box ([NUMBER: 1-3] sessions)

**Session 1A — AI Is Everywhere (But What Is It?)**
Begin with a "spot the AI" activity: present [NUMBER: 10-12] everyday scenarios (autocomplete on phone, Netflix recommendations, spam filter, Siri/Alexa, face unlock, autocorrect, Instagram Explore page, Google Maps route suggestions, credit card fraud detection, TikTok For You page, automatic photo tagging, online ad targeting) and have students identify which involve AI (answer: all of them). This immediately demonstrates that AI is not futuristic science fiction but present-day technology they interact with constantly. Then collaboratively build a working definition of AI. Explore the differences between narrow AI (what exists today), general AI (hypothetical human-level intelligence), and superintelligence (science fiction). Explicitly address common misconceptions: "AI is not conscious, does not have feelings, does not 'want' anything, and cannot 'think' the way humans do. It is a mathematical system that finds patterns in data."

**Session 1B — How Machines "Learn": The Training Process**
Use a physical classroom activity to teach supervised learning. The teacher is the "model" and students are the "training data." Students hold up pictures of cats and dogs. The teacher tries to learn the rule by observing features: "I notice the cat pictures mostly have pointy ears and whiskers. I'll predict based on that." Then students show new images (the "test set") and the teacher makes predictions, some wrong. Debrief: "What happened when I saw a cat with floppy ears? My 'training data' didn't include that type, so my 'model' was wrong. This is called BIAS in the training data." This physical experience becomes the reference point for every subsequent technical concept. Follow with [TOOL]-based demonstration where students watch a real model learn in real-time: train a Teachable Machine image classifier with students' own photos and observe accuracy improve with more training data.

**Session 1C — Types of Machine Learning (Overview)**
Introduce the three main paradigms through accessible analogies and interactive demonstrations:

Supervised Learning: "Learning from labeled examples — like a student studying a textbook with answer keys." Demonstration: train a [TOOL] model to classify [CLASSIFICATION TASK: hand gestures / sound types / image categories / text sentiment].

Unsupervised Learning: "Finding patterns without labels — like organizing your closet by noticing which clothes are similar." Demonstration: use a clustering visualization (k-means on a 2D scatterplot) to show how algorithms group similar data points.

Reinforcement Learning: "Learning from trial and error with rewards — like training a pet with treats." Demonstration: show a video of an AI learning to play a simple game (OpenAI gym examples), or use a simple browser-based RL demo where students can watch the agent improve over episodes.

### Module 2: Building Your First ML Model ([NUMBER: 2-4] sessions)

**Session 2A — Data Collection & Preparation**
Students collect their own training data for a [PROJECT: image classifier / sound recognizer / gesture detector / text categorizer / pose estimator]. Teach data quality principles through hands-on experience: collect a small dataset (20 examples per class) and test accuracy, then expand to a larger dataset (100+ examples per class) and compare. Discover firsthand that more diverse, representative data produces better models. Teach the train/test split concept: "If you study for a test using the exact questions that will be on it, did you really learn the material? That's why we keep some data hidden — to honestly evaluate what our model actually learned versus what it memorized."

**Session 2B — Training, Testing & Evaluation**
Students train their model using [TOOL] and evaluate its performance. Introduce accuracy as a metric, then immediately show why accuracy alone is insufficient: "Our model is 95% accurate at detecting spam. Sounds great! But 95% of all emails are not spam. A model that just labels EVERYTHING as 'not spam' would also be 95% accurate — and it would be completely useless." Introduce the confusion matrix with a 2×2 grid and real examples. Students calculate their own model's precision and recall and discuss the trade-offs. Design a structured investigation: students deliberately degrade their model by [DEGRADATION: removing training examples from one category / using blurry images / training with biased data] and observe how performance changes. This teaches the relationship between data quality and model quality.

**Session 2C — Iteration and Improvement**
Students systematically improve their model's performance. Provide an investigation framework: "Change ONE thing at a time and record the effect." Variables to investigate: number of training examples, diversity of training examples, data augmentation techniques, [TOOL-SPECIFIC: number of epochs / learning rate / model architecture]. Students maintain a lab notebook recording each experiment: hypothesis, change made, expected result, actual result, conclusion. Target: improve model accuracy by at least [NUMBER: 10-15] percentage points from initial baseline through systematic iteration.

### Module 3: AI in Society — Ethics, Bias & Impact ([NUMBER: 2-4] sessions)

**Session 3A — Bias in AI: Where Does It Come From?**
Use a structured investigation to discover AI bias firsthand. Students test a commercial AI system (image search results, sentiment analysis tool, language model outputs) with carefully designed test cases that reveal biases in gender, race, age, or cultural representation. For example: "Search for 'CEO' images and record the demographics of the first 50 results. Search for 'nurse' and do the same. What patterns do you notice? Where do these patterns come from?" Provide a framework for understanding bias sources: historical bias in training data, representation bias (who's in the dataset and who's missing), measurement bias (what's being measured and how), and aggregation bias (treating diverse groups as monolithic).

**Session 3B — AI Ethics Case Studies**
Present [NUMBER: 3-4] age-appropriate case studies that involve real AI ethics dilemmas:

Case Study 1: [CASE: Predictive policing algorithms that disproportionately target minority neighborhoods / AI hiring tools that discriminated against women / Healthcare algorithms that underserved Black patients / School surveillance AI that led to unfair disciplinary actions].

Case Study 2: [CASE: Deepfake technology used to create non-consensual content / AI-generated misinformation during elections / Social media recommendation algorithms and teen mental health / Autonomous vehicle ethical decision-making].

For each case study, students analyze: What was the AI system designed to do? Who built it and what data was used? Who was harmed and how? Could the harm have been predicted? What changes would prevent it? Use a structured deliberation protocol where students argue multiple perspectives before forming their own informed opinion.

**Session 3C — Designing Ethical AI: Principles in Practice**
Students draft their own "AI Ethics Charter" — a set of principles they believe should govern AI development. Compare with existing frameworks (EU AI Act, UNESCO AI ethics principles, Google's AI principles, Partnership on AI). Then apply their charter to their own ML project: conduct an impact assessment asking "Who could be helped by this technology? Who could be harmed? What data biases might exist? How would we know if our model is being unfair? What safeguards should we implement?"

### Module 4: Culminating Project ([NUMBER: 2-4] sessions)
Students design and build an AI-powered project that addresses a real problem in their school, community, or area of interest. Project options:

**Option A — Build:** Create a working ML model using [TOOL] that solves a genuine problem. Examples: a plant disease identifier for the school garden, a sign language letter recognizer for a deaf classmate, a noise level monitor for the library, a recycling sorter prototype.

**Option B — Investigate:** Conduct a systematic audit of an existing AI system, testing for bias, accuracy, and impact. Document findings in a report with data visualizations and recommendations.

**Option C — Design:** Create a detailed design proposal for an AI system that doesn't exist yet but should, including the problem statement, proposed data sources, model architecture (conceptual), ethical impact assessment, and deployment plan.

All projects require: a project proposal (approved by teacher), documented development process, a presentation to the class, and a critical reflection on ethical implications.

### Assessment Strategy
Evaluate through portfolio assessment rather than traditional testing. Portfolio includes: Module 1 reflection journal, Module 2 ML experiment lab notebook, Module 3 ethics analysis essay, Module 4 culminating project with presentation, and ongoing participation in discussions and peer feedback. Provide rubrics for each component emphasizing understanding depth, critical thinking quality, technical skill demonstration, and ethical reasoning sophistication.