AI & Machine Learning

Understanding Intelligence in the Digital Age

Prerequisites: Basic programming and web concepts, computational thinking

Focus: From Analytic to Generative AI

Learning Objectives

By the end of this module, you will be able to:

Distinguish between Analytic AI and Generative AI applications
Explain the three-step machine learning process and common pitfalls
Understand neural networks as foundational technology for modern AI
Effectively communicate with Large Language Models using structured prompting
Apply information literacy skills in AI-enhanced search environments

I

Introduction to Artificial Intelligence

20-25 minutes • Top-down approach: Broad AI landscape to focused applications

Understanding AI Through Two Primary Categories

We'll start with the big picture of AI, then focus on specific types and applications

Two Primary Categories of AI

Analytic AI

Definition: AI systems designed to identify patterns, trends, and apply rules or access knowledge systems without dynamic generative ability

Core Function: Analysis, classification, prediction based on existing data

Recommendation systems (Netflix, Spotify)
Fraud detection algorithms
Medical diagnostic tools
Traditional search engines

Generative AI

Definition: AI systems designed for creation and content generation

Core Function: Producing new content, text, images, code, etc.

ChatGPT, Claude (text generation)
DALL-E, Midjourney (image generation)
GitHub Copilot (code generation)

Applications Across Domains

Healthcare

Analytic: Diagnostic AI for pattern recognition in medical imaging

Generative: Medical report generation and treatment plan creation

Education

Analytic: Learning analytics and performance tracking

Generative: Personalized content creation and adaptive learning materials

Business

Analytic: Market analysis and trend prediction

Generative: Marketing copy generation and product descriptions

II

Machine Learning as Analytic AI

25-30 minutes • Bottom-up approach: Building understanding from fundamentals

Positioning ML within AI Framework

Machine Learning as a subset of Analytic AI
Focus on pattern recognition and prediction from data

The Three-Step ML Process

View Document ↗️

1

Data Collection

Gathering relevant, representative datasets

Quality considerations: completeness, accuracy, relevance

2

Training

Study Analogy: "Like studying for a test using practice problems"

Algorithm learns patterns from training data through iterative improvement

3

Prediction

Study Analogy: "Applying what you learned to solve new problems"

Making predictions on new, unseen data with performance evaluation

ML in Action: Real Examples

Netflix Movie Recommendations

View Document ↗️

Email Spam Detection

View Document ↗️

These examples demonstrate how the three-step process applies to different domains

Common ML Pitfalls

Overfitting

Cram Study Analogy: "Memorizing specific practice problems instead of understanding concepts"

Model performs well on training data but fails on new data

Real-world impact: Poor generalization to novel situations

Bias in Training Data

Biased Study Material Analogy: "Studying from materials that only cover certain perspectives"

Models inherit and amplify biases present in training data

Real-world impact: Discriminatory outcomes in hiring, lending, criminal justice

III

Neural Networks: Foundation for Modern AI

15-20 minutes • From simple concepts to complex applications

Conceptual Overview

Inspired by biological neural networks
Interconnected nodes (neurons) processing information
Pattern recognition through layered processing

Neural Networks: The Committee Analogy

Neural nets are analogous to committees that vote

Their weights determine how much their votes count in the final decision

View Document ↗️

Input layer: Junior members with equal votes analyzing key features (edges, corners, lines, curves)
Hidden layer 1: Specialists focusing on specific aspects (shape, texture, size)
Hidden layer 2: Senior analysts with broader insights for complex pattern recognition
Output layer: Decision makers announcing final weighted decision (Dog 85%, Cat 12%, Bird 3%)

From Simple to Complex

Bottom-up Progression

Single Neuron → Networks → Deep Networks

Bridge to Generative AI

Neural networks serve as the foundation for Large Language Models

Transition from Analytic AI applications to Generative AI capabilities

Activity: Training a Neural Net

Train Your Pet Recognition System!

Using Google's Teachable Machine, we'll create a basic pet classifier

Create basic pet classes (dogs, cats) and populate with images
Test with sparse training using students' shared pet images
Add more training data on specific dogs and cats from pre-existing labeled sets
Observe improved predictions on students' images
Demonstrate bias: unknown objects predicted as the class with more training data

Learning Objective

Experience firsthand how training data quality and quantity affects model performance

IV

Large Language Models and GPTs

30-35 minutes • Understanding and effectively using generative AI

Understanding LLMs as Generative AI

• High-throughput creative capacity
• Token-based prediction mechanisms
• Emergent abilities from scale

The Prediction Reality

Critical Understanding

When LLMs cite sources or provide information, they are making predictions based on training data, not accessing real-time databases

Implications:

Accuracy varies and requires verification
Sources may be hallucinated or misattributed
Fact-checking remains essential
Models can sound confident while being incorrect

Natural Language Interface Advantage

Communicate in natural language rather than specific code/syntax

Democratizes access to AI capabilities and enables rapid prototyping

Effective Prompting with CLEAR

A structured approach to communicating with AI

C

Clarity

Specific, well-defined goals

L

Logistics

Practical details, input types, steps, desired output

E

Examples

Demonstrate intended output, specify what you DON'T want

A

Audience

Target population characteristics and needs

R

Refinement

Iterative improvement through feedback

CLEAR Activity: Phase 1

Basic Prompt (5 minutes)

Students create a simple prompt and observe the results

Example Basic Prompt:

                        "Tell me about Python basics"
                    

Expected Result

Generic, broad responses that may not meet specific needs

Observe how vague prompts lead to vague responses

CLEAR Activity: Phase 2

CLEAR-Enhanced Prompt (7 minutes)

Apply the CLEAR framework to transform the basic prompt

Clarity: "Create a comprehensive Python basics reference guide"
Logistics: "Include fundamental data types, methods, keywords, and control structures with example usage"
Examples: "Format as a glossary with code snippets for each concept"
Audience: "For computer science students familiar with basic programming concepts"
Refinement: "Output as structured markdown with clear sections and subsections"

CLEAR Activity: Phase 3

Iteration (3 minutes)

Refine prompts based on initial outputs

Refinement Strategies:

Add specific formatting constraints (tables, bullet points)
Request different output lengths or detail levels
Specify tone or writing style
Include counter-examples of what NOT to include
Add context about intended use case

Key Insight

Iteration is crucial - first attempts rarely produce optimal results

Each refinement teaches you more about effective prompting

Information Literacy in the AI Era

How Search Has Changed with RAG

Retrieval-Augmented Generation (RAG)

LLMs accessing and synthesizing information from databases

Integration with search engines (Google's AI Overviews, Bing Chat)

Traditional Search

Returns list of links to relevant pages

User evaluates and synthesizes information

AI-Enhanced Search

Provides synthesized answers from multiple sources

AI pre-processes and summarizes information

Critical Information Literacy Skills

Why It Matters

AI systems can perpetuate misinformation and present confident-sounding but incorrect information

Verification Strategies:

Cross-reference sources: Don't rely on a single AI response
Check primary sources: Verify original research and data
Understand AI limitations: Knowledge cutoffs and training biases
Question confident claims: Higher confidence ≠ higher accuracy
Use traditional search: Compare AI summaries with original sources

Example: Google's AI Overviews

Synthesized information appears at top of search results - looks authoritative but may contain errors

Always scroll down to check original sources

Assessment Opportunities

Formative Assessment

CLEAR framework prompting exercise
Classification of AI examples as Analytic vs. Generative
ML process step identification
Neural network committee analogy explanation

Summative Assessment Options

Design a prompting strategy for a specific domain problem
Analyze potential biases in a given ML scenario
Compare traditional search vs. AI-enhanced search for research tasks
Create an information literacy checklist for AI-generated content

Summary & Next Steps

Key Takeaways

✓ AI encompasses both Analytic and Generative capabilities
✓ ML follows a three-step process with common pitfalls to avoid
✓ Neural networks power modern AI through layered pattern recognition
✓ CLEAR framework improves AI communication effectiveness
✓ Information literacy remains critical in AI-enhanced environments

Ready for AI Ethics

You now have the foundational understanding of AI systems

Next: Degenerative AI - Exploring the Dark Side of AI