Chapter 13

13.1 Artificial Intelligence

• Definitions:

  • Artificial Intelligence (AI): Development and application of algorithms and models to mimic human thought processes.

  • Example: Neural networks are a primary model in AI, crucial for tasks such as data classification, prediction, and output generation.

• Applications of Artificial Intelligence:

  • Self-driving cars

  • Facial recognition

  • Productivity tools (e.g., GitHub Copilot)

  • Movies and TV shows generation

  • Online shopping (customer service bots, recommendations)

• Key Concepts:

  • Participation Activity 13.1.2:

  1. AI mimics human thought using algorithms and models.

  2. AI has many applications in the real world.

  3. Inputs for AI include Text, Video, Images (All of the above).

  4. AI systems can:

     • Write responses to prompts/questions

     • (Do not depend on training data, have emotions).

13.1.3 AI Domains

• Six Major Domains of AI:

  1. Machine Learning: Uses algorithms for predictions, identifying patterns in data.

  2. Computer Vision: Extracts meaning from images/videos, interprets visual information.

  3. Natural Language Processing: Understands and interprets human language/text.

  4. Knowledge Representation: Framework for representing knowledge and information.

  5. Automated Reasoning: Techniques to solve conceptual problems using logic.

  6. Robotics: Involves design, construction, and operation of robots for tasks.

• Machine Learning Example: Predicting medical conditions based on patient health history.

• Participation Activity 13.1.4: Demonstrates task assignment to respective AI domains:

  • Describe retail product features: Natural Language Processing

  • Predict purchases: Machine Learning

  • Track customer movements: Computer Vision

13.2 Deep Learning

• Deep Learning:

  • Describes models with numerous parameters (e.g., Artificial Neural Networks).

  • Models difficult to interpret; capable of capturing complex relationships between inputs and outputs.

  • Rise of deep learning linked to advanced computational resources since the 2000s.

• Activity 13.1.5: Parameters of various models:

  • Early Neural Networks: 1,000 parameters

  • GPT models: Increasing from 117 million (GPT) to 1.7 trillion parameters (GPT-4).

• Participation Activity 13.1.6:

  • Deep learning models capture complex relationships.

  • “Deep” in deep learning refers to layers in models.

  • Deep learning is used in all domains of AI.

• Ethical Considerations: AI's growth raises job concerns — 75% of respondents fear job opportunities will decrease due to AI. When used ethical, AI aids daily lives.

• Guidelines for AI Use:

  • Check for accuracy

  • Avoid direct copying of AI outputs

  • Recognize biases and outdated results in AI content

  • Be transparent about AI usage.

13.2 Machine Learning Overview

• Machine Learning: A subset of AI focused on algorithms/models for outcomes and pattern recognition.

  • Involves computer science, mathematics, statistics for predictions.

  • Applications include image recognition, business analytics, and healthcare.

• Machine Learning Process:

  1. Model Training: Estimate parameters from training data.

  2. Model Validation: Check algorithm assumptions, refine parameters.

  3. Model Evaluation: Test performance using unseen data.

  4. Model Interpretation: Explain decision-making processes in models.

• Participation Activity 13.2.2:

  1. Machine Learning predicts using algorithms and data.

  2. Mathematical functions describing relationships are called models.

  3. Optimization minimizes prediction errors during model training.

• Machine Learning Types:

  • Supervised Learning: Predicts output based on known inputs.

  • Unsupervised Learning: Finds patterns without known outputs.

  • Reinforcement Learning: Adjusts actions based on past responses for optimal outcomes.

13.3 Computer Vision

• Definition: Algorithms and models designed to extract information from images/videos and generate new content.

  • Applications include self-driving technology, facial recognition, and image editing.

• Major Computer Vision Tasks:

  1. Image Classification: Identifying an image category.

  2. Image Segmentation: Dividing images into distinct regions.

  3. Object Detection: Locating and classifying objects in an image.

  4. Image Captioning: Generating descriptive text based on an image.

  5. Text-to-Image Generation: Creating images based on text descriptions.

  6. Image-to-Image Generation: Modifying images based on existing images.

• Convolutional Neural Networks (CNNs):

  • Use convolutional layers to filter images, identifying features like edges.

  • A pooling layer reduces dimensions, followed by flattening the image for neural network input.

• Limitations:

  • Requires substantial training data and computational power.

  • Often lacks interpretability.

13.4 Natural Language Processing (NLP)

• Definition: AI subfield focusing on algorithms to interpret and generate human language through structured data extraction from unstructured text.

• Key Components:

  • Tokenization: Splits text into tokens.

  • Stemming: Reduces words to their base form.

  • Lemmatization: Similar to stemming but uses a dictionary for base forms.

  • Part-of-Speech Tagging: Categorizes words as nouns, verbs, etc.

• Main Applications:

  • Text Classification: Assigning labels to text.

  • Sentiment Analysis: Classifying sentiment behind a text.

  • Text Summarization: Creating condensed versions of longer texts.

  • Text Generation: Generating new text based on prompts.

  • Language Translation: Translating text from one language to another.

• Transformers: Incorporates self-attention for sequential data analysis; foundational for large language models like BERT/GPT.

• Ethical Considerations: Involves biases in training data affecting output fairness and accuracy. Regulatory frameworks emerging for ethical AI use.

• Conclusion: The continuous evolution of AI and its applications present significant benefits alongside ethical implications that require attention and regulation.