1: Fundamentals of AI and ML of scored content

Learning Objectives

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

• Explain basic AI concepts and terminologies, including when AI/ML is and is not appropriate, and the methods used to put a model in production.
• Define foundational AI terms (AI, ML, deep learning, neural networks, computer vision, NLP, model, algorithm, training and inferencing, bias, fairness, fit, LLM, GenAI, agentic AI), and select the appropriate AI/ML techniques (regression, classification, clustering) for a stated use case, while mapping pipeline stages to relevant AWS services (Amazon Bedrock, Amazon Q, Amazon QuickSight Q, Kiro, Amazon SageMaker AI).
• Describe similarities and differences between AI, ML, GenAI, deep learning, and agentic AI, identify real-world AI applications, and explain fundamental MLOps concepts including experimentation, repeatability, scalability, technical debt, production readiness, model monitoring, and model re-training.
• Describe the four inferencing modes (batch, real-time, asynchronous, serverless), the capabilities of AWS managed AI services (Amazon SageMaker AI, Amazon Transcribe, Amazon Translate, Amazon Comprehend, Amazon Lex, Amazon Polly), and the model and business metrics used to evaluate ML systems (accuracy, precision, recall, F1, cost per user, development cost, ROI).
• Differentiate the data types used in AI models (labeled vs. unlabeled, tabular, time series, image, text, structured vs. unstructured), and decide when traditional ML or a foundation model fits, given regulatory, explainability, and operational constraints.
• Describe supervised, unsupervised, and reinforcement learning, and walk through the AI/ML development lifecycle from business framing to retraining.
• Identify practical use cases for AI and the components of an AI/ML pipeline (data collection, processing, feature engineering, training, evaluation, deployment, monitoring).
• Recognize applications where AI/ML adds value (assistive decisioning, scalability, automation) and describe sources of foundation models (open source pre-trained models, training custom models).

Executive Summary

• AI is the umbrella, ML is a learned subset of AI, deep learning is a multi-layer neural subset of ML, generative AI uses large pretrained models to produce content, and agentic AI wraps a model in a goal-driven planning loop with tool use. Knowing the hierarchy is worth several exam points by itself.
• AWS exposes four inferencing modes on Amazon SageMaker AI (real-time, serverless, asynchronous, batch transform), and Amazon Bedrock provides synchronous InvokeModel, streaming InvokeModelWithResponseStream, and asynchronous batch inference. The right mode is driven by payload size, latency budget, and request shape.
• Choosing between a traditional ML model and a foundation model is a regulatory and operational decision as much as a technical one. Explainability, deterministic behavior, dataset volume, and total cost of ownership often push regulated workloads to classical algorithms.
• Production AI on AWS depends on MLOps discipline: versioned data and models, repeatable pipelines, monitored endpoints, and a retraining trigger. Without it, model drift will quietly degrade business metrics.

Assumptions

• All AWS service names follow the AWS naming convention from section K of the system prompt (Amazon for managed services, AWS for management/governance services).
• Region examples assume us-east-1 unless stated otherwise; quotas and feature availability vary by Region and should be checked against the AWS service quotas console for production design.
• The reader has practitioner familiarity with AWS Identity and Access Management (IAM), Amazon S3, AWS Lambda, and the AWS shared responsibility model; basic AWS terminology is not redefined.
• Code samples are illustrative; fictional account IDs (111122223333), bucket names (acme-ml-prod), and Region (us-east-1) are used consistently.