AI, ML, and AWS AI Services

Before going deep into generative AI, it helps to place it in the wider AI and machine-learning landscape and to know the managed AWS AI services that solve common tasks out of the box. This chapter reviews the core definitions and then tours the AWS services for search, language, speech, documents, and model building.

Verify service details

AWS service features, editions, billing, and Region availability change over time. The service-fact tables below are a quick reference; confirm specifics in the current AWS documentation before you design or budget around them.

Core definitions

Artificial Intelligence (AI) is the field of computer science devoted to solving cognitive problems associated with human intelligence, such as learning, problem-solving, and pattern recognition.

Machine Learning (ML) is the science of building algorithms and statistical models that perform tasks without explicit instructions, relying instead on patterns and inference learned from data. ML algorithms process large amounts of historical data, find patterns in it, and use those patterns to predict outcomes for new inputs. For example, you could train a medical application to flag cancer in x-ray images by learning from millions of past scans and their diagnoses.

Generative AI is a type of AI that creates new content and ideas, including conversations, stories, images, video, and music. Like all AI it is powered by ML models, but generative AI specifically uses very large models pre-trained on vast data, the foundation models (FMs) introduced in Chapter 1: Introduction to Generative AI. A generative model reuses what it learned to solve new problems, for example, learning English vocabulary and then composing a poem. Organizations apply it to digital assistants, media creation, product design, and more.

The nesting, again

AI is the broad field, ML is the data-driven subset that learns from examples, and generative AI is the slice of ML built on large foundation models. AWS offers managed services at each level, from task-specific AI APIs to full ML platforms.

Intelligent search: Amazon Kendra

Amazon Kendra is an ML-powered enterprise search service. Instead of matching keywords, it understands the context and meaning of a query and finds relevant information across scattered sources, document repositories, SharePoint, Salesforce, databases, file shares, and more. It learns an organization’s own vocabulary, can answer questions, and can return summaries rather than just a list of documents, which speeds up finding information and making decisions. It works using natural language processing (NLP) and ML over your content.

Table 1 Amazon Kendra at a glance

Attribute	Detail
Billing	Billed per hour for each Kendra index.
Editions	Enterprise Edition and Developer Edition.
Infrastructure type	Regional
Management type	Self-managed (service)
Service category	Machine learning

Text and language: NLP services

Natural Language Processing (NLP) is the area of AI focused on analyzing, understanding, and generating human language. AWS offers several task-specific, fully managed text and speech services you can call from the console or an API:

Service	What it does
Amazon Transcribe	Speech-to-text: converts spoken audio into written text (captions, dictation, call transcripts).
Amazon Polly	Text-to-speech: turns written text into lifelike spoken audio.
Amazon Textract	Extracts text, forms, and tables from scanned documents and images (beyond simple OCR).
Amazon Translate	Machine translation between languages for fast, scalable localization.

These map onto the single-modality tasks in AI Literacy and Responsible Use (speech-to-text, text-to-speech, image-to-text) delivered as managed APIs.

Conversational interfaces: Amazon Lex

Amazon Lex lets you build conversational interfaces, chatbots and voice assistants, into your applications. It uses the same deep-learning technology behind Amazon Alexa to provide natural language understanding (NLU) and automatic speech recognition. You define the conversation flow and language models (the intents), and Lex handles the NLP and speech recognition; when a user speaks or types, Lex identifies their intent and returns it to your app so it can respond or act. This makes applications more accessible by letting users interact in natural language instead of menus and forms.

Table 2 Amazon Lex at a glance

Attribute	Detail
Billing	Pay-as-you-go: charged for speech/text requests, speech intervals, and automated chatbot-design training minutes.
Infrastructure type	Regional
Management type	Fully managed
Service category	Machine learning

Document insights: Amazon Comprehend

Amazon Comprehend uses NLP to extract insights from text with no preprocessing required. It analyzes UTF-8 text to recognize entities, key phrases, language, sentiment, and other elements, so you can scan a document repository for key phrases, monitor social feeds for product mentions, or discover the topics in a set of documents. It relies on a continuously trained, pre-trained model, so you do not supply training data.

Benefits include: powerful NLP through a simple API (no text-analysis expertise needed), deep-learning-based accuracy that improves as AWS retrains across domains, and the scale to analyze millions of documents. This is the managed-service counterpart to the unstructured-text problem many organizations face in retail, finance, healthcare, and beyond.

Table 3 Amazon Comprehend at a glance

Attribute	Detail
Billing	Charged by the Comprehend option chosen and the type of API call.
Infrastructure type	Regional
Management type	Fully managed
Service category	Machine learning

Building models: SageMaker and Bedrock

Two services anchor ML on AWS, and they sit at different layers:

• Amazon SageMaker is the platform for building, training, and deploying your own ML models. It provides pre-built algorithms and models as starting points, so you can pick one and customize it without being an ML expert.

• Amazon Bedrock provides pre-trained foundation models through an API, which you can fine-tune to your needs without training from scratch. It is the backbone of this book; see Chapter 1: Introduction to Generative AI.

In short: reach for SageMaker when you want to build and operate models yourself, and Bedrock when you want to consume and adapt foundation models. The task-specific services above (Kendra, Lex, Comprehend, Transcribe, Polly, Textract, Translate) sit one level higher still, ready-made AI for a specific job.

Key takeaways

• AI -> ML -> generative AI is a nesting; AWS has managed services at each level.

• Amazon Kendra is intelligent, meaning-aware enterprise search.

• Transcribe, Polly, Textract, Translate are task-specific NLP/speech/document services; Amazon Lex builds chatbots and voice assistants; Amazon Comprehend extracts insights from text.

• SageMaker builds and operates custom models; Bedrock delivers and fine-tunes foundation models.

Attribution

Service descriptions are summarized from AWS training and documentation. Confirm current features, editions, and pricing on the official AWS documentation before relying on them.