What is AI?
Unit 2 of 8 · Module 1: Plan and prepare to develop AI solutions on Azure
Read this lesson
What we mean by AI
So, what do we actually mean by AI? Broadly, it's software that behaves in ways that feel human: it takes in information, makes sense of it, and responds. The definition has shifted over the years, as the technology has changed. Today, most AI solutions are built on machine learning models. A model captures the relationships hidden in huge amounts of data. That's what lets an application interpret input in different formats, reason over it, and generate a response or a prediction.
Generative AI and agents
Let's start with the capability everyone's talking about: generative AI. It's built on large language models, or LL Ms, that create original responses to natural language prompts. That's what powers chat apps and AI-assisted content creation. And increasingly, it's the foundation for agents. An agent combines a language model with focused instructions that define its job, and tools it can use to find knowledge and automate tasks. Model, instructions, tools: that's an agent.
Natural language processing
Those language models didn't come from nowhere. They grew out of natural language processing, a long-established field that uses statistical and semantic models to make sense of text. Think documents, emails, and social media posts. Generative models can handle a lot of that work now, but specialized text analysis still has its place. Techniques built on term frequency, plus task-specific models for classification, sentiment analysis, and summarization, are often the sharper tool.
Computer speech
Next, speech. Recognizing and synthesizing speech lets apps and agents talk with people naturally, with voice in and voice out. Modern speech models cope with background noise, interruptions, and many languages and accents. And speech isn't just for conversation. It's how you transcribe and analyze live or recorded audio, and how text gets read aloud, or translated in real time.
Computer vision
Then there's computer vision: understanding images, video, and live camera feeds. Picture an automated checkout that recognizes what's in a shopper's basket, so nobody has to scan a barcode or type in a quantity. And generative models are increasingly multimodal. They don't just read images; they can create them too, along with video.
Information extraction
Put these capabilities together and you get information extraction. Language reasoning, document understanding, vision, and speech, combined to pull key details out of documents, forms, images, and recordings. Take an expense claims app. It reads a scanned receipt and extracts the purchase date, each line item, and the total cost, automatically.
From capabilities to services
So that's the landscape: generative AI and agents, language, speech, vision, and information extraction. Here's why it matters for planning. Once you know which capabilities your application needs, it's much easier to identify the right AI services to provision, configure, and use. Next, we'll look at where those services live.
Facts from Microsoft's documentation: What is Microsoft Foundry?; What are Foundry Tools? (CC BY 4.0), rewritten in Loop's own words and examples. Narration, animation, and quiz by Loop. Microsoft, Azure, and Microsoft Foundry are trademarks of Microsoft. Loop is independent and not affiliated with or endorsed by Microsoft.