What Is AI, Really? - AI Fundamentals

What This Course Will Do For You

If you're here, you've probably been hearing about AI everywhere -- at work, on the news, from your kids, from that one coworker who won't stop talking about ChatGPT. Maybe you've nodded along. Maybe you've even tried one of these tools and felt like you were missing something everyone else seemed to get.

This course is for you.

By the end of these 8 modules, you won't just know what AI is. You'll use it regularly and understand what it's doing when you do. You'll know which tools to reach for, what to ask them, when to trust their answers, and when to double-check. You'll go from "I keep hearing about AI" to "I use AI every day and understand what it's doing."

Here's how it works:

8 modules, each building on the last. No skipping around needed.
Things to try immediately in every module. This isn't a lecture series. It's a doing series.
No prerequisites. No coding. No math. No prior experience assumed. If you can use email, you can do this course.
No hype. AI is powerful and useful. It is also overhyped, often misunderstood, and occasionally wrong. You'll get an honest picture.

Each module takes about 15 minutes to read, plus however long you spend experimenting. That part is up to you. But the reading itself is designed to fit into a coffee break.

Ready? Let's start at the very beginning.

The One-Sentence Answer

People ask "What is AI?" and get answers that range from a shrug to a philosophy lecture. Let's skip both.

AI is a program that finds patterns in data and uses those patterns to generate new outputs.

That's it. Everything else -- the neural networks, the deep learning, the billions of parameters -- is detail about how it does that. But the what is simple: pattern finding, then pattern using.

Think about spell check. It has seen millions of correctly spelled words, so it knows that "accomodate" is probably a misspelling of "accommodate." It found a pattern (correct spelling patterns) and uses it to generate an output (a correction suggestion). AI does the same thing, just with vastly more data and more complex patterns.

When ChatGPT writes you an email, it's not "thinking" about what to say. It's using patterns it learned from reading billions of words to predict what words should come next in that particular situation. When Midjourney creates an image, it's using patterns from millions of images to predict what pixels should go where.

One sentence. Pattern finding. Pattern using. That's AI.

How AI Actually Works (The Simple Version)

Now that you know the one-sentence answer, let's see how it actually happens. There are two phases, and understanding them clears up a lot of confusion.

Phase 1: Training

Before AI can do anything useful, it needs to learn. It learns by example -- lots and lots of examples.

Imagine you wanted to learn to write well, and someone handed you every book ever published, every newspaper article ever written, every email ever sent, and every Wikipedia page. You read all of it. You don't memorize it word for word, but you start to notice things: how formal letters are structured, how recipes list ingredients before instructions, how news articles put the most important information first.

That's essentially what happens during AI training. The program is fed enormous amounts of data -- text from the internet, images from photo databases, audio recordings -- and it learns the statistical relationships within that data. Not the meaning. The patterns.

"Statistical relationships" sounds technical, but it just means: after seeing the word "peanut" billions of times, the program learns that "butter" follows it more often than "bicycle." After seeing millions of photos of cats, it learns that pointy ears and whiskers tend to appear together. It's counting and correlating at a scale no human could.

This training phase takes months and costs millions of dollars in computing power. That's why only a handful of companies can build these models from scratch. But once training is done, the result is a fixed model -- a big, frozen collection of learned patterns, ready to be used.

Phase 2: Inference

Inference is what happens when you use AI. You type a prompt, and the model generates a response. Here's what's actually going on:

You provide an input -- a question, a prompt, a description.
The model looks at the patterns it learned during training.
It predicts what should come next, one piece at a time.

For text AI, "one piece at a time" means one word (or part of a word) at a time. It's literally predicting the next word, then the next, then the next, hundreds of times to build a full response. For image AI, it's predicting groups of pixels. For voice AI, it's predicting sound wave patterns.

The Key Insight

Here's the part that really matters, and it's worth reading twice:

AI doesn't understand anything. It's pattern matching at enormous scale.

When ChatGPT writes a compelling paragraph about climate change, it's not because it understands climate science. It's because it has seen millions of discussions about climate change and can predict what a good paragraph on that topic looks like. The output seems intelligent because the patterns it learned came from intelligent sources -- humans writing on the internet. But the AI itself has no comprehension, no beliefs, no knowledge in the way you think of knowledge.

This distinction matters more than anything else in this course. It's the difference between using AI effectively and being misled by it. We'll come back to it again and again.

The Phone Autocomplete Analogy

If you've ever used autocomplete on your phone -- where it suggests the next word as you type -- you've used a tiny, simple version of the same technology. Your phone's keyboard learned patterns from everything you've typed: after "I'll see you," it probably suggests "there" or "soon" or "tomorrow." It doesn't know what you mean. It just knows what usually comes next.

Large language models like ChatGPT work on the same principle. The difference is scale: instead of learning from your text messages, they learned from the entire internet. Instead of suggesting one word, they can generate whole paragraphs, essays, and conversations. But the core mechanism is the same: predict what comes next based on patterns.

Scale changes what's possible, but it doesn't change what's happening under the hood.

What AI Is NOT

Understanding what AI isn't is just as important as understanding what it is. There's a lot of mythology out there, and it gets in the way of using AI well. Let's clear up the big ones.

It is NOT conscious, thinking, or alive

AI doesn't have thoughts, feelings, desires, or awareness. When ChatGPT says "I think" or "I feel," it's generating text that follows the pattern of how humans express thoughts and feelings. It's performing a very convincing impression of a thinking being without actually being one.

This is easy to forget, because the outputs are so humanlike. But remember: the patterns came from human writing. The AI learned to sound human because it was trained on humans. The humanity is in the training data, not in the AI itself.

It is NOT accessing the internet in real time (usually)

This surprises a lot of people. Most AI models have a training cutoff -- a date after which they haven't seen new information. ChatGPT, for example, doesn't know about things that happened after its last training update unless it has a feature specifically designed to search the web.

Some AI tools now include web search as an add-on feature. When you use those, the AI can look things up. But the core model itself is a snapshot of what it learned during training. It's like a very well-read person who hasn't picked up a newspaper since last year.

This matters because it means AI can give you outdated information. If you ask about something that changed recently, the AI might confidently give you an old answer. Always check the date if currency matters.

It is NOT always right

This is the one that trips people up the most. AI can generate answers that sound completely confident and are completely wrong. The term for this is "hallucination," and it's not a bug -- it's an inherent feature of how the technology works.

Remember, AI is predicting what should come next based on patterns. Sometimes the most likely-looking pattern isn't the correct answer. The AI doesn't have a fact-checking step. It doesn't "know" what's true and what isn't. It just knows what sounds plausible.

A hallucination might be small -- getting a date wrong by a year. Or it might be big -- inventing a book that doesn't exist, or confidently describing a legal case that never happened, or making up a recipe that would taste terrible. The confident tone stays the same regardless of accuracy.

This doesn't mean AI is useless. It means you should treat it like a knowledgeable but occasionally unreliable assistant. Great for drafts, ideas, and starting points. Not great for final answers you haven't verified.

It is NOT replacing humans

AI is a tool, not a replacement. This is worth saying plainly because the news loves the "robots taking your job" narrative, and it's misleading.

Yes, AI can automate specific tasks -- writing first drafts, summarizing documents, generating image options, answering common questions. Tasks, not jobs. A job is a collection of tasks, many of which require judgment, relationships, context, and adaptability that AI doesn't have.

The people who benefit most from AI aren't the ones who fear it or the ones who worship it. They're the ones who learn to use it as a tool alongside their own judgment. A calculator didn't replace accountants. It made them faster and more accurate. AI is doing the same thing for knowledge work.

Three Types of AI You'll Actually Encounter

AI comes in many forms, but for most people, three categories cover almost everything you'll use day to day. Understanding these categories helps you know what to expect from each tool and which one to reach for.

Text AI

What it does: Generates and understands human language. You type, it types back.

Examples: ChatGPT, Claude, Gemini, Copilot

What it's good for:

Writing -- emails, reports, blog posts, resumes, cover letters
Summarizing -- long documents, meeting notes, articles
Explaining -- breaking down complex topics in simple terms
Brainstorming -- generating ideas, lists, options
Editing -- improving text you've already written
Answering questions -- general knowledge, how-to guidance

What it's not good for:

Anything requiring verified, up-to-the-minute facts without checking
Tasks requiring personal judgment it doesn't have context for
Anything where a wrong answer could cause real harm (medical, legal, financial decisions)

Text AI is probably where you'll spend most of your time, and it's where this course focuses most of its attention. It's the most versatile category and the one most likely to change how you work day to day.

Image AI

What it does: Creates and edits images from text descriptions. You describe it, it draws it.

Examples: DALL-E, Midjourney, Stable Diffusion, Adobe Firefly

What it's good for:

Creating original images -- illustrations, concept art, mockups
Exploring visual ideas quickly -- iterate on designs in seconds
Editing images -- removing objects, extending backgrounds, changing styles
Generating variations -- take one concept and see twenty versions

What it's not good for:

Pixel-perfect design work (it's generative, not precise)
Consistent characters or branding across multiple images (it varies each time)
Replacing professional photography or design when quality matters
Anything where accuracy matters (it can distort text, hands, architectural details)

Voice AI

What it does: Understands and generates spoken language. You talk, it listens and talks back. Or it listens and writes it down. Or it reads text out loud.

Examples: Siri, Alexa, Google Assistant, Otter.ai, Apple Dictation, transcription tools

What it's good for:

Quick tasks -- setting timers, reminders, simple questions
Transcription -- turning spoken words into text
Text-to-speech -- having written content read aloud
Translation -- real-time speech translation in some tools

What it's not good for:

Complex multi-step requests (voice assistants struggle with nuance)
Anything requiring privacy (these devices listen, by design)
Detailed, nuanced conversations (text AI is much better for this)

Each of these categories will get deeper treatment later in the course. For now, the important thing is just knowing they exist and roughly what each one does.

Your First AI Conversation

Time to stop reading and start doing. This section walks you through having your first conversation with AI. If you've already used ChatGPT or a similar tool, feel free to skip ahead or try the prompts anyway -- you might be surprised by what you get.

Step 1: Go to ChatGPT.com

Open your browser and go to chatgpt.com. This is the free version of the most popular text AI. You don't need to download anything.

Step 2: Create a Free Account

Click "Sign up." You can use your Google account, Microsoft account, or an email address. The free tier is more than enough for everything in this course.

Step 3: Try These Five Prompts

Type each one exactly as written, read the response, then try the next one. Don't overthink it -- just see what happens.

Prompt 1: "Explain quantum computing to me like I'm 12"

This is a classic "explain something complex simply" prompt. Notice how the AI adjusts its language to match the "like I'm 12" instruction. It won't be perfect, but it'll be surprisingly readable. That adaptability is one of AI's strongest features.

Prompt 2: "Write a polite email declining a meeting"

Now you're seeing AI do real work -- generating something you could actually use. The email will be professional and reasonable. It won't know the specifics of your situation, but it'll give you a solid first draft you could customize in seconds. That's the workflow: AI drafts, you refine.

Prompt 3: "What are 5 quick dinner ideas with chicken, rice, and broccoli?"

This shows AI's ability to work with constraints. You gave it three ingredients and asked for ideas. It'll give you five plausible recipes. Some might be good. Some might be boring. But you went from "what's for dinner?" to "here are some options" in about three seconds.

Prompt 4: "Summarize the key points of the US Constitution in 200 words"

Notice the specific word count. AI can follow format instructions reasonably well -- not perfectly, but close enough. Also notice that it handles the summarization task competently. It's pulling from its training data, which includes plenty of content about the Constitution. For well-documented topics, AI tends to be more reliable.

Prompt 5: "Help me plan a 3-day weekend trip to Chicago"

This one shows AI at its most useful: a planning assistant that generates structured, detailed output instantly. You'll get a day-by-day plan with restaurants, attractions, and logistics. Is it the best possible Chicago itinerary? No. Is it a perfectly good starting point that would have taken you an hour to research? Absolutely.

What You'll Notice

After trying these five prompts, you'll probably notice three things:

It's fast. Each response takes a few seconds. What would have taken you 30 minutes of writing or researching happens almost instantly. Speed is one of AI's genuine advantages.

It's helpful. The outputs are genuinely useful starting points. Not final products -- you'll want to review and customize -- but solid first drafts that save you the blank-page problem.

It's not perfect. You'll spot something off. Maybe a recipe that sounds weird. Maybe a Chicago recommendation that's closed. Maybe a summary that misses something important. This is normal. This is expected. This is why you stay in the loop as the editor, not the author.

That last point is the most important thing to take away from your first conversation. AI is not a magic answer machine. It's a very fast, very helpful draft machine. The final product always needs a human.

Key Takeaways from Module 1

Here's what you should walk away with:

AI is pattern matching at scale. It finds patterns in enormous amounts of data, then uses those patterns to generate new outputs. It doesn't understand, think, or reason. It predicts.
AI has two phases: training and inference. Training is the expensive, months-long process of learning patterns from data. Inference is what happens when you use it -- the model predicting output based on what it learned.
AI can be confidently wrong. Because it generates plausible-sounding output based on patterns, not truth, it can and does produce incorrect information. Always verify anything that matters.
Three types of AI cover most everyday use: text, image, and voice. Text AI (ChatGPT, Claude, Gemini) is the most versatile and where you'll likely spend the most time.
AI is a tool, not a replacement. It's best used as a fast first draft generator and idea partner, with a human always reviewing and refining the output.

What's Next

In Module 2, we step back and look at the bigger picture: The AI Landscape. You'll learn who the major players are, how the tools differ from each other, and which ones are worth your time. You'll also learn why the same question can get very different answers from different AIs -- and how to pick the right one for the job.

By the end of Module 2, you won't just know what AI is. You'll know which AI to use.