01What AI Really Is
Cut through the hype. Understand AI, machine learning, deep learning and generative AI — what each one is, how they relate, and what they can and can't do for you.
What you'll learn
- Explain the difference between AI, machine learning, deep learning and generative AI
- Recognize the kinds of problems AI is good (and bad) at
- Describe the main types of generative AI models a business can use today
- Build a realistic mental model you'll use for the rest of the Academy
The nesting-doll picture
People use “AI” to mean a dozen different things. The clearest way to understand the vocabulary is as a set of nested circles, each one a more specific version of the last.
| Term | What it means | Everyday example |
|---|---|---|
| Artificial Intelligence | The whole field: any computer system doing things we'd normally call “intelligent” — perceiving, reasoning, deciding, using language. | A chess engine, a route planner, a chatbot |
| Machine Learning (ML) | Systems that learn patterns from data instead of following hand-written rules. | Your email spam filter; Netflix recommendations |
| Deep Learning | ML using large neural networks with many layers. Powers modern vision, speech and language. | Face unlock, voice typing, photo search |
| Generative AI | Deep learning models that create new content — text, images, audio, video, code. | Claude, ChatGPT, Gemini, image generators |
| Large Language Model (LLM) | A generative model trained on huge amounts of text that predicts the next word-piece (token). | The “brain” inside every modern chatbot and AI agent |
Rules vs. learning: why ML changed everything
Traditional software is written as explicit rules: if the invoice is over ₱50,000, require a manager's approval. That works well when the rules are clear. It breaks down for fuzzy tasks — “is this email a complaint?”, “what's in this photo?”, “summarize this contract” — where writing every rule by hand is impossible.
Machine learning flips the approach. Instead of writing rules, you show the system thousands or millions of examples and it learns the patterns itself. Large language models took this to an extreme: they learned from a large share of the public internet, books and code, and in the process picked up grammar, facts, reasoning patterns, styles and even basic programming.
What generative AI can do today
Text
Write, rewrite, summarize, translate, classify, extract data, answer questions, brainstorm.
Code
Write and fix code, explain codebases, build small apps and automations, write spreadsheet formulas.
Vision
Read receipts and screenshots, describe photos, check product images, digitize handwritten forms.
Speech
Transcribe meetings and calls (speech-to-text) and speak naturally (text-to-speech).
Images & video
Generate product mockups, social graphics, concept art and short clips.
Actions
When connected to tools, AI can search, update spreadsheets, send messages and run multi-step tasks — this is an agent.
What AI is genuinely good at — and where it struggles
Strong
- Language tasks: drafting, rewriting, summarizing, translating
- Turning messy input into structured output (emails → table rows)
- Classifying and routing (which department? how urgent?)
- Explaining concepts and answering questions from provided documents
- First drafts of almost anything — fast
- Working 24/7 at consistent quality once set up well
Weak / risky
- Exact facts it wasn't given — it may hallucinate (confidently invent)
- Recent events after its training cutoff (unless given search tools)
- Precise arithmetic over big datasets (use code or a spreadsheet)
- Knowing your business specifics unless you tell it
- High-stakes judgment with no human review
- Tasks where success can't be checked
A mental model that works: the brilliant new hire
Treat an LLM like an exceptionally well-read new employee on their first day. They're fast, articulate and know a bit about almost everything. But they:
- know nothing about your business, clients, prices or policies until you tell them;
- will sometimes fill gaps with plausible guesses instead of asking;
- do dramatically better work with clear instructions, examples and source material;
- need supervision on anything important — at least until you've verified their work.
Almost every technique in this Academy — prompting, RAG, tools, guardrails, human approval — is a way of managing that new hire well.
Narrow AI, general AI and the hype
Everything available today is still narrow in an important sense: models are very capable across many language tasks, but they have no persistent goals, don't learn from your conversation unless a system saves memory for them, and can fail in ways a human wouldn't. You'll hear terms like AGI (artificial general intelligence) in the news. For practical business use, ignore the debate and focus on one question: “Can this tool reliably do this specific task, and can I check the output?”
Key takeaways
- AI ⊃ Machine Learning ⊃ Deep Learning ⊃ Generative AI. LLMs are the generative models for text.
- ML learns patterns from examples instead of hand-written rules — that's what makes fuzzy language tasks automatable.
- LLMs excel at language and structure; they struggle with unprovided facts, exact math and unchecked high-stakes decisions.
- Manage AI like a brilliant new hire: give context, examples and review.
Knowledge check
0 / 3Q1Which statement best describes machine learning?
ML systems learn from data instead of explicit hand-written rules.
Q2An LLM confidently states a fake statistic. What is this called?
Hallucination is when a model produces plausible-sounding but false information.
Q3Which task is the riskiest to fully hand off to an LLM without review?
High-stakes decisions about people need human judgment and oversight.