AI vs Machine Learning vs Deep Learning

Part 1 defined AI broadly. This part draws the real, precise boundaries between AI, Machine Learning, and Deep Learning — three genuinely nested concepts, not three separate, competing technologies.

The real, nested relationship

Artificial Intelligence (the broadest real category)
  └── Machine Learning (a specific, real approach to achieving AI)
        └── Deep Learning (a specific, real technique within ML)

Every deep learning system is genuinely a machine learning system, and every machine learning system is genuinely a form of AI — but the reverse isn't true. A rule-based chess engine from the 1990s is real AI, but it's not machine learning at all, since it followed explicit, hand-written rules rather than learning a pattern from real data.

Machine Learning: learning a real pattern from data, not being told the rule

Traditional programming: a human writes the explicit rule
  if email contains "free money": mark as spam

Machine learning: a human provides real, labeled examples
  (thousands of real emails, each marked spam or not-spam), and
  the system learns the actual pattern itself

This is the real, defining shift Machine Learning represents — instead of a programmer manually encoding every rule for detecting spam, a real ML system is shown many labeled real examples and learns statistical patterns that generalize to new, unseen emails. Genuine ML systems predate the current AI wave by decades — spam filters, recommendation systems, and credit scoring models have used real ML techniques for a long time before large language models existed.

Deep Learning: Machine Learning using real, layered neural networks

Machine Learning (the broad category) includes several real
  techniques:
  - Decision trees
  - Support vector machines
  - Neural networks — including DEEP neural networks (many real
    layers), which is specifically what "deep learning" refers to

Deep Learning specifically means using neural networks with many real, stacked layers (part 6 covers what a layer actually does) — "deep" refers literally to the real depth of that layered structure, not to any qualitative claim about how sophisticated or intelligent the result is. Deep learning is the real, specific technique behind image recognition, speech recognition, and — most relevant to this series — large language models.

Where large language models actually sit in this hierarchy

AI
  └── Machine Learning
        └── Deep Learning
              └── Large Language Models (this series' actual focus,
                    starting in part 5) — a specific, real application
                    of deep learning to language

Every large language model this series discusses — the ones behind real products like ChatGPT and Claude — is a genuine, specific instance of deep learning, which is itself a genuine, specific instance of machine learning, which is itself a genuine, specific instance of AI. Understanding this real nesting is what makes a sentence like "is ChatGPT AI or Machine Learning?" resolve cleanly — it's honestly both, simultaneously, at different real levels of the same hierarchy.

Why this precision actually matters for a real project

A real, imprecise brief: "add some AI to the product"
A real, precise brief: "add a large language model feature that
  answers customer questions using our product documentation"

The imprecise version leaves genuinely enormous room for interpretation — a simple, hand-coded recommendation rule technically qualifies as "AI" under part 1's broad definition. Being precise about which real layer of this hierarchy a project actually needs — as this series will be from part 5 onward, focused specifically on large language models — is what turns a vague goal into an actually buildable, scoped real feature.

Next: how Generative AI actually works — the real, specific capability that made large language models genuinely useful for creating new content, not just classifying existing content.