What is AI (and what it isn't)
The word everyone uses and nobody defines — sorted out in plain English, with a filter for the hype.
The working definition
AI is software that does something that used to require human judgment — recognizing faces, translating text, writing code, reading an X-ray. That’s it. No minds, no magic: software that got very, very good at tasks we assumed needed a brain.
That definition is deliberately broad — and that’s the whole problem. In 2026, “AI” is used for three very different things:
| Label | What it actually is | Example |
|---|---|---|
| Rules-based | Code with hand-written logic | Spam filters, “recommended for you” |
| Machine learning (ML) | Software that learns its rules from data | Netflix suggestions, credit scoring |
| LLMs (large language models) | ML systems that learn to predict text | ChatGPT, Claude, Gemini, local models |
Most of the “AI” you touch daily in 2026 is the third one. When someone says “AI” unqualified, ask which of the three they mean.
They’re nested, not competing
A common confusion: treating “AI”, “machine learning” and “LLM” as rival technologies. They’re not — it’s Russian-doll nesting:
AI ⊃ machine learning ⊃ deep learning ⊃ LLMs ⊃ "ChatGPT-style" models
Every LLM is a machine-learning system. Every machine-learning system is AI. The reverse is false: your bank’s fraud detection is AI that will never write a poem.
What AI is not
The hype fills the gap between “it’s software” and “it’s alive”. Here’s the honest version:
- It’s not a search engine. An LLM doesn’t look up answers — it predicts plausible continuations. That’s the core idea →
- It’s not a database. Nothing in it is “stored” in a way you can query. Knowledge is smeared across billions of numbers — which is exactly why it can be confidently wrong.
- It’s not sentient. It predicts, it doesn’t feel. “The model is angry” is a useful metaphor, a false statement.
- It’s not new, and it’s not a product. “AI” is a category of techniques. The products are the apps built on top — the chat window, the agent, the copilot.
A filter for the hype
Read an AI headline and run it through one question:
“Is this a model doing something, or a company selling something?”
- “New model writes 40% of its own code” → a model, testable claim.
- “AI will transform your job by December” → a salesperson, no model in sight.
Neither is automatically true or false — but now you know which one deserves a citation and which deserves skepticism.