01 / What

What is AI, really?

No math. No buzzwords. Just a straight answer to the question everyone’s too polite to ask out loud.

AI IN ONE SENTENCE

AI is software that learns patterns from huge amounts of examples, then uses those patterns to make predictions — the next word, the likely answer, the best match — instead of following rules a person wrote by hand.

The plain version

For decades, software did exactly what it was told. A programmer wrote the rules — if this, then that — and the computer followed them. Useful, but brittle. It couldn’t handle anything nobody had thought to write a rule for.

Modern AI flips that around. Instead of being given rules, it’s shown millions of examples and figures out the patterns on its own. Show it enough emails marked “spam” and it learns what spam tends to look like — no one had to spell it out.

The systems making headlines today — the ones that write, summarize, and answer questions — are doing a sophisticated version of exactly that: predicting what should come next, based on everything they’ve seen before.

It depends who you ask

NINE PERSPECTIVES

The word “AI” means something genuinely different to every institution that uses it. That isn’t a communications problem — it’s a power problem. Whoever controls the definition controls the conversation. Tap each to expand.

{{ d.num }} {{ d.tag }}{{ d.headline }} +

{{ d.body }}

{{ d.quote }}

{{ d.why }}
THE CORE PROBLEM

These nine aren’t competing versions of one truth. They reflect genuinely different interests and stakes. Until we share a stable vocabulary for what these systems actually are, the confusion isn’t a bug in how people think about AI — it’s a structural feature of the conversation.

Four words worth knowing

GLOSSARY
MACHINE LEARNING

The broad idea: software that improves at a task by learning from data, rather than being explicitly programmed for it.

MODEL

The thing that does the learning. After training, it’s a big file full of patterns you can feed new inputs to and get predictions back.

LLM

A “large language model” — trained on enormous amounts of text to predict the next word. It’s what powers chatbots and writing assistants.

TRAINING

The expensive, one-time process of showing a model its examples so it can learn the patterns. Using it afterward is cheap and fast.

And what it isn’t

It isn’t thinking

It has no understanding or intent. It’s pattern-matching at a scale that can look like understanding — which is exactly why it’s easy to over-trust.

It isn’t always right

A confident answer can still be wrong. These systems predict what sounds plausible, not what’s true. Verification is still your job.

It isn’t coming for everything

It’s very good at some narrow things and useless at others. Knowing the difference is most of the battle.

THE ECOSYSTEM

“AI” is an umbrella. Here’s what’s under it.

When people say “AI,” they could mean any of a dozen different fields. Most of them are nested inside one another — and most of the recent excitement lives in one small corner. A quick map so the next vendor pitch makes sense.

ARTIFICIAL INTELLIGENCE
MACHINE LEARNING
DEEP LEARNING
GENERATIVE AI · LLMs
The chatbots and image tools in the headlines live here — the innermost layer.
A

AI is the whole ambition: machines doing things that used to need a human mind.

B

Machine learning is the approach that won: instead of hand-written rules, learn the patterns from data.

C

Deep learning stacks those patterns in “neural networks” — the leap that made the last few years possible.

D

Generative AI is one thing deep learning turned out to be great at — and what most people now picture when they hear “AI.”

A toolbox, not one brain

Most people picture AI as a single super-intelligent system. In reality it’s a toolbox — specialized models for text, images, video, speech, and more, wired together into workflows. One drafts, one pictures, one transcribes, one calls APIs. The power comes from the orchestration, not any one part.

THE BUILDING BLOCKS

NINE COMPONENT TYPES

Language Models

Take text in (now often images and audio too) and produce text out — writing, summarizing, translating, answering, coding. What most people mean by “AI.”

CHATGPT · CLAUDE · GEMINI · LLAMA

Image Generation

Turn a text prompt into an image — concept art, mockups, marketing visuals, rapid iteration. Trained on millions of image-text pairs.

DALL·E · MIDJOURNEY · STABLE DIFFUSION

Video Generation

Create short clips and animation from text or reference footage. Computationally expensive and still short — but improving fast.

RUNWAY · SORA

AI Agents

Don’t just answer — plan and act. Given a goal, an agent breaks it into steps, calls tools and APIs, and iterates until it’s done.

AUTOGEN · LANGCHAIN AGENTS

Speech

Two directions: speech-to-text for transcription and captions, and text-to-speech for narration and voice assistants. Chain them and you get a “talking” AI.

WHISPER · ELEVENLABS

Multimodal Models

One model that takes text, images, and audio together — no separate pipeline. Closer to how people actually communicate.

GPT-4O · CLAUDE 3.5 · GEMINI PRO

Retrieval (RAG)

Bolt a search step onto a model so answers are grounded in your own documents — current, specific, and cited. Only as good as the index behind it.

PERPLEXITY · GLEAN · ENTERPRISE SEARCH

Memory & Personalization

Store preferences and past interactions so responses carry context across sessions. Useful — and a real privacy question about who owns that history.

ASSISTANT MEMORY · CRM PROFILES

Governance & Safety

The oversight layer: detect bias, filter harmful output, watermark generated media, audit usage. What keeps the rest accountable at scale.

MODERATION · WATERMARKING · AUDITING

Orchestration — the glue

Frameworks that connect models, search, and APIs into one working application. Without them, every AI product is hand-written glue code.

LANGCHAIN · AUTOGEN

Hardware — the engine

It all runs on chips tuned for parallel math. Access to compute is now an economic and geopolitical lever — and a real energy cost.

NVIDIA GPUS · GOOGLE TPUS

Two faces of AI

In daily life it shows up two very different ways — and the difference is the whole ballgame for trust.

THE AI YOU REACH INTO

Tools you deliberately open. You prompt it, you read the output, you decide what to keep. You’re in control and you know AI is involved.

CHATBOTS · IMAGE TOOLS · CODE ASSISTANTS · TRANSCRIPTION
THE AI THAT REACHES INTO YOU

Ambient systems shaping your world in the background — usually without you noticing. Hidden objectives, no real consent, quiet feedback loops. The kind worth watching closely.

FEED RANKING · AD TARGETING · RECOMMENDATIONS · YOUR MAP’S ROUTE

Useful rule of thumb: the component matters less than the question. What is it doing, how does it fail, and who’s checking? That holds no matter which corner of the toolbox a system comes from.

NEXT — 02 / WHY
So why should you care?