· 5 min read

What is AI?

#ai-literacy #ai #agents
A blank ruled index card with an arrow and a stack of order slips covered in tally marks, inside a dashed boundary with a check mark; below, white tulips in a kraft-paper cone
On this page

Sam wants to send flowers to his wife, Elena, for her birthday on Friday. He can spend $60, including delivery and tax. The flowers must arrive that day. Elena loves tulips.

He opens a flower-delivery site. The first row is called “Birthday favorites”, and every bouquet in it is roses. Lower down, another row says: “Customers who bought tulips also bought…”

A computer chose both of these rows. Is that AI? And does the site know Sam?

Sam and Elena are not real people. But their problem is familiar: buying something specific, with a fixed budget, by a fixed date.

This is the first of eight short posts. They follow Sam from asking a chatbot for help to letting it buy the flowers. (A chatbot is a program you talk to by typing messages.) In this post, I’ll give a short introduction to AI, and a brief history of how we got here.

AI is older than your computer

If you first used AI through ChatGPT, it may seem new. It isn’t. John McCarthy created the term “artificial intelligence” in a 1955 proposal for a workshop at Dartmouth College. The workshop met in the summer of 1956. The Computer History Museum calls it the event that started the field. That was 25 years before IBM’s personal computer.

To understand Sam’s two rows, it helps to look at two ideas that have shaped AI for decades: giving computers rules, and helping them find patterns in data. Below are some important moments, shown next to the changes in how everyone else came to use computers.

DENDRAL and Amazon show two different kinds of AI. One follows knowledge that experts wrote down. The other finds patterns in data. Sam’s two rows help us see the difference.

Two ways to make a computer decide

Think about a new assistant in a flower shop. On the first day, the manager gives them a card: “For birthdays, show roses first.” Later, the most experienced worker gives them more rules, like which flowers shouldn’t travel in hot weather.

After a few months, the assistant notices something nobody told them: people who buy tulips often buy a vase too. They learned that pattern by watching.

A computer can work in both ways. But be careful: one simple rule, like the birthday row, is normal software, not AI. DENDRAL shows the AI version: many “if-then” rules from chemistry experts, and a computer that reasons through them.

In a system that learns from past orders, nobody has to choose every pair by hand. The system finds the links in past purchases. Amazon explains the idea like this: “item B is related to item A if customers who buy A are unusually likely to buy B as well.”

One product can use both, plus normal software rules. So this is how I describe AI:

AI is the part of computing that tries to make machines do things we connect with intelligence: find patterns, reason, and understand or write language. Over the years, people have tried different ways to do this. Writing rules by hand and learning from data are two of the biggest.

People disagree about where normal software ends and AI begins. Many products call themselves AI. I find it more useful to ask: what kind of decision is this?

Does it know Sam?

Not from these two rows. In this example, neither row uses any information about Sam or Elena. The birthday rule put roses first. The learned pattern links tulips with what other customers bought. Knowing what people usually buy together is not the same as knowing what one person wants.

Either kind of AI could use personal information, if the site had it. These two rows don’t show Sam whether it does. What a system keeps about you, and who can see it, comes in a later post.

A question to ask

The next time a product says it uses AI, ask what it actually does. Does it follow rules someone wrote? Does it find patterns in data? Or does it combine several ways? And how does anyone know its results are useful?

If you build or manage these systems, ask one more thing. Each part needs someone responsible for it. Who checks that the rules still match the job? Who checks that the learned patterns give people useful results?

These two rows can suggest bouquets. But they can’t answer Sam’s full request: tulips, under $60 including delivery and tax, delivered on Friday.

So Sam opens a chatbot and types exactly that. In a few seconds, it replies in everyday language, with a bouquet and a price. Where does that answer come from? And why can it sound correct but still be wrong? That is the next post.

Sources

Sources checked on 25 October 2026.

continue reading

Is your data layer ready for agents? The case for Lakebase

Agents need a disposable workspace and a durable record. Here's where Databricks Lakebase fits, in plain terms, and what it still leaves to you.

When the database's newest user is an agent

When a whole company books travel through agents, the database gets a new kind of user: many, fast, and often wrong. What that asks of the data layer, and why branching alone isn't enough.

What an agent remembers

An agent books your flight, then crashes before writing it down. When it restarts, does it book again? What an agent remembers, and why some of it has to survive a crash.