If AI Isn’t New, Why Is It Everywhere Now?

You tell an app, “Write a friendly invitation to my birthday at home on Saturday.” A draft appears. “Make it warmer,” you add. The wording changes.

The surprise in this fictional example isn’t just the draft. You can explain what you want in your own words, without special commands or complicated menus.

AI can feel like an overnight invention. Yet decades of work sit behind that conversation. Both its capabilities and how we access them have changed.

The phrase “artificial intelligence” appeared in the title of a research proposal in 1955. The workshop held at Dartmouth College in the United States the following year is considered a founding milestone. Researchers explored questions such as whether machines could use language and solve problems.

Useful everyday systems took decades to develop, with periods of optimism followed by disappointment.

Artificial intelligenceⓘ — short explanation was already part of daily life before chat apps became familiar. Think of your inbox: while you read a friend’s invitation, a system was trying to filter unwanted messages.

In a Gmail announcement dated 9 July 2015, Google described how it was already using systems that learned from examples to filter spam.

Imagine showing a system emails labelled “spam” or “not spam.” It learns relationships between words and other signals to help classify new messages. For example, “gift” might appear in both a deceptive advert and a friend’s birthday message. That word alone may not be enough.

Learning relationships from examples like these is called machine learningⓘ — short explanation. It can still make mistakes: your friend’s message might be filtered out.

Several developments reinforced each other: more digital examples to learn from, more powerful computers to process them, and better learning methods to use those resources.

“Computers got faster” is only part of the answer. More computing power doesn’t automatically turn inaccurate or unsuitable data into useful information. Data quality, methods and evaluation matter too.

For users, another change was how to interact with these systems. OpenAI introduced ChatGPT as a research preview on 30 November 2022. People could type a question, then ask a follow-up. AI didn’t begin that day; conversation became a more visible way to try it.

Back to the invitation. Generative AIⓘ — short explanation uses relationships between words and phrases learned from text to create a draft that responds to your request. For example: “I’m celebrating my birthday at home on Saturday. Come over for a laugh and a catch-up. I’d love to see you!”

Ask for a shorter version and it might write, “Birthday at mine this Saturday. Come join us!” Here, its job is to create content, rather than filter an incoming message.

These concepts are connected: generative AI also uses machine learning. One describes an approach to learning; the other describes the ability to create content.

Filtering and writing can coexist in one app. Conversation isn’t always the best interface, either. You wouldn’t want to discuss every email individually just to keep your inbox tidy.

When assessing a new app, look for two kinds of change. Does it do something it couldn’t do before? Or does it make an existing capability easier to use? Both can be useful, for different reasons.

For a two-minute experiment, try this request in a chat app you can access:

Check three things: Does the tone suit you? Is it within 40 words? Has it added a time or address you didn’t provide? Then ask for one change: “Use simpler wording.” No personal details are needed.

This shows how easy conversational guidance can be, while reminding you that easy access doesn’t guarantee a perfect result. If you prefer not to use an app, write the request and checking criteria on paper.

AI’s visibility today reflects years of progress meeting an everyday way to ask for help. Understanding how a feature changes your task matters more than memorising dates.

What does “learning from examples” mean? Next, we’ll compare giving a computer explicit rules with showing it examples.