What is "AI"?

Cutting through the hype

"AI" is a marketing term, not a scientific one

In the past, "artificial intelligence" was a term used in academia to describe attempts to create systems with human-like capabilities for thought and reason.

However since 2023, it has been transformed into a marketing buzzword, and is now applied to anything; from TVs to vacuum cleaners. Features that have existed for years are now re-labelled as "AI" in order to sell them. Companies are adding "AI" to their products because it improves the share price.


How it works

AI companies have a vested interest in making "AI" seem extremely complicated and mysterious. But modern AI is basically a very expensive, very fancy version of predictive text.

On your smartphone if you type "good m", predictive text guesses that you probably want to type "good morning". It's been trained on what words come after others, and it guesses based on what is most likely.


How "AI" is created

Creating "large language model"-based AI can be broken down into three steps:

1. Get a huge amount of text

Download it from the web, from newspapers, books, forums. Typically companies have not cared about getting permission.

The aim is to find as many examples of natural-sounding text.

2. Find patterns in the text

Find which words are often said after others. This is called "training" and costs hundreds of millions of dollars and as much electricity as small countries use in a year.

3. Generate likely responses to an input

When given some text, the program will now output text that is the most statistically likely way to respond to that given text.

That's it.