What is AI, and what it isn’t

AI is not one thing. It keeps changing.

Ask five people what AI means. You may get five different answers.

Some may say ChatGPT. Others may say robots or smart software.

The meaning of AI has changed over time.

This lesson looks at how AI changed and which ideas became important.

Give a machine enough rules, and it can act smart.

In 1956, researchers met at Dartmouth College.

They believed intelligence could be described using rules.

If humans wrote the rules, a computer could follow them.

For many years, AI was mostly about logic and rules.

This worked for simple tasks. Computers could solve puzzles, play games, and follow instructions.

Every rule needed another exception.

Take the rule “birds can fly.” Penguins cannot fly. Ostriches cannot fly.

A bird with a broken wing cannot fly either.

Real life has many exceptions.

More exceptions meant more rules. The systems became difficult to manage.

Humans had to manually add too much knowledge. This became a major problem for early AI.

Stop writing rules. Give the machine examples.

Researchers tried a different approach.

Instead of writing every rule, they gave computers lots of examples.

The computer could then find patterns in those examples.

To teach it about cats, you could show it many pictures of cats and other animals.

Learning means adjusting numbers to make fewer mistakes.

The model makes a prediction.

It checks how wrong the prediction was.

Then it changes some internal numbers.

It repeats this many times. The predictions slowly improve.

This simple process is at the heart of machine learning.

More layers can learn more complex patterns.

Neural networks contain layers.

Each layer works with information from the previous layer.

Early layers may detect edges.

Later layers may detect shapes, faces, or objects.

This is the “deep” in deep learning. The network contains many layers.

Bigger models started working better.

New model designs made it easier to train on huge amounts of data.

Researchers also used more powerful computers.

As models became larger, their performance often improved.

This led companies to build much bigger AI systems.

We still do not know how far this can continue or what bigger models cannot solve.

AI moved from choosing answers to creating them.

Older machine learning systems mostly chose between options.

Is this email spam? Is this image a cat? Which product should we recommend?

Newer systems can create things.

They can generate text, images, audio, video, and code.

The answer is not always clear.

Some systems clearly feel like AI.

Others are harder to classify.

Try sorting the examples below into “AI” and “not AI”.

There is no perfect answer.

A useful definition is this: AI systems learn patterns from data and use those patterns on new inputs.

AI does not want anything.

AI does not have feelings.

It does not have personal goals.

When a chatbot says “I understand how you feel”, it is generating a suitable response.

It is not actually feeling anything.

People often feel that a responsive machine has a mind. This is called the ELIZA effect.

When AI becomes common, we stop calling it AI.

Chess programs were once seen as advanced AI.

So were speech recognition, spam filters, route planning, and autocomplete.

Today, we often call these normal software.

The label AI usually moves toward newer technology.

This is one reason AI can be difficult to define. The meaning keeps changing.

A simple definition of AI.

For this course, use this definition.

AI is a system that learns patterns from data and uses them on new inputs.

It does not need to think like a human.

It can use learned patterns to make predictions or create new content.

Lesson 02 looks inside a modern AI system and shows how it works step by step.

1956 — Logic

Intelligence is rules, written by hand. Elegant, and it broke on the first exception nobody anticipated.

1980s — Expertise

Thousands of rules from human experts. Collapsed under maintenance before it collapsed commercially.

1990s — Learning

Stop writing rules; learn them from examples. Quietly won, and is still the definition that matters.

2012 — Depth

Layers that build their own features. Vision and speech went from unsolved to ordinary within five years.

2017 — Scale

An architecture that rewarded more data and more compute, further than anyone predicted.

Today — Generation

The same machinery producing sequences rather than picking labels. New output, old mechanism.