Patterns become predictions.
Give a system observations, a way to compare them, and feedback about its mistakes. Learning is the gradual adjustment of that comparison.
An interactive history of artificial intelligence
How machines learned to notice.
Artificial intelligence did not arrive in a single flash of code. It grew through ideas about logic, learning, language, data, and the stubborn mystery of thought itself.
1943–today
Eight decades of ideas
10 moments
That shifted the field
1 live model
Waiting for your examples
Start with a useful definition
AI is a collection of ways to make a machine act on patterns.
Different eras asked different questions. Can a machine reason from rules? Learn from examples? Recognize an image? Generate language? The methods changed, but the central ambition remained: build systems that can turn information into useful decisions.
Rules
Early researchers wrote knowledge down explicitly: if this, then that.
Data
Later systems found their own regularities across many examples.
Scale
More computing power made larger models and richer tasks practical.
Interactive simulation
Train a historical machine.
Click examples, then teach the model
model://1958/perceptron
ready
FEATURE SPACE · X / Y
Try it: add a few examples of each class. The faint field is the model's current guess.
1958
The Perceptron
A learnable boundary: it shifts after seeing labeled examples, but only separates simple patterns.
SELECT AN ERA BELOW TO CHANGE THE KIND OF MODEL IN THE LAB.
A non-linear timeline
The field kept reinventing its questions.
Selected signal
1958
The machine became a learner, at least in principle.
Learn from examples
The Perceptron
Psychologist Frank Rosenblatt proposed a trainable network inspired by simple neurons. It could change a set of weights after mistakes, making learning a concrete engineering process.
What changed
A model could adjust from feedback.
In the lab
A single boundary moves through the field.
The recurring ingredients
Progress came from a conversation between ideas and machinery.
There is no one “AI.” Every breakthrough brought a new tradeoff: more knowledge, more examples, more computation, more flexible architectures, or a better way to test what a system had learned.
01
Representation
How should a system hold a fact, an image, a sentence, or a relationship in a form it can use?
02
Learning signal
What tells the machine that a choice was good, bad, surprising, or incomplete?
03
Generalization
The real test is not remembering old examples. It is doing something sensible with a new one.
04
Human judgment
Every system reflects choices about data, goals, errors, access, and what outcomes matter.
The story is still being written
Every model is a record of what we decided to teach it.
Go back to the learning lab, switch eras, and notice what each kind of model can and cannot see. AI history is not a straight climb toward intelligence—it is a growing map of possible tools.