The Evolution of Artificial Intelligence

From Logic
to Intelligence

Seventy years from a single question — can machines think? — to machines that reason, create, and act. Scroll through the acceleration.

Begin
1950 — Present

The line that learned to think

Each milestone closer than the last. Watch the gaps collapse as we approach now.

The Question 1950

The Turing Test

Alan Turing asks whether a machine can imitate human thought well enough to fool us. The benchmark that haunts the field is born.

The Field Is Named 1956

The Dartmouth Workshop

A summer of mathematicians coins the term "Artificial Intelligence." A discipline gets a name and a mission.

First Conversation 1966

ELIZA

The first chatbot mirrors your words back as questions. Crude pattern-matching — yet people confided in it. The illusion of understanding arrives early.

The Knowledge Era 1980s

Expert Systems Boom

Hand-coded rules encode human expertise into software. Industry pours in. For a moment, it feels like reasoning is solved.

The Collapse 1987–93

The AI Winter

The hype outruns the results. Funding evaporates, labs shutter, the promise freezes over. The field learns humility the hard way.

Machine Beats Master 1997

Deep Blue defeats Kasparov

IBM's machine beats the world chess champion. Brute-force search topples a human mind at its own game. The world takes notice again.

The Revolution 2012

AlexNet & ImageNet

A deep neural network shatters the image-recognition benchmark. Deep learning ignites — and everything after accelerates from here.

Intuition, Learned 2016

AlphaGo beats Lee Sedol

A machine masters Go — a game thought to require intuition. Move 37 stuns the world's best. Strategy is no longer ours alone.

The Architecture 2017

"Attention Is All You Need"

The Transformer is published. A single architecture unlocks scale — and quietly becomes the engine behind everything that follows.

Scale Speaks 2020

GPT-3

175 billion parameters. Language modelled at a scale that surprises even its makers. Fluency emerges from sheer size.

Mainstream 2022

ChatGPT

AI walks out of the lab and into a hundred million pockets in two months. The fastest-adopted product in history. Everything changes overnight.

Multimodal & Agentic 2023–24

GPT-4 & the Agent Era

Models that see, hear, and reason across modalities — then begin to use tools, write code, and act in steps toward goals.

You Are Here NOW

Frontier reasoning & agents in the world

Models that think before they answer, plan across long horizons, and act through software on our behalf. The gaps have collapsed to weeks. The question is no longer can machines think — but what we ask them to do.

Then → Now

The distance travelled

ELIZA matched patterns in a single line. Today's systems reason across thousands. The same field — a different universe.

Conversation
1966 → Now
Pattern-match → Reasoning
Echoes your wordsChains thought to a goal

ELIZA reflected sentences back. Today's agents form plans, test hypotheses, and self-correct mid-task.

Scale
~10¹² ×
Parameters
Hand-written rulesTrillions of weights

From a few hundred coded rules in expert systems to models learning from a meaningful fraction of the internet.

Adoption
2 months
To 100M users
Decades in labsGlobal in weeks

It took the telephone 75 years to reach 100M people. ChatGPT did it in two months.

Capability
Sees · Thinks · Acts
From single-task to general
One narrow gameCode, vision, tools, plans

Deep Blue played chess and nothing else. Today's frontier models cross domains and operate software on their own.

The next milestone isn't on this line yet.
You'll watch it arrive.

For seventy years, intelligence was something only we possessed. That assumption is now the most interesting open question of our time.