Chapter Four - The First Winter

CHAPTER FOUR

The First Winter

EVERY BOOM HAS its bust. Every golden age has its reckoning.

For artificial intelligence, that reckoning hit hard in the 1970s.

What began as a revolution of logic and language turned into a graveyard of broken promises. Systems didn’t scale. Models didn’t generalize. Expectations skyrocketed — and results flatlined.

The age of thinking machines collapsed under its own weight.

They called it the AI Winter.

What went wrong?

Simple: the machines weren’t smart.

They couldn’t deal with ambiguity. They couldn’t learn new concepts without being manually reprogrammed. They couldn’t translate language beyond textbook phrases. They couldn’t adapt to new environments. They couldn’t even tell a cat from a toaster unless someone coded the difference by hand.

What AI researchers had actually built was a pile of brittle logic trees and symbol manipulators — impressive in the lab, useless in the wild.

The gap between promise and performance was enormous.

And everyone noticed.

In 1969, Marvin Minsky — ironically one of AI’s biggest hype men — co-authored a book with Seymour Papert that effectively destroyed the perceptron.

In Perceptrons, they mathematically proved that single-layer neural networks couldn’t solve even basic visual problems, like distinguishing shapes in different positions. It wasn’t just a criticism — it was a eulogy.

Funding bodies read that book like a death certificate.

Neural nets were dead. AI was over.

By the mid-’70s, the U.S. and U.K. governments began slashing AI research budgets. DARPA pulled back. Commercial interest vanished. The field was ridiculed in academia. Graduate students stopped applying. Journalists stopped writing. AI became a cautionary tale — the story of tech’s greatest overpromise.

A second winter hit in the late 1980s, when expert systems (we’ll get to those soon) also failed to live up to the hype.

This was no longer a cycle of excitement.

This was a pattern.

The core problem wasn’t the ambition — it was the metaphor.

Researchers had assumed that intelligence was just logic. That if you could write enough rules, you could simulate thought. But the human brain doesn’t run on symbols. It runs on patterns. Neurons. Experience. Noise.

Trying to model a brain with rules was like trying to simulate a hurricane with a spreadsheet.

Nature doesn’t care about neat categories.

But here’s the twist: AI didn’t die in the winter. It waited.

A few researchers — outcasts, visionaries, lunatics — kept the flame alive. They kept tinkering with neural networks. They kept questioning assumptions. They watched as Moore’s Law drove up computing power and storage. They bided their time.

Because they knew something others didn’t:

The future wasn’t about rules.

It was about data.