Everyone remembers where they were when they first tried ChatGPT. Maybe you asked it to write a poem. Explain quantum physics like you’re five. Draft that awkward email to your boss. And somewhere between amazement and existential dread, you thought:

Where the hell did this come from?

Here’s the answer nobody tells you: ChatGPT was 70 years in the making. It’s the survivor of two “AI winters,” billions in wasted funding, careers destroyed by overpromising, and decades when saying you worked in “artificial intelligence” was like admitting you believed in UFOs.

This is that story—the real one. No jargon. No hype. Just the wild, improbable, sometimes hilarious journey of humanity’s longest-running technological obsession.

1950: A Dangerous Question

It started with Alan Turing asking the question that launched a thousand research papers: “Can machines think?”

Turing—the British genius who cracked Nazi codes and essentially invented computer science—didn’t want to debate philosophy. He proposed a simple test: if a machine can fool a human into thinking it’s human through conversation alone, who cares if it’s “really” thinking? Call it intelligent and move on.

Six years later, a group of young hotshots gathered at Dartmouth College for a summer workshop. Their pitch to funders was breathtaking in its confidence: “Every aspect of learning or any other feature of intelligence can be so precisely described that a machine can be made to simulate it.”

They needed a sexy name. One organizer, John McCarthy, suggested “artificial intelligence.” It stuck. They figured they’d crack it in a few months. They were off by about seven decades.

The Golden Age of Overpromising

Early AI actually delivered some wins. In 1956, a program called Logic Theorist proved mathematical theorems—one proof was so elegant that its creators submitted it to a math journal. The journal rejected it. Co-authorship with a machine was apparently a step too far.

Then came ELIZA in 1966—a chatbot that mimicked a therapist by simply reflecting your words back at you. (“I feel sad.” “Why do you say you feel sad?”) It was a parlor trick. But people loved it. The creator’s own secretary asked him to leave the room so she could talk to ELIZA privately. He was horrified by what he’d unleashed.

Flushed with success, researchers made predictions that would haunt them. “Within 20 years, machines will do any work a man can do.” “The problem of AI will be substantially solved within a generation.” Government money flooded in.

Winter Is Coming (The First One)

Then the checks bounced.

Turns out, getting computers to do “simple” human things—recognize a face, understand sarcasm, not walk into walls—was insanely hard. The computers of the 1970s had less processing power than your microwave. A brutal 1973 British government report concluded AI had failed to meet its “grandiose objectives.”

Funding vanished. Researchers left the field. “AI winter” had arrived. For a decade, admitting you worked on artificial intelligence was career suicide.

The ’80s: Electric Boogaloo (AI Winter 2)

AI got a second chance with “expert systems”—programs that captured human expertise in narrow domains. Diagnose diseases! Configure computers! Business magazines went nuts. Japan bet a billion dollars on it. Companies scrambled to build AI departments.

You can guess what happened next. Expert systems were expensive, brittle, and useless outside their tiny specialty. By 1987, the market collapsed. AI Winter 2.0. Researchers once again learned to mumble when asked what they did for a living.

The Nerds Keep Working

Here’s the thing about winters: spring eventually comes.

While the hype died, stubborn researchers kept tinkering. They stopped trying to program intelligence directly and started letting machines learn from data. In 1997, IBM’s Deep Blue beat chess champion Garry Kasparov—not by “thinking” but by evaluating 200 million moves per second through sheer brute force. Kasparov accused them of cheating.

Meanwhile, something massive was happening quietly. The internet was generating oceans of data. Graphics chips designed for video games turned out to be perfect for AI calculations. Computing power was doubling every couple of years. The ingredients were coming together.

2012: The Year Everything Changed

Neural networks—computer systems loosely inspired by the brain—had been around since the 1950s but never worked at scale. Then, around 2012, researchers figured out how to stack them deep and feed them massive data.

The results were stunning. Image recognition went from “impressive science project” to “better than humans.” So did speech recognition. Translation. One by one, problems that had stumped researchers for decades started falling like dominoes.

In 2016, Google’s AlphaGo defeated the world champion at Go—a game with more possible positions than atoms in the universe. One move, now famous as “Move 37,” was so unexpected that the champion left the room to compose himself. Commentators called it “beautiful.” A machine had made a creative, artistic move.

2017: “Attention Is All You Need”

Google researchers published a paper with an almost cocky title: “Attention Is All You Need.” It described a new architecture called the Transformer—a way for AI to understand language by grasping how every word in a sentence relates to every other word.

Translation: AI could finally get context. It understood that “bank” means different things in “river bank” and “bank robbery.” It could follow a long argument. It could write paragraphs that actually made sense.

OpenAI took this architecture and went full send. GPT-1 in 2018. GPT-2 in 2019 was so good they initially refused to release it, fearing weaponized misinformation. GPT-3 in 2020 had 175 billion parameters and could write essays, code, poetry.

Then came November 30, 2022. ChatGPT launched. A million users in five days. And suddenly, 70 years of AI history stopped being a niche academic subject and became everyone’s problem to figure out.

The Punchline

So why does this history matter?

Because the next time AI hype sounds too good to be true—or too scary to be real—remember: this field has a 70-year track record of being wrong. Wrong about timelines. Wrong about what’s hard. Wrong about what’s easy. The researchers who kept working through the winters, when AI was a punchline, are why you can now generate images with a text prompt.

We’re not at the end of this story. We’re probably still in the opening act. Strap in.

—

Rashad Haque

November 2025