Before ChatGPT: The (Brief) History Of AI

In partnership with

Welcome to this week’s Deep-Fried Dive with Fry Guy!

In these long-form articles, Fry Guy conducts in-depth analyses of cutting-edge artificial intelligence (AI) developments and developers. Today, Fry Guy dives into what came before ChatGPT. We hope you enjoy!

Do More. Spend less on SaaS.

Launch, grow, and scale your company faster with Notion.

Thousands of startups rely on Notion to move quickly, stay aligned, and replace multiple tools. Whether you're building a wiki, managing projects, or writing documentation, Notion is your all-in-one workspace.

Get up to 6 months of the new Plus plan + unlimited AI, for free!

To redeem your Notion for Startups offer, simply visit the Notion for Startups page and apply.

Get the offer

Artificial intelligence might seem like a 21st-century breakthrough, but the truth is that ChatGPT’s 2022 debut was decades in the making. When OpenAI released ChatGPT on November 30, 2022—as a friendly chatbot interface to a powerful GPT-3.5 language model—it marked the culmination of over 70 years of research and innovation.

In this fun, simplified, and informative journey, we’ll travel back in time to answer the question: What came before ChatGPT?

THE DAWN OF THINKING MACHINES (1950s–1960s)

The story begins in 1950, when British mathematician Alan Turing asked a deceptively simple question: “Can machines think?” This led to the now-famous Turing Test—a way to gauge a machine’s intelligence by how well it could hold a conversation without giving itself away. It was more philosophy than engineering at the time, but it lit a spark in the tech community.

A few years later, in 1956, a group of ambitious researchers met at Dartmouth College. Following Turing’s question, these researchers believed that machines could, in principle, learn and reason just like humans. Together, they coined the term “artificial intelligence.” With that bold declaration, AI was born—not just as an idea, but as a field of research.

The early experiments were rule-heavy and surprisingly charming. Marvin Minsky built a “learning machine” with vacuum tubes and rats. Arthur Samuel created a checkers program that learned from experience to compete against humans. Meanwhile, Shakey the Robot—the first robot to perceive its surroundings—was bumbling around halls at Stanford, clunkily making decisions and navigating rooms like a toddler in a cardboard costume. And in 1966, MIT’s ELIZA program mimicked a therapist by bouncing user input back as questions. It didn’t understand anything—but people felt understood.

HOPE, HYPE, AND HARD LESSONS (1970s–1980s)

By the ’70s, AI had swagger. Researchers were building “expert systems”—programs that mimicked expert human decision-making by applying hand-coded “if-then” rules to specific problems. DENDRAL analyzed chemical data to identify molecular structures, while MYCIN asked doctors questions and recommended antibiotic treatments. Although these systems often made mistakes, the potential wowed industries—and soon, companies were investing heavily in AI.

But those early systems had a fatal flaw: they couldn’t learn. Every update meant manually adding rules. As projects ballooned in complexity, enthusiasm gave way to frustration. By the late 1980s, funding dried up. This cold snap became known as “the AI winter”—a period when hopes dimmed and AI lost its luster.

LEARNING TO LEARN (1990s–2000s)

In the ’90s, AI traded rigid rules for data. Machine learning emerged as the new frontier: instead of telling a computer how to think, you showed it examples and let it figure out the patterns for itself.

Neural networks—systems that could learn from data to make predictions—started to shine. Through this new method of learning, AI started to read handwriting and could remember information across time.

Then came some huge turning points that changed how everyone thought about machines. One notable turning point came in 1997, when IBM’s Deep Blue beat world chess champion Garry Kasparov. This shocked the tech world and general public: could robots actually “out-think” humans? AI was cool once again.

In the 2000s, tech advances supercharged AI. Graphics processing units (GPUs), originally built for video games, were repurposed to train neural networks much faster. Then, in 2011, IBM’s Watson—which analyzed vast amounts of data to answer questions—won Jeopardy!, competing against two famous human competitors. AI was learning fast, and it was catching people’s attention!

WHEN BIGGER BECAME BETTER (2010s)

Once AI started learning, it never stopped. In 2012, a neural network called AlexNet blew away the competition in an image recognition challenge called ImageNet. What made it special? It could scan millions of pictures and learn to recognize objects—like cats, cars, or faces—far better than anything before. It used a method called “deep learning,” where layers of digital “neurons” process data and spot patterns, kind of like a simplified brain that perceives shapes, colors, and textures. This win was a wake-up call: if you gave neural networks enough data and computing power, they could outperform traditional methods by a mile.

That breakthrough sparked a creative explosion. Soon, AI was being applied to all kinds of domains. It began composing music, translating languages, and even making images from scratch. One clever approach, called GANs (Generative Adversarial Networks), used two AIs in a kind of digital showdown. One AI tried to create fake images, while the other tried to catch them. Over time, both got so good that the fake images became nearly indistinguishable from real ones.

Even with these advances, however, understanding language remained tough. Words have context, tone, and layers of meaning. Older models struggled with longer passages or nuanced phrasing. That changed in 2017 when Google introduced the Transformer—a new model that could read a sentence and figure out which words mattered most, no matter how far apart they were. It was a huge leap forward for AI’s ability to understand human language patterns, and it set the stage for everything that followed.

MEET GPT: THE WORD PREDICTION PRODIGY (2018–2020)

Tech geeks saw potential in these models. Innovators like Elon Musk, Ilya Sutskever, and Sam Altman started a company called OpenAI (ever heard of it?) to capitalize on these advancements, “for the good of humanity.” In 2018, they released GPT-1, a transformer trained on human writing from the internet to predict the next word in a sentence. It wasn’t flashy, but it laid the groundwork for one of the greatest tech breakthroughs we have ever seen.

In 2019, GPT-2 made waves by showing it could generate surprisingly convincing paragraphs of text—stories, news-style blurbs, even simple explanations—just by being given a short prompt. It worked by predicting one word at a time, based on what came before.

Then came GPT-3 in 2020, and it took things to a whole new level. It could write essays, summarize long documents, draft poems, and even generate simple computer code. All it needed was a sentence or two to get started. It was incredibly capable—but still tricky to access unless you had permission to use OpenAI’s program. Progress was being made, and a major breakthrough was looming.

FROM LABS TO OUR FINGERTIPS (2021–2022)

OpenAI was determined to make AI more conversational and accessible for everyone. The company trained models to better follow human instructions with something called InstructGPT, refining outputs using human feedback.

Then, in late 2022, OpenAI launched ChatGPT, a friendly chat interface powered by GPT-3.5. It looked like a messaging app—but with a super-intelligent pen pal on the other side. People asked it for recipes, resumes, bedtime stories, bug fixes—you name it. It quickly rose to popularity, hitting a million users in just five days! Suddenly, AI wasn’t a lab curiosity—it was part of daily life. This opened the floodgates for the conversational models that we know and love today!

ON THE SHOULDERS OF GIANTS

So, what came before ChatGPT? A long chain of ideas and breakthroughs, each building on the last. From Alan Turing’s early philosophical musings in 1950, to the enthusiastic goals set at Dartmouth in 1956, the rollercoaster of expert systems and AI winters, the resurgence through machine learning, and finally the eruption of deep learning and transformers—every era contributed something essential. We gained an understanding of language, learning, perception, and reasoning bit by bit. ChatGPT was not a breakthrough that came out of nowhere—it was only possible because of all of these stepping stones.

In a familial sense, you could say ChatGPT had many “grandparents”: a therapist chatbot from the 60s (ELIZA) taught it the art of conversation; rule-based expert systems showed the value of knowledge (and the pitfalls of hard-coded rules); Deep Blue proved machines can tackle games and think ahead; the Transformer taught it to attend to every word you say; and the GPT series gave it a world-class education by reading the internet. Each breakthrough made ChatGPT a little more feasible.

So today, when you chat with an AI model like ChatGPT, you’re engaging with the product of all these historical milestones. It might feel like talking to the future, but it’s also like talking to a synthesis of human knowledge and AI history. And while ChatGPT’s launch was a landmark, the story of AI isn’t over—far from it. New, advanced models are emerging every day, built on the same foundations.

As we marvel at what AI can do now, it’s worth remembering the long road of trials, errors, and genius ideas that got us here. ChatGPT stands on the shoulders of giants, proving that revolutionary things can happen when the right pieces come together at the right time—even if it takes over half a century!

He’s already IPO’d once – this time’s different

Spencer Rascoff grew Zillow from seed to IPO. But everyday investors couldn’t join until then, missing early gains. So he did things differently with Pacaso. They’ve made $110M+ in gross profits disrupting a $1.3T market. And after reserving the Nasdaq ticker PCSO, you can join for $2.80/share until 5/29.

Invest today for $2.80/share.

_{This is a paid advertisement for Pacaso’s Regulation A offering. Please read the offering circular at invest.pacaso.com. Reserving a ticker symbol is not a guarantee that the company will go public. Listing on the NASDAQ is subject to approvals. Under Regulation A+, a company has the ability to change its share price by up to 20%, without requalifying the offering with the SEC.}