Did AI exist before ChatGPT?

Why Everyone Thinks ChatGPT Started AI (And Why That's a Problem)

If you first heard about artificial intelligence in December 2022, you're not alone — but you're also about 70 years late. The field of AI was formally founded at the 1956 Dartmouth workshop, where John McCarthy coined the very term "artificial intelligence" (according to historical accounts). ChatGPT didn't start AI; it started something much narrower, and mistaking the two costs businesses real money.

The problem with treating ChatGPT as AI's origin point is that it makes you overlook tools you already own. As MIT xPRO's analysis explains, everyday products "quietly integrated AI to improve efficiency, personalization, and decision-making long before the current generative AI boom":

• Google's RankBrain, which has interpreted complex search queries since 2015
• Spam filters and bank fraud detection systems that analyze patterns in real time
• Netflix and Amazon recommendation engines driving purchase behavior
• Siri (2011) and Alexa (2014), which brought natural language processing to consumers

The key to understanding what actually changed in November 2022 is the distinction between traditional AI and generative AI. Traditional AI is reactive — it analyzes data to make predictions within pre-defined rules, but it "lacks the ability to create new content." Generative AI is proactive: it produces novel text, images, and code from learned patterns. Your spam filter never wrote you an email; ChatGPT can.

Even within OpenAI's own lineage, ChatGPT was a latecomer. Release timelines show GPT-1 launched in June 2018, GPT-2 in February 2019, and GPT-3 — with 175 billion parameters versus GPT-2's 1.5 billion — in May 2020, a full two and a half years before ChatGPT's public preview. And in non-language domains, world-beating AI arrived even earlier: AlphaGo defeated 18-time Go champion Lee Sedol 4–1 in March 2016, a match watched by over 200 million people (Wikipedia documents).

This matters for business decisions. If ChatGPT feels like the beginning, you assume AI adoption means buying new chatbots. If you understand the real timeline, you recognize that analysis and creation are different AI capabilities with different uses — and that the specialized systems approach (expert systems like MYCIN in the 1970s, game engines like AlphaGo) is how AI has always delivered value. That same pattern continues today: modern agents gain capability through specialized skills, and SkillzBench is building a free directory and MCP connector to help teams discover, verify, and benchmark them. The lesson from seven decades of AI history: the tools that create headlines are rarely the ones doing the quiet work.

Seven Decades of AI: From Dartmouth to Deep Blue

The short answer to "did AI exist before ChatGPT?" is yes — by roughly 70 years. The field was formally born in 1956 at the Dartmouth workshop, where John McCarthy coined the term "artificial intelligence" and Allen Newell, Herbert Simon, and J.C. Shaw presented the Logic Theorist, the first AI program concept, according to the documented history of the field.

The optimism of that era — captured in the founding belief that "every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it" — drove decades of functional systems long before anyone typed a prompt into a chat window, as historical accounts detail.

Functional AI arrived in waves, each with a distinct job:

• ELIZA (1966) — MIT's Joseph Weizenbaum built the first chatbot, which simulated therapy by repurposing user answers into questions; many users believed they were talking to a human.
• Expert systems (1970s–80s) — MYCIN handled medical diagnosis and XCON configured machines at Digital Equipment Corporation, embedding AI in real business workflows.
• Deep Blue (1997) — IBM's chess computer beat world champion Garry Kasparov, reviewing 200 million potential moves per second.
• Watson (2011) — IBM's system defeated Jeopardy! champions, bringing AI reasoning to mainstream audiences.
• AlphaGo (2016) — Google DeepMind's program beat 18-time world Go champion Lee Sedol 4–1, a match watched by over 200 million people, using deep neural networks combined with Monte Carlo tree search and reinforcement learning.

The AlphaGo milestone deserves special weight. Go has roughly 10^170 possible board configurations — a googol times more complex than chess — and DeepMind itself notes the victory came "a decade before experts thought possible," with techniques still used in today's AI systems.

This history wasn't a smooth climb. After 1956's optimism, the 1974 Lighthill report and U.S. Congressional pressure ended government funding, triggering an "AI winter" that stretched from the late 1970s to the early 1990s. Expert systems and Japanese initiatives then revived the industry into a billion-dollar enterprise by the late 1980s — a boom-and-bust cycle worth remembering as today's agent hype accelerates.

That pattern repeats now: just as businesses once adopted expert systems like XCON to automate specialized tasks, today's AI agents gain capability through specialized skills. SkillzBench is building a free directory and MCP connector for discovering, verifying, and benchmarking those agent skills — the same cycle of specialization, one generation later.

The AI You Were Already Using Before ChatGPT

If you ran a business any time in the last decade, AI was already working for you every single day — you just never called it that. Spam filters, recommendation engines, and fraud alerts were quietly making decisions on your behalf years before anyone typed a prompt into a chatbot.

Consider what your workday already looked like. Google launched RankBrain in 2015, an AI system that interpreted complex search queries — meaning even your Google searches ran on machine learning seven years before ChatGPT appeared. Netflix and Amazon recommendations, Siri (2011), and Alexa (2014) all brought natural-language and pattern-matching AI into ordinary workflows, as MIT xPRO's analysis documents. Your bank flagged suspicious transactions with AI. Radiology teams used machine learning to analyze X-rays, MRIs, and CT scans.

• Spam filters — machine learning classifying millions of emails before they reach you
• Google RankBrain (2015) — AI interpreting ambiguous search queries
• Netflix and Amazon recommendations — predictive models driving revenue
• Siri and Alexa — natural language assistants in every pocket and office
• Bank fraud detection and medical imaging — AI making high-stakes calls quietly, for years

The reason none of this felt like "AI" is the distinction between traditional and generative systems. Traditional AI analyzes data and makes predictions within predefined rules — it "lacks the ability to create new content," as the same MIT piece puts it. ChatGPT felt revolutionary because it generated text, not because it was the first intelligence in your toolkit. As one MIT instructor put it, generative AI "has been the gateway for many people to finally engage with AI" — the entry point, not the origin.

Even OpenAI's own history proves the point. GPT-1 launched in June 2018, GPT-2 followed in February 2019, and GPT-3 arrived in May 2020 — a full two and a half years before ChatGPT's November 30, 2022 launch, per the documented GPT timeline. That's a 4.5-year gap between OpenAI's first AI model and the chatbot that made it famous. GPT-3 already had 175 billion parameters; what ChatGPT added was a friendlier interface, follow-up questions, and better handling of inappropriate prompts, as Coursera's history of AI explains.

The lesson for business owners is simple: AI adoption isn't new, and you were probably further along than you thought. The real shift is from AI that analyzes to AI that acts — and today's agents extend that trajectory. If you're exploring what agents can do for your workflows, agent skills are the practical next step: SkillzBench is building a free directory and MCP connector so you can discover, verify, and benchmark skills before trusting them with real work.

What ChatGPT Actually Changed — And What It Means for Your AI Strategy

ChatGPT didn't invent AI — it changed who could use it. When it launched on November 30, 2022, it was OpenAI's fourth major language model in roughly four and a half years, following GPT-1 (2018), GPT-2 (2019), and GPT-3 (2020), which packed 175 billion parameters compared to GPT-2's 1.5 billion.

So what actually changed? According to MIT xPRO, generative AI served as "the gateway for many people to finally engage with AI." The distinction matters for your strategy. Traditional AI — the spam filters, fraud detection, Netflix recommendations, and Google's RankBrain (launched 2015) that most businesses already relied on — is reactive, analyzing data within pre-defined rules. Generative AI is proactive: it creates new content rather than merely analyzing existing data.

That shift from analysis to creation is why ChatGPT felt like a revolution when it was really an inflection point. And the pattern behind it offers a practical lesson for anyone building with AI agents today.

Look at the systems that came before: expert systems like MYCIN for medical diagnosis and XCON for configuring machines at Digital Equipment Corporation each solved one problem well. Deep Blue reviewed 200 million potential chess moves per second to beat Kasparov in 1997 — but it couldn't write an email. AlphaGo defeated 18-time world champion Lee Sedol 4–1 in March 2016 using deep neural networks and reinforcement learning, and DeepMind notes its planning techniques are still used in today's AI systems — yet it played exactly one game.

Each breakthrough was narrow by design, and that narrowness is precisely what made it powerful.

Today's AI agents follow the same logic. A general-purpose agent becomes genuinely capable not by getting bigger, but by stacking specialized skills — instruction packages that teach it to handle one workflow reliably. The catch, as the field is learning, is quality: research has found roughly 26% of published agent skills carry at least one vulnerability, so which skills you attach matters as much as which model you run.

That's the problem SkillzBench is being built to solve. It will offer a free, independent directory of agent skills — continuously crawled from GitHub, version-pinned, security-scanned, and benchmarked in a sandbox with and without each skill so you can see the actual Lift in percentage points. A free MCP connector will let you toggle exactly which skills each agent sees. Nothing is live yet, but you can join the waitlist now to browse and test free agent skills the moment it launches.

Seventy years of AI history points the same direction: capability comes from specialization. Choose your agent's skills the way the field always has — deliberately, and with evidence.