Back to Knowledge Base >>

The History of AI and Key Milestones

Seventy-five years of bold ideas, two winters, one generative spring, and a handful of breakthroughs that quietly rewired how machines think: how AI went from theory to ChatGPT.

By |Published On: July 6, 2026|Last Updated: July 6, 2026|Categories: |
The History of AI and Key Milestones

ChatGPT can feel like it appeared out of nowhere. It didn’t. Every system generating images, writing code, or answering questions today rests on 75 years of ideas, most of which failed before they worked. Twice, the field’s own hype outran its results, and funding collapsed in what researchers now call an “AI winter.” Twice, it came back stronger. That pattern, boom, bust, breakthrough, is the real story here, and it’s the best lens for judging whether today’s generative AI wave is different, or just the third loop of the same cycle.

Defining Each Discipline  1940s–1950s

Mathematicians and engineers begin asking whether machines can reason, and give the field a name.

1943

McCulloch & Pitts publish the first computational model of a neural network, showing how simple artificial neurons could perform logical operations.

1950

Alan Turing publishes “Computing Machinery and Intelligence,” proposing the Turing Test as a practical way to ask whether a machine’s conversation is indistinguishable from a human’s.

1955–56

John McCarthy coins the term “artificial intelligence.” The Dartmouth Summer Research Project, organized by McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon, is widely regarded as founding AI as a formal discipline.

Golden years & first winter  1960s–1970s

Early wins fuel confidence that general machine intelligence is close, until funding and computing power run out.

1966

ELIZA, MIT’s early natural-language chatbot, simulates a psychotherapist, a forerunner of today’s conversational AI.

1969

SRI International builds Shakey, the first mobile robot able to perceive its environment, plan, and reason about its own actions.

1970s

Predictions of near-term general intelligence go unmet. Funding from DARPA and others is sharply cut, triggering the first AI winter.

Expert systems & second winter  1980s–1990s

A commercial wave of rule-based systems rises, then falls when it can’t scale beyond narrow tasks.

1980s

Expert systems, rule-based programs replicating specialist decision-making, drive a new commercial wave, especially in medicine and engineering.

Late 1980s

Expert systems prove expensive and brittle outside narrow use cases, triggering a second AI winter.

1997

IBM’s Deep Blue defeats world chess champion Garry Kasparov, proof machines can outperform humans at complex, rules-based reasoning.

Machine learning era  2000s–2010s

Statistical learning replaces hand-coded rules, supercharged by the internet’s data and growing compute.

2006

Geoffrey Hinton and colleagues popularize “deep learning,” reviving multi-layer neural networks at scale.

2011

IBM’s Watson defeats human champions on Jeopardy!, showcasing advances in natural language processing.

2012

AlexNet dramatically outperforms rivals in the ImageNet competition, widely credited with kickstarting the modern deep learning boom.

2016

DeepMind’s AlphaGo defeats world champion Lee Sedol at Go, a milestone for AI reasoning and intuition.

Generative AI era  2017–Present

A new architecture, the Transformer, powers a leap in language and image generation, bringing AI to the mainstream.

2017

Google researchers introduce the Transformer architecture in “Attention Is All You Need”, the foundation for nearly all modern large language models.

2020

OpenAI releases GPT-3, a 175-billion-parameter model demonstrating strong general-purpose language ability.

2021-22

DALL·E, Midjourney, and Stable Diffusion bring generative AI to visual content.

Nov 2022

ChatGPT launches as a free research preview, reaching roughly 100 million users in two months and bringing generative AI to mainstream attention.

2023-24

GPT-4, Google’s Gemini, Anthropic’s Claude, and Meta’s open-weight Llama models intensify competition; GPT-4o unifies text, audio, and vision.

2024-25

Two shifts move AI from “answer machine” to “action machine.” Reasoning models (OpenAI’s o1/o3, and comparable models from other labs) are trained to work through multi-step problems before responding, trading speed for accuracy on tasks like coding, math, and analysis. At the same time, AI agents start handling multi-step work with less human supervision: researching a topic across sources, drafting and revising a document, or executing a workflow across several tools, then reporting back rather than waiting for a prompt at every step. For organizations, this is the shift from “AI that drafts an email” to “AI that manages a process,” which is also where governance questions get harder: what decisions is the agent allowed to make on its own, and how is that tracked?

Why this history matters

The pattern is consistent: a breakthrough in theory or architecture sparks a wave of optimism and investment, followed by a sober reassessment once real-world limits appear. Today’s generative AI boom is the latest, and largest, of these waves, driven by the convergence of compute, internet-scale data, and the Transformer. For any organization evaluating AI adoption, it’s a reminder that capability is one chapter in a longer story, and that well-governed adoption tends to outlast the hype cycle.

Sources

  • Turing, A. M., “Computing Machinery and Intelligence,” Mind, 1950
  • McCulloch, W. & Pitts, W., “A Logical Calculus of the Ideas Immanent in Nervous Activity,” 1943
  • Vaswani, A. et al., “Attention Is All You Need,” Google Research, 2017
  • Coursera, “The History of AI: A Timeline of Artificial Intelligence”
  • TechTarget, “The History of Artificial Intelligence: Complete AI Timeline”
  • CIO Dive, “The Rise of Generative AI: A Timeline of Triumphs, Hiccups and Hype”
  • OpenAI & Reuters, official release notes (ChatGPT, GPT-4, GPT-4o, o1/o3)

Further learning

AI Professional Certification

Recognising the pattern of AI’s historical development is the first step. DASCIN’s AI Professional certification prepares professionals to critically evaluate, govern, and adopt AI within their organisations, ensuring capability is built on sound judgement rather than momentary hype.
Explore Certification

Knowledge - Certification - Community

Other Knowledge Articles