From Dartmouth to Deep Learning: A Complete Guide to AI's History and Core Concepts

AI’s history is often told as a single, smooth line from 1956 to ChatGPT, but the real story includes two genuine funding collapses, a decades-long detour through symbolic reasoning, and a founding ambition that was, by any honest measure, wildly overconfident. This guide ties together where the field actually began, the specific milestones that defined each era, why it nearly died twice, and the core vocabulary you need to make sense of any of it.

Where it actually began

The term “artificial intelligence” wasn’t coined casually or gradually — it has a precisely documented origin. Computer scientist John McCarthy introduced it in a 1955 proposal for what became the 1956 Dartmouth Workshop, and who actually coined the term artificial intelligence? traces exactly why that specific term was chosen over existing labels like “cybernetics.” The workshop itself is remembered less for a single breakthrough achieved that summer and more for founding the field as a distinct discipline — see what was the Dartmouth Workshop and why does it matter? for the founding proposal’s genuinely audacious central claim: that essentially every aspect of intelligence could, in principle, be described precisely enough for a machine to simulate.

The program most historians point to as the field’s actual first — the Logic Theorist, presented at that same Dartmouth gathering — is covered in what was the first program considered AI by researchers?

The milestones that defined each era

AI’s history moves in a handful of genuinely discrete leaps rather than smooth continuous progress. Deep Blue’s 1997 chess win over Garry Kasparov mattered less for its technical approach — brute-force search, not learning — than for its symbolic weight, demonstrating a machine could outperform the best human mind in a domain long associated with deep strategic thought. The 2012 ImageNet result, where a deep neural network called AlexNet dramatically outperformed prior approaches, is the moment most researchers point to as the true start of the modern deep learning boom — see what made ImageNet and the 2012 deep learning breakthrough so significant? for why that specific combination of data, compute, and neural network architecture mattered so much.

The other pivotal technical moment came in 2017, when a Google research paper introduced the transformer architecture. What was the actual breakthrough behind the transformer architecture? explains the “attention” mechanism at its core — the innovation that made today’s large language models technically possible by letting models weigh relationships across an entire input simultaneously rather than processing it sequentially.

Why the field nearly died twice

It’s easy to assume AI progress has been a steady upward line, but the field went through two genuine funding collapses, sometimes called “AI winters.” What caused the first AI winter? walks through the mid-1970s collapse, driven by overpromised results and critical reports like the UK’s Lighthill Report. The second collapse, in the late 1980s and early 1990s, followed the commercial disappointment of expert systems and the parallel collapse of specialized AI hardware — both funding winters shared the same underlying pattern: promised capability outrunning what the technology could actually deliver.

Given that history, it’s a reasonable question whether today’s boom could end the same way. Are we at risk of another AI winter happening now? lays out both sides honestly: supporters of continued growth point to much deeper real commercial adoption than existed before prior winters, while skeptics point to diminishing returns from scaling and historically overpromised timelines as reasons for caution.

The vocabulary you actually need

A handful of terms get thrown around constantly, often without being defined. What’s the actual difference between AI, machine learning, and deep learning? clears up the nesting: AI is the broadest umbrella, machine learning is a specific approach within it, and deep learning is a further, more specific subset using multi-layered neural networks. What is a neural network, explained without the jargon? breaks down the actual building blocks — simple processing units, adjustable weights, and a training process that gradually tunes those weights toward better outputs.

Finally, what does narrow AI versus general AI actually mean? addresses one of the most consistently misused distinctions in public discussion: every AI system in use today, including the most capable large language models, is still considered narrow AI by most researchers, since none demonstrate the kind of robust, flexible, human-comparable general reasoning the term “AGI” actually describes.

Bottom line

AI’s history is a story of genuine leaps followed by genuine collapses, not a straight line — and the core vocabulary that gets used to describe where the field stands today only makes sense once you understand which specific era and which specific technical breakthrough it’s describing.