Theme 01 · Historical · Start Here

The History of AI:
From 1943 to Now

Artificial intelligence did not appear suddenly in 2022. It was built over eight decades — with breakthroughs, collapses, revivals, and paradigm shifts. Understanding this history makes the current moment legible: what's genuinely new, what's hype we've seen before, and why the stakes feel different this time.

Why this matters for the rest of the site Every story on Maxvaria — about labor, ownership, environmental cost, political governance — is a chapter in this longer history. The decisions being made right now about who controls AI, who benefits from it, and who bears its costs are not new decisions. They follow patterns visible across all eight eras below.
1943 – 1956 Foundations Origin

The mathematical and philosophical groundwork for artificial intelligence is laid by a small group of mathematicians, logicians, and neuroscientists. The field does not yet have a name.

Key Milestones
  • 1943McCulloch and Pitts publish the first mathematical model of a neuron — the conceptual ancestor of every neural network that followed.
  • 1950Alan Turing publishes "Computing Machinery and Intelligence" and proposes the Turing Test — can a machine think?
  • 1956The Dartmouth Conference coins the term "Artificial Intelligence." John McCarthy, Marvin Minsky, Claude Shannon, and others gather for a summer to define a new field.
Why it matters The Dartmouth Conference set an ambition — human-level reasoning in machines — that was wildly optimistic and has driven the field ever since, including its periodic collapses when the ambition outpaced the capability.
1957 – 1974 The Golden Age & First Hype Cycle Boom

Enormous optimism. Early programs solve algebra problems, speak English, play checkers. Researchers make bold predictions — machines will surpass human intelligence within 20 years. Government funding flows freely. The promises vastly exceed what the technology can actually deliver.

Key Milestones
  • 1957Frank Rosenblatt builds the Perceptron — an early neural network that learns to classify images. The New York Times predicts machines will walk, talk, and think.
  • 1965ELIZA, the first chatbot, created by Joseph Weizenbaum at MIT. Many users believe it understands them. Weizenbaum is disturbed by this.
  • 1966ALPAC report concludes machine translation is going nowhere. Early warning that AI's self-promotion was ahead of its capability.
Why it matters The hype cycle of the 1960s looks remarkably like today's — bold predictions, media excitement, government and investor enthusiasm. The pattern of overclaiming followed by reckoning has repeated every 15–20 years.
1974 – 1980 The First AI Winter Winter

Funding dries up. The Lighthill Report in the UK concludes AI has failed to deliver on its promises. DARPA cuts funding. The field contracts severely. Researchers who stay are considered eccentric.

Key Milestones
  • 1973The Lighthill Report concludes AI research has disappointed — machines could handle toy problems but not the complexity of the real world.
  • 1974US and UK governments dramatically cut AI research funding. Labs close. Researchers move to other fields.
Why it matters AI winters are what happens when investment outruns capability. The question many researchers ask today — are we in another bubble? — has historical precedent. The winters were real, painful, and lasted years.
1980 – 1987 The Expert Systems Boom Boom

A new approach revives the field: instead of general intelligence, build narrow systems that encode expert human knowledge in specific domains. Corporations invest heavily. Japan launches a major national AI initiative. Commercial AI briefly appears to have arrived.

Key Milestones
  • 1980XCON, an expert system for configuring DEC computers, saves the company $40M/year. Corporate AI investment surges.
  • 1981Japan's Ministry of Trade launches the Fifth Generation Computer project — a $850M national AI program meant to leapfrog the US.
  • 1985The AI industry reaches $1 billion in annual revenue. Companies like Symbolics, Lisp Machines, and IntelliCorp attract major investment.
Why it matters Expert systems were brittle — they worked inside narrow domains but couldn't generalize. The corporate and government investment that flooded in during the boom evaporated when this became clear. Sound familiar?
1987 – 1993 The Second AI Winter Winter

Expert system hardware becomes obsolete overnight. The Lisp machine market collapses. DARPA again cuts funding. Japan's Fifth Generation project quietly fails. The second winter is deeper and more disillusioning than the first because so much more money had been committed.

Key Milestones
  • 1987The market for specialized AI hardware collapses — $500M industry gone in a year as desktop computers outperform it at a fraction of the cost.
  • 1991Japan's Fifth Generation project ends without achieving its goals. The US Strategic Computing Initiative winds down.
  • 1993DARPA cuts AI funding again. The term "AI" becomes toxic in funding circles — researchers rebrand their work as "machine learning" or "knowledge engineering."
Why it matters The rebranding that happened here — AI becoming "machine learning" to avoid the stigma — is a pattern worth knowing. Marketing language changes; the underlying technology and its limitations often don't.
1993 – 2011 The Statistical & Machine Learning Era Paradigm Shift

A fundamental change in approach: instead of trying to encode human knowledge as rules, let systems learn patterns from data. The internet provides vast training data for the first time. Computing power grows. Progress is steady but unsexy — happening in research labs, not headlines.

Key Milestones
  • 1997IBM's Deep Blue defeats chess world champion Garry Kasparov. The first time a computer beats a human champion at a major intellectual game.
  • 1998Yann LeCun develops convolutional neural networks for handwriting recognition — the direct ancestor of modern image AI.
  • 2002Roomba launched — practical AI in a consumer product. AI begins quietly entering daily life without the label.
  • 2006Geoffrey Hinton publishes work on deep belief networks — beginning the quiet revival of neural networks that would transform everything.
Why it matters The people who kept working through two winters — Hinton, LeCun, Bengio — are the reason modern AI exists. The breakthroughs of 2012 onward were built on decades of unfunded, unglamorous work.
2012 – 2017 The Deep Learning Revolution Breakthrough

A genuine paradigm break. Deep neural networks trained on massive datasets and powerful GPUs begin outperforming every prior approach — not incrementally, but dramatically. Image recognition, speech, translation — one by one, systems exceed human performance. The research community is stunned. Investment follows.

Key Milestones
  • 2012AlexNet wins the ImageNet competition by a shocking margin — 26% error rate vs 16% for the neural network. The deep learning era begins.
  • 2014Google acquires DeepMind for £400M. The AI talent acquisition race begins in earnest.
  • 2016AlphaGo defeats Go world champion Lee Sedol. Go was believed to be a decade away from computer mastery. The field recalibrates its expectations upward.
  • 2017Google Brain publishes "Attention Is All You Need" — introducing the Transformer architecture that underlies every major AI system today.
Why it matters AlphaGo's victory is the moment the general public began taking AI seriously again. The Transformer paper is arguably the most consequential technical publication of the 21st century so far — everything that followed was built on it.
2017 – 2022 The Transformer Era Breakthrough

The Transformer architecture enables a new approach: train enormous models on vast amounts of text, and emergent capabilities appear that nobody explicitly programmed. Language models begin to write, reason, translate, and summarize at unprecedented quality. The scale of computation — and the infrastructure required to support it — grows by orders of magnitude.

Key Milestones
  • 2018Google releases BERT — a large language model that dramatically improves search and translation. AI begins powering products used by billions.
  • 2020OpenAI releases GPT-3 — 175 billion parameters. Researchers are startled by its ability to write coherent prose, answer questions, and complete code.
  • 2021DeepMind's AlphaFold 2 solves the protein folding problem — a 50-year scientific challenge. Immediate implications for drug discovery and disease research.
  • 2022DALL-E 2, Stable Diffusion, and Midjourney bring AI image generation to the public. For the first time, millions of non-technical people interact directly with generative AI.
Why it matters This is the era when the infrastructure story begins in earnest. Training GPT-3 required thousands of specialized GPUs running for weeks. The data center buildout, the energy consumption, the water cooling — all of it traces back to the computational demands of this era.
2022 – 2025 Generative AI Goes Public Now

ChatGPT reaches 100 million users in two months — the fastest consumer product adoption in history. Generative AI enters every industry simultaneously. Investment reaches unprecedented levels. The infrastructure arms race begins: data centers, power grids, water systems, and semiconductor supply chains are pushed to their limits. The public conversation about AI — its benefits, costs, governance, and risks — becomes unavoidable.

Key Milestones
  • 2022ChatGPT launched November 30. 1 million users in 5 days, 100 million in 2 months. The public AI era begins.
  • 2023Microsoft invests $10B in OpenAI. Google rushes Bard to market. The AI investment arms race goes into overdrive. Kenyan workers who trained ChatGPT's safety filters speak publicly about their conditions.
  • 2024Frontier models — GPT-4, Claude 3, Gemini Ultra — demonstrate reasoning, coding, and analysis capabilities that begin affecting professional labor markets. Data center construction announcements multiply. Community opposition begins organizing.
  • 2025AI attributed to 54,836 job losses tracked in the US alone. Community moratorium efforts grow from 8 to 78 active campaigns. Sanders/AOC introduce the AI Data Center Moratorium Act. The questions of who owns AI, who bears its costs, and who governs its development move from research papers to town halls.
Why it matters — and what comes next We are inside this era now, without the clarity of hindsight. What we know from history: the hype cycles are real, but so are the genuine breakthroughs. The winters came when investment outpaced capability and the promised returns didn't materialize. Whether the current moment is a genuine inflection point or another cycle depends partly on technology — and partly on decisions being made right now about governance, ownership, and who bears the costs.
The Pattern Across Eight Eras

AI history follows a repeating structure: a technical breakthrough generates optimism, optimism attracts investment, investment outpaces capability, a reckoning arrives. What's different this time is scale — the investment is larger, the capability gains are more real, and the infrastructure being built is physical and permanent in ways that code never was.

The data centers going up now will exist for decades. The energy and water systems being stressed by them will bear those consequences long after the hype cycle, whatever shape it takes, has resolved. History suggests humility about predictions. It also suggests that the structural questions — who owns this, who governs it, who pays for it — matter more than the technical ones.