The History of AI

The first formal description of how neurons could compute logic — laying the theoretical foundation 80 years before ChatGPT.

The paper that asked 'Can machines think?' and proposed the Imitation Game (later called the Turing Test) as a way to answer it.

John McCarthy, Marvin Minsky, Claude Shannon, and Nathaniel Rochester convene the conference that names the field. Their proposal predicted 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.' They expected it to take one summer.

The first hardware neural network. The New York Times reported it would 'be able to walk, talk, see, write, reproduce itself, and be conscious of its existence.' It could classify simple patterns.

Joseph Weizenbaum's chatbot at MIT uses pattern matching to mimic a Rogerian therapist. Users become emotionally attached to it — the first demonstration that humans anthropomorphize AI.

The first robot to reason about its own actions. Shakey combined vision, movement, and problem-solving. It moved at about 2 meters per hour.

British mathematician James Lighthill's report to the UK Science Research Council concludes that AI had failed to achieve its 'grandiose objectives.' Funding collapses across the UK.

DARPA slashes AI budgets. UK follows. Research continues in universities but the commercial promise evaporates.

Rule-based systems like XCON (used by DEC to configure computers) prove that AI can deliver commercial value in narrow domains.

Rumelhart, Hinton, and Williams publish the paper that makes training multi-layer networks practical. This idea, ignored for 20 years, becomes the foundation of modern deep learning.

Companies that spent millions on AI systems find them expensive to maintain and limited in scope. The specialized hardware market (Lisp machines) dies.

Researchers rebrand their work as 'machine learning,' 'computational intelligence,' or 'knowledge systems' to avoid the stigma.

IBM's chess computer wins a six-game match against the world champion. It used brute-force search, not learning — but it proved machines could outperform humans at complex tasks.

Geoffrey Hinton's group at Toronto shows that deep neural networks can be trained layer by layer. This paper reignites the field that became modern AI.

IBM's Watson defeats champions Ken Jennings and Brad Rutter. The system combined NLP, knowledge retrieval, and probabilistic reasoning — a preview of what LLMs would do at scale.

Krizhevsky, Sutskever, and Hinton's deep neural network crushes the competition in image recognition. This is the moment deep learning becomes undeniable.

Ian Goodfellow invents Generative Adversarial Networks — two neural networks competing against each other to generate increasingly realistic images.

Google researchers publish the architecture that powers GPT, BERT, and every modern LLM. The key insight: attention mechanisms can replace recurrence entirely, enabling massive parallelization.

OpenAI's 175-billion parameter model shows that scale alone can produce capabilities nobody explicitly programmed — writing code, translating languages, and reasoning about novel problems.

The fastest-growing consumer application in history. AI goes from research curiosity to dinner table conversation overnight.

Models that see images, write code, reason through problems, and pass professional exams. The capability ceiling keeps rising.

Meta's LLaMA, Mistral, and others approach frontier performance at a fraction of the cost. The moat question becomes central.

Autonomous systems that can browse the web, write and execute code, manage workflows, and operate tools with minimal human oversight.

Regulatory frameworks are taking shape. AI-generated content exceeds human-created content online. The labor market is shifting. And nobody knows what happens next — which is why forecasting matters.