This is Phase 0, Episode 2 of the OCDevel AI course: a narrative history of artificial intelligence, built around two throughlines. The first is a pendulum between symbolic, logic-based AI and connectionist, learn-from-data AI, where each era's failure seeded the next era's breakthrough. The second is the hype cycle that produced two named "AI winters."
We start in prehistory: the golem and Talos myths, Hobbes calling reasoning "reckoning," Leibniz's dream of settling arguments by calculation, and Boole's Laws of Thought turning logic into algebra. Foundations follow: Turing's machine (1936) and his imitation game (1950), and the McCulloch-Pitts neuron (1943).
The field is named at the Dartmouth workshop (1956). The symbolic golden years bring the Logic Theorist, the Physical Symbol System Hypothesis, ELIZA, SHRDLU, and Rosenblatt's perceptron. Then the first winter: Minsky and Papert's Perceptrons and the Lighthill report. The expert-systems boom (DENDRAL, MYCIN, XCON, Japan's Fifth Generation) ends in a second winter.
The connectionist revival arrives with backpropagation (1986), then SVMs and Deep Blue (1997). Deep learning ignites with ImageNet, AlexNet (2012), and AlphaGo (2016). Finally the transformer (2017), GPT-3, foundation models, ChatGPT, and today's agents. Closing lesson: Sutton's Bitter Lesson.
News brief: Microsoft's seven MAI models at Build, the Mayo Clinic healthcare model, NVIDIA's Cosmos 3, Anthropic's IPO filing, a Trump AI executive order, and xAI's Grok updates.