How Machines Learn From Mistakes: The Perceptron (1958)

In 1958 a newspaper said the Navy had built a computer that would "read and grow wiser". What had it really learned? Whether a mark sits on the left or the right of a card. This video untangles how a machine can learn from its own mistakes. Room 4 of Tech Untangled: AI explained from zero, one room at a time. No jargon, no hype, every fact sourced. ━━━━━━━━━━━━━━━━━━━━ IN THIS EPISODE • Frank Rosenblatt's perceptron (1958), step by step, with a left-or-right card game • Why a "weight" is just how much something counts • What a neural network is: many of these little cells linked together • Arthur Samuel's checkers program (1959): looking ahead, scoring boards, learning from its own games • Why the famous "without being explicitly programmed" line is not in his paper ━━━━━━━━━━━━━━━━━━━━ CHAPTERS 00:00 A 1958 headline vs a left-or-right card 01:03 One brain cell: the perceptron 01:20 The sticker game 02:57 Where this breaks 03:17 Weights and neural networks 03:37 1959: Arthur Samuel's checkers program 04:57 Playing against itself 05:33 The definition that isn't in his paper 06:05 Your turn 06:23 Recap 06:39 Next room: 1966 ▶ Watch the series in order:    • Tech Untangled: AI from zero   ▶ Previous room: How Did AI Get Its Name? :    • How Did AI Get Its Name?   ▶ Next room: 1966, the chatbot that turned your words into questions, and people told it their secrets ▶ Subscribe so you don't miss a room:    / @techuntangledai   Narration uses a synthetic voice. Script, research and illustrations by Becher Zribi. Sources: Rosenblatt (1958) Psychological Review 65(6); Samuel (1959) IBM Journal of Research and Development 3(3); Cornell Chronicle (2019).