1704: John Locke dies, leaving behind the blank slate
On 28 October 1704, the English philosopher John Locke died at Oates, the Essex home of his friends Sir Francis and Lady Masham, where he had spent his last years cataloguing plants and revising his writings. He left behind no laboratory, no machine, and no mathematics — only an argument about where knowledge comes from. That argument would travel, by a long and winding road, straight into the heart of artificial intelligence.
In his 1689 An Essay Concerning Human Understanding, Locke rejected the idea that the mind arrives pre-loaded with innate ideas, as Descartes and Leibniz supposed. Instead, he proposed that the mind begins as a tabula rasa, a blank slate, filled in only through sensation and reflection — through experience. Knowledge, on this view, is built, not given.
This empiricist wager shaped two centuries of thought before it ever touched a machine. It fed into associationist psychology, which held that ideas connect through repeated pairing rather than innate logic. It resurfaced in behaviorism's insistence that learning is shaped entirely by stimulus and response. And it reappeared, transformed, in the early neural-network tradition:
- Donald Hebb's 1949 learning rule — "cells that fire together wire together" — was explicitly a mechanism for building knowledge from experience.
- Frank Rosenblatt's Perceptron, trained on examples rather than programmed with rules, embodied Locke's wager in silicon.
- Today's deep learning systems are tuned not by hand-coded knowledge but by exposure to vast training data — a mechanized, mathematical descendant of "nothing in the intellect that was not first in the senses."
Locke, of course, intended none of this. He was arguing against theologians and rationalists about the origins of human ideas, not anticipating gradient descent. But the fault line he drew — innate structure versus learning from experience — is exactly the one that still runs through AI research, in debates over how much inductive bias a model needs built in versus how much it can discover purely from data. Large language models, trained on oceans of text rather than hand-written grammars, are in a real sense Locke's blank slates at industrial scale, even as critics argue that some innate scaffolding remains indispensable. Three centuries on, the question Locke posed at his death bed in Essex is still being tested, daily, in every training run.