1928: Frank Rosenblatt, father of the Perceptron, is born

On 11 July 1928, Frank Rosenblatt was born in New Rochelle, New York. A psychologist by training, Rosenblatt became one of the most consequential — and most controversial — figures in the early history of artificial intelligence when he invented the Perceptron, one of the first artificial neural networks capable of learning from examples.

Rosenblatt built on earlier work by Warren McCulloch and Walter Pitts on artificial neurons, but he went further: he designed an algorithm that let a network of simple, threshold-based units adjust their own connection weights after each example, gradually improving at a task without being explicitly programmed to perform it. In 1958, with funding from the US Office of Naval Research, he unveiled the Mark I Perceptron, a room-sized machine built to recognise simple visual patterns using a 20x20 grid of photocells wired to adjustable resistors.

The press coverage was extraordinary for its time. The New York Times reported in 1958 that the Navy expected the Perceptron to eventually "walk, talk, see, write, reproduce itself and be conscious of its existence," a level of hype that set unrealistic expectations for a field still in its infancy. When Marvin Minsky and Seymour Papert published their 1969 book "Perceptrons," which mathematically demonstrated the limits of single-layer networks — most famously, their inability to learn the XOR function — funding and enthusiasm for neural network research collapsed for over a decade, a period historians now call the first "AI winter" for connectionist approaches.

Rosenblatt died on 11 July 1971 — his 43rd birthday — in a boating accident in the Chesapeake Bay, before he could see his ideas vindicated. It would take until the 1980s, with the popularisation of multi-layer networks and the backpropagation algorithm, for the field to rediscover the promise of the architecture he pioneered.

• 1958: Rosenblatt's Mark I Perceptron is demonstrated at Cornell Aeronautical Laboratory.
• 1969: Minsky and Papert's "Perceptrons" curtails funding for neural network research.
• 1986: Backpropagation, popularised by Rumelhart, Hinton and Williams, revives multi-layer neural networks.

Every modern deep learning system — from image classifiers to large language models — descends conceptually from the simple learning rule Rosenblatt built into the Perceptron: adjust your weights based on your errors, and improve. He did not live to see neural networks become the dominant paradigm in AI, but the museum marks his birth, and the anniversary of his death on the same date, as one of the field's great, unresolved ironies.