1958: The Perceptron is unveiled to the press

No single milestone in computing or AI history is definitively dated to July 12, so this entry turns to the closest well-documented event: the public unveiling of the Perceptron to journalists on July 7, 1958, reported days later, on July 8, in the New York Times under the headline "New Navy Device Learns By Doing."

The Perceptron was built by psychologist Frank Rosenblatt at the Cornell Aeronautical Laboratory, with funding from the U.S. Office of Naval Research. It was an early attempt to model, in hardware, how a network of simple neuron-like units could learn from examples rather than follow explicit rules. The first version, the Mark I Perceptron, used a grid of 400 photocells wired to adjustable resistors, so that repeated exposure to labeled images could strengthen or weaken connections until the machine correctly classified simple patterns, such as cards marked on the left or right.

The press conference produced some now-famous overreach. Journalists reported that the Navy expected future perceptrons to walk, talk, see, write, reproduce themselves, and even be conscious of their own existence. None of that happened on Rosenblatt's timeline, but the underlying idea, that intelligence might emerge from adjustable weighted connections trained on data, proved remarkably durable.

The Perceptron's story is also a caution against hype. In 1969, Marvin Minsky and Seymour Papert published a rigorous critique showing that simple perceptrons could not solve certain basic problems, such as XOR, and enthusiasm for neural networks cooled for over a decade. It took the development of backpropagation and multi-layer networks in the 1980s, and far more data and computing power decades later, for the perceptron's descendants to become genuinely useful.

Every modern neural network, from the models recognizing speech on a phone to today's large language models, still relies on the same basic idea Rosenblatt demonstrated in 1958: adjustable connections tuned by exposure to examples. The gap between a press conference's bold predictions and a technology's real, slower path to usefulness is a pattern worth remembering whenever new AI breakthroughs are announced today.