The machine that would grow wiser

Frank Rosenblatt was a psychologist at the Cornell Aeronautical Laboratory in Buffalo, New York. In 1958 the journal Psychological Review published his paper on the perceptron, a simple model of how a network of brain-like cells might store and recognise patterns.

The key idea was that the machine was not given rules. It was shown examples and told when it got them wrong, and it adjusted the strength of its connections until it got them right. In July 1958 the US Navy, which funded the work, showed it in action: running on an IBM 704 computer, it learned after about 50 tries to tell cards marked on the left from cards marked on the right.

The press went much further than the demonstration. The New York Times headline called it the "embryo" of a computer designed to read and grow wiser. The New Yorker magazine called it the first real rival to the human brain.

In 1969 Marvin Minsky and Seymour Papert published a book, Perceptrons, showing that a single-layer perceptron simply cannot learn some kinds of pattern, however long it trains. Interest and funding for this style of research dropped sharply for years. Rosenblatt died in a sailing accident in 1971, long before his approach came back.

Both sides were partly right. The critique was correct about the simple, one-layer machine. But Rosenblatt's core idea, learning from examples by adjusting connections, is the ancestor of the neural networks inside every modern chatbot. It needed more layers, a better way to train them, more data and far faster computers. All of those came later.

For a chatbot user today, the perceptron is a warning about hype cycles. A real but small result became a headline about thinking machines, and the backlash buried a good idea for a generation. AI news still swings like this, from "it will change everything by next year" to "it is all a bubble". Neither headline is a good guide to what a tool can do for you.

The idea behind this caseLesson: How to keep up without burning out →

1. 1Separate what a system was shown to do from what people predict it will do.
2. 2Look for the actual demonstration behind a headline: what task, how well, under what conditions.
3. 3Treat both "AI will soon do everything" and "AI is useless hype" as claims that need evidence.
4. 4Test a tool on a small piece of your own work before you believe either the hype or the backlash.