Where this idea comes from
The Perceptron
In 1958 Frank Rosenblatt, a psychologist from the Cornell Aeronautical Laboratory in Buffalo, New York, showed reporters a program he called the Perceptron. Running on an IBM 704 computer, it learned to tell punched cards marked on the left from cards marked on the right after about fifty tries, nudging its weights each time it was wrong — much the same idea this room uses.
One thing to hold: A stack of punched cards, each with a mark on the left or the right side.
Why it matters: The next day the New York Times reported that the Navy expected the machine to “walk, talk, see, write, reproduce itself and be conscious of its existence.” It could sort cards. In 1969 Marvin Minsky and Seymour Papert’s book Perceptrons showed that a single layer of these neurons can never learn some simple patterns — for one neuron like this room’s, because it can only draw one straight line. Networks with many neurons in layers, trained a different way, came later.
What we don’t know: How much Perceptrons slowed neural-network research is still argued about: some say it stopped the field for a decade, others that money and computers were the real limits.
Dates:
- 1958 The Perceptron learns to tell left-marked cards from right-marked ones.
- 1958 A newspaper says the Navy expects it to walk, talk and be conscious.
- 1969 Perceptrons shows what one neuron can never learn.
- 1986 A way to train many neurons in layers is published.