he first perceptron was a program. The second was a machine, built so that the learning would happen in wires and motors you could point at.

what it is

The Smithsonian's record (National Museum of American History, 1958) describes the Mark I as built in 1958 at the Cornell Aeronautical Laboratory in Buffalo, New York, with funding from the Office of Naval Research and the Rome Air Development Center. It has three kinds of unit, following Rosenblatt's theory: sensory (S) units, a grid of photocells that took in an image; association (A) units; and response (R) units. Connections were made on a plugboard, and the adjustable weights were potentiometers. It went to the Smithsonian in 1967 and is in the collection of the National Museum of American History, not currently on display.

why a machine

A program on a general-purpose computer could be dismissed as arithmetic. A machine that changed its own wiring made the claim physical. It also made the limits visible: a single layer of adjustable weights can only learn some kinds of distinction, as Minsky & Papert later proved.

The book of the theory is (Rosenblatt, 1962). The newspaper story of the first demonstration is in the archive.

in the tour: the room of machines that learn has a placard called "The perceptron, 1958" ···

Sources

1. National Museum of American History (1958). Electronic Neural Network, Mark I Perceptron. Object record 1987.0819.01, Smithsonian Institution. americanhistory.si.edu/collections/object/nmah_334414
2. Rosenblatt (1958). The perceptron: A probabilistic model for information storage and organization in the brain. Psychological Review. doi:10.1037/h0042519
3. Minsky & Papert (1969). Perceptrons. Cambridge, MA: MIT Press.
4. Rosenblatt (1962). Principles of Neurodynamics. Washington, DC: Spartan Books.

see also