n July and October 1948 the Bell System Technical Journal printed a long paper in two parts by a Bell Labs mathematician of thirty-two. It defined the information in a message, the capacity of a channel, and what noise costs, and it is very readable.

the document

Shannon, "A Mathematical Theory of Communication". A year later it was reprinted as a book with an introduction by Warren Weaver (Shannon & Weaver, 1949), and the title changed from "A" to "The". A transcription of the paper is hosted by the Harvard mathematics department: people.math.harvard.edu (PDF).

three sentences from it

“The fundamental problem of communication is that of reproducing at one point either exactly or approximately a message selected at another point.”
— Shannon 1948, introduction

“These semantic aspects of communication are irrelevant to the engineering problem.”
— the same paragraph

“If the base 2 is used the resulting units may be called binary digits, or more briefly bits, a word suggested by J. W. Tukey.”
— the next page

The second sentence is the one people argue about. Shannon was not saying meaning does not matter. He was saying the engineer has to design for every message that might be sent, not the one that will be, so the theory measures choice, not sense.

what to read first

Part I, section 3, "The series of approximations to English". Shannon generated text from the statistics of English, a letter or a word at a time, each step using a little more context. The zero-order sample is letters at random:

“XFOML RXKHRJFFJUJ ZLPWCFWKCYJ FFJEYVKCQSGHYD QPAAMKBZAACIBZLHJQD.”
— approximation 1, symbols independent and equiprobable

By the sixth step, choosing each word from the ones that follow the last word in English, it reads:

“THE HEAD AND IN FRONTAL ATTACK ON AN ENGLISH WRITER THAT THE CHARACTER OF THIS POINT IS THEREFORE ANOTHER METHOD FOR THE LETTERS THAT THE TIME OF WHO EVER TOLD THE PROBLEM FOR AN UNEXPECTED.”
— approximation 6, second-order word approximation

Why it lasts. That table is a small language model, built by hand from frequency counts. Every text generator since works on the same principle with vastly more context, and inherits the same property: the output is shaped by what usually comes next, not by what is true. See hallucination. Shannon came back to the question in 1951 with people as the predictors (Shannon, 1951).

in the tour: the rack room has a case labelled "Shannon, 1948" ···

Reference

1. Shannon (1948). A Mathematical Theory of Communication. Bell System Technical Journal. doi:10.1002/j.1538-7305.1948.tb01338.x
2. Shannon & Weaver (1949). The Mathematical Theory of Communication. Urbana: University of Illinois Press.
3. Shannon (1951). Prediction and Entropy of Printed English. Bell System Technical Journal. doi:10.1002/j.1538-7305.1951.tb01366.x

see also