The Internet

The internet is a network of networks joined by a shared agreement — the TCP/IP protocol suite — rather than by any central switchboard. Messages are broken into small packets, each stamped with a destination address, and every machine along the route simply forwards packets toward that address; there is no master exchange whose failure stops the whole system. Packet switching was conceived independently by Paul Baran at RAND and Donald Davies at Britain's National Physical Laboratory in the early 1960s, and Vint Cerf and Bob Kahn turned it into TCP/IP, which became the network's sole language on 1 January 1983.

For all of prior history, moving information meant moving atoms — a letter, a book, a courier — and cost climbed with distance. The internet collapsed that cost to approximately zero: sending a message across the planet became indistinguishable from sending it across a room. In doing so it accidentally assembled the largest corpus of recorded human language and behavior ever created — a side effect that would matter enormously once machines learned to read it.

The ARPANET, funded by the U.S. Defense Department's ARPA, sent its first message on 29 October 1969 — the "LO" that opens this node — linking four university sites through refrigerator-sized packet-switching computers called Interface Message Processors. Through the 1970s that single network had to be joined to others built on different hardware — radio links, satellite links, campus and corporate LANs — and Cerf and Kahn's TCP/IP solved it by demanding nothing of the underlying networks except that they carry packets, pushing all reliability out to the endpoints. On "flag day", 1 January 1983, every ARPANET host switched to TCP/IP at once, and the internet as a single coherent system began.

The World Wide Web (1989–91) was built on top of it as one application among many, and email, streaming, and cloud computing followed — whole industries that assume a universal, always-on network reaching every device. By making the entire body of recorded human language and behavior addressable in one place, it also quietly assembled the training data that deep learning and large language models would later feed on. Almost every downstream node in this corpus after 1990 presupposes it.