In the conference hall at the hotel was an array of terminals whose software permitted interactions with computers hundreds or thousands of miles away. The invitees were encouraged to experiment—try out an air traffic control simulator, play chess against an electronic foe, explore databases. There had been some problems in setting up the demonstrations. At one point, a file meant to go to a printer in the hall was mistakenly directed to a robotic turtle, resulting in a wild dance. But it all worked when it had to, convincing the doubters and engaging their interest so effectively that, as one of the organizers said, they were "as excited as little kids." Within a month, traffic on the network increased by two-thirds. In a few more years, ARPANET hooked up with other networks to become a connected archipelago called the Internet. By the end of the 20th century, more than 100 million people were thronging Internet pathways to exchange e-mail, chat, check the news or weather, and, often with the aid of powerful search engines to sift for useful sites, navigate the vast universe of knowledge and commerce known as the World Wide Web. Huge electronic marketplaces bloomed. Financial services, the travel industry, retailing, and many other businesses found bountiful opportunities online. Across the world, the connecting of computers via the Internet spread information and rearranged human activities with seismic force.

All this began in an obscure branch of the U.S. Department of Defense called the Advanced Research Projects Agency, or ARPA. In the 1960s a number of computer scientists at universities and research laboratories across the country received ARPA funding for projects that might have defense-related potential—anything from graphics to artificial intelligence. With the researchers' needs for processing power steadily growing, ARPA decided to join its scattered mainframes into a kind of cooperative, allowing the various groups to draw on one another's computational resources. Responsibility for creating the network was assigned to Lawrence Roberts, a young computer scientist who arrived at ARPA from the Massachusetts Institute of Technology in 1966.

Roberts was aware of a promising approach in the ideas of an MIT classmate, Leonard Kleinrock, and he later learned of related work by two other communications experts, Paul Baran and Donald Davies. Kleinrock had written his doctoral dissertation on the flow of messages in communications networks, exploring the complexities of moving data in small chunks. At about the same time, Baran proposed a different kind of telephone network, which would turn the analog signal of a telephone into digital bits, divide the stream into blocks, and send the blocks in several different directions across a network of high-speed switches or nodes; the node nearest the destination would put the pieces back together again. Davies proposed a similar scheme, in which he called the chunks or blocks "packets," as in packet switching, and that name stuck.

Roberts, for his part, was convinced that the telephone system's method of routing signals, called circuit switching, was poorly suited for linking computers: to connect two callers, a telephone switch opens a circuit, leaving it open until the call is finished. Computers, however, often deliver data in bursts and thus don't need full-time possession of a connection. Packet switching seemed the obvious choice for ARPA's network, not only enabling several computers to share a circuit but also countering congestion problems: when one path was in heavy use, a packet could simply take another route.