The Idea Factory: Bell Labs and the Great Age of American Innovation

by Jon Gertner

  Q: You do other things besides analyze communications, though, don’t you, Dr. Shannon? Didn’t I see an article in Scientific American by you about a chess-playing machine? Don’t tell me the Laboratories is interested in chess.

  Shannon: No—but interested in potentiality of computer machines. They can solve complex problems in a few minutes.

  Q: But how does chess fit into the picture?

  Shannon: A test to see what we can do with computing machines.

  Q: Would a chess-playing machine of the type you are talking about play perfect chess?

  Shannon: No, make same kind of errors as human players….

  Q: Would you say that these computing machines are capable of “thinking”?

  Shannon: Depends on how you define “thinking”—memory, decisions, but must all be programmed into the machine.

  Q: What sort of things will the Telephone Company use the computers for?

  Shannon: The telephone exchange itself is a type of computer.1

  At the time, Shannon hadn’t yet built a machine; computers of the era, even the simplest ones, were bulky and complex, and so he admitted it would be too expensive to use a computer for “so trivial a problem” as chess. But it’s unlikely he actually believed the problem to be trivial, and not long afterward he did in fact build a primitive chess computer. Shannon’s first paper on the subject—the one from which Scientific American adapted an article—happened to be the first paper ever written on chess programming. Much like his work on cryptography and information, it combined philosophical and mathematical elements, exploring the purpose of a chess machine as well as the logical theory behind its possible mechanisms. It also contained something unusual: an explanation that was meant to clarify why computer chess, a radical notion in 1949, could prove useful. “It is hoped that a satisfactory solution of this problem will act as a wedge in attacking other problems of a similar nature and of greater significance.”2 If you could get a computer to play chess, in other words, you could conceivably get it to route phone calls, or translate a language, or make strategic decisions in military situations. You could build “machines capable of orchestrating a melody,” he suggested. And you might be able to construct “machines capable of logical deduction.” Such machines could be useful as well as economical, he offered; they could ultimately replace humans in certain automated tasks.

  Almost surely, these justifications for the chess program would have placated his bosses at the Labs, worried as they were that some of their researchers could stray too far from the subject of communications. Regulatory anxiety created a background hum for the top managers at Bell Labs: What if government officials became alert to the fact that the phone monopoly was using its customers’ dollars to fund the research of a game-playing eccentric? Shannon’s offering—a few crumbs of justification—were about as much as he would give. He would acknowledge that building devices like chess-playing machines “might seem a ridiculous waste of time and money. But I think the history of science has shown that valuable consequences often proliferate from simple curiosity.”3 “He never argued his ideas,” Brock McMillan says of Shannon. “If people didn’t believe in them, he ignored those people.”

  “WHAT DO YOU GIVE A GUY like Claude for Christmas?” his wife, Betty, asks. “He liked Erector Sets, and so I bought the biggest Erector Set I could find and it was 50 bucks.”4 At first, Shannon began to stay up all night with the Erector Set building small machines. “He built a little turtle that walked around the house,” Betty says. “It would bump into something and back off and walk the other way. And then he built the mouse.” The mouse was named Theseus. It was a small object, built of wood with copper wire whiskers, which was intended to search for a piece of electronic “cheese” within a maze that Shannon also built. Theseus was named in winking recognition of the mythical Greek hero who found a way to navigate a labyrinth ruled by a deadly minotaur.

  Shannon did the project at home—just as he had the information paper—thus making it a surprise to almost everyone on the Labs staff when he brought it in.5 But in fact the mouse was far less ingenious than the maze, which was about the size of a small kitchen table and which Shannon had fitted with sliding aluminum panels so its pattern could be easily reconfigured. “I decided that my mouse would be basically a bar magnet, moved by means of an electromagnet under the floor of the maze,” Shannon explained. “The bar magnet, covered by a mouse-like shell, could be turned, and when it hit a wall of the maze could signal a computing circuit. The computer would then cause the mouse to try a different direction.”

  Watching the mouse trot through the maze in its first run-through wasn’t exactly breathtaking. Theseus would move slowly through the labyrinth until he hit a wall; he would bump against the wall and then turn in various directions until he finally found an open route. He didn’t move quickly. But he was unerring in ultimately calculating which way to go. What was interesting about Theseus was not that he could successfully navigate the aluminum labyrinth. What was interesting was that he (or, more precisely, the relays underneath the floor of the maze) could learn while he navigated the maze, and could likewise remember the location of walls and routes and whether it made sense to turn north, east, west, or south. Therefore, on his second try, Theseus could make it through the maze much more quickly—perhaps in as little as fifteen seconds. Or Shannon could actually pick up Theseus and place him anywhere in the maze and he could find his way out.

  Among the researchers at the Labs the mouse was wildly popular. The legal department was less enthusiastic, seeing little value in the patent they obtained for Shannon on the gadget’s circuitry.6 This was, after all, a communications company—and one that was limited by the federal government to the telecommunications business. But complaints aside, Theseus was a boon for the Labs in making Shannon a minor celebrity in a way that information theory never had. The Labs produced a short movie about the maze and the mouse, hosted by a thin and dapper Shannon, who narrated in his folksy midwestern manner. And soon Time magazine came calling, too. “Theseus Mouse is cleverer than Theseus the Greek,” the magazine’s writer noted, “who could not trust his memory but had to unwind a ball of string to guide him out of the labyrinth.”7 In large part, this observation matched Shannon’s intuition that machines would someday be smarter than men in some respects. Shannon often rhapsodized about the human brain and the inimitable processing power of its billions of neurons. But he believed without question that machines had the potential to do calculations and perform logical operations and store numbers with a speed, efficiency, and accuracy that would soon dwarf our own. It was only a matter of time. Often, he said, he was rooting for the machines.

  “I wish you could come east some time so I could show you the beautiful new laboratories we have here at Murray Hill, and some of the many new scientific developments in progress,” Shannon wrote to Irene Angus, a beloved high school science teacher of his, just after the Time story appeared in 1952. “Probably the most important is work on the transistor, the small germanium device which competes with the vacuum tube. I consider it very likely the most important invention of the last fifty years. As soon as the final manufacturing ‘bugs’ are out and large scale production starts, electronic devices that we only dream about today will become realities.” Shannon told his teacher that the electronic mouse Time described was “a demonstration device to make vivid the ability of a machine to solve, by trial and error, a problem, and remember the solution.” His fondest dream, he explained, was to now try and build a machine that actually thinks, learns, and communicates with humans.

  He added, “I am currently trying to design a machine that will be able to repair itself.”8

  BY THE EARLY 1950S, Shannon’s admirers from around the world began to seek him out. They wrote to the oracle at Bell Labs to ask about computers or chess or information theory, and then tried to tease out what he was thinking and why he was thinking it. Requests also came through official channels: In London in 195
0, during one of his periodic swings through Europe, Mervin Kelly wrote to Ralph Bown at Murray Hill to say that a British contingent at the Imperial College of Science and Engineering was insisting on hosting Shannon for a visit. “They hope to build a Communication Theory Conference around Shannon,” Kelly said. “His work has made a great impression here.”9 The following year, the director of the Central Intelligence Agency, Walter Bedell Smith, contacted Kelly for help with cryptography. “We urgently need the assistance of Dr. Claude E. Shannon of your Company who, we are informed on the best authority, is the most eminently qualified scientist in the particular field concerned.”10

  Still, most of the letters to Shannon came from academics or chess enthusiasts or weekend tinkerers, schoolchildren or hobbyists wanting to know more about Theseus. On occasion Shannon would answer the letters; more often than not, he would let them languish in piles and folders in his office. He would frequently receive letters from some of the most notable scientists in the world. And these, too, would languish. David Slepian recalls that the letters would eventually get herded into a folder he had labeled “Letters I’ve procrastinated in answering for too long.” On rare occasions when Shannon did reply to someone whose original query he had pushed aside, he would begin, I am sorry to be so slow in returning this, but…. It seemed lost on Shannon that the scientist who had declared that any message could be sent through any noisy channel with almost perfect fidelity was now himself a proven exception. Transmissions could reach Claude Shannon. But then they would fail to go any farther.

  Information theory, in the meantime, was getting ready for the masses. In 1953, one of the premier science journalists of the era, Francis Bello of Fortune magazine, profiled Shannon along with Norbert Wiener, an MIT mathematician who was putting forward theories on the command and control of machines, a discipline closely related to Shannon’s work on information. Wiener called his work cybernetics. “Within the last five years a new theory has appeared that seems to bear some of the same hallmarks of greatness,” Bello wrote. “The new theory, still almost unknown to the general public, goes under either of two names: communication theory or information theory. Whether or not it will ultimately rank with the enduring great is a question now being resolved in a score of major laboratories here and abroad.”11 Bello didn’t say whether Shannon was working toward any kind of resolution, and in fact he wasn’t. At the Labs, Shannon had continued to work on various aspects of information theory—on coding, for instance—but he was increasingly drawn to computing. Often his ideas were incorporated into the machines he was building—machines that were constructed for research, for amusement, or for both. Just as he had challenged people to think of information as a word bereft of meaning, he seemed to be challenging people to see whether the things he was building had any deeper significance. Game-playing machines “may seem at first an entertaining pastime rather than a serious scientific study,” he said at the time; he noted that there was “a serious side and significant purpose to such work, and at least four or five universities and research laboratories have instituted projects along this line.”12 Shannon didn’t say, however, whether the things he was building on occasion—one was a large desk calculator, known as THROBAC, that did calculations only in roman numerals13—had any deeper purpose apart from curiosity or wit. Often they were merely responses to somewhat mundane questions: Could it be built? How would you build it? Shannon’s self-proclaimed “ultimate machine,” for instance, seemed a jesting commentary on the subject of the meaning of his tinkerings. It was a wooden box with a single switch. A user hit the switch to turn it on, the box top opened and a mechanical hand reached out and turned the switch off, then the hand retreated into the box and the top closed.

  In his speeches in the 1950s, Shannon seemed to make the point that he was not necessarily interested in automated machines per se. He was interested in how machines interact with other machines (as in the telephone switching system) and how they interact with human operators (as in a chess machine). In the latter instance, there was a psychological aspect that seemed curious to him: “We hope that research in the design of game playing machines will lead to insights in the manner of operation of the human brain.”14 To his colleagues at Bell Labs, who had firsthand experience of Shannon’s electronic endeavors, this rang true. Shannon’s games and machines from those years weren’t only about engaging opponents in matches where a computer was pitted against man. Sometimes they were about engaging opponents in games that depended not only on outwitting them but in subtly deceiving them. (Or making fun of them: One game of Shannon’s had a computer make sarcastic comments after each move.) Many of these were games in which the unsuspecting player would invariably lose—and thus Shannon, as the games’ creator, would win.

  His friend David Slepian recalls one of Shannon’s electronic games that was peculiar on two accounts.15 First, the computer was programmed to take an absurd and arbitrary amount of time to calculate its next move, all to give its human opponent a false impression of formidable strategizing. Second, the design of the game board was done so artfully, and with such mathematical precision, that the opponent didn’t realize it was created so that the computer would have extra squares on which to move its pieces. “The person playing it from the other side couldn’t notice that one side, from one direction, was smaller,” Slepian says. Slepian played Shannon’s computer several times. He couldn’t win. No one could. And obviously that was interesting as well as pleasing to Shannon. “My characterization of his smartness is that he would have been the world’s best con man if he had taken a turn in that direction,” Slepian says.

  In the early 1950s, one of the few people at the Labs whom Shannon would actually seek out at lunchtime was David Hagelbarger, a self-described “tall, skinny kid” from Ohio who went to the University of Michigan and got a PhD under Robert Millikan at Caltech. Hagelbarger always wore a bow tie, in part because he liked to work in the Labs’ machine shop: So dressed, he wouldn’t have to worry, while working a lathe or drill press, about the safety hazard of a dangling cravat. His mechanical skills were in part what drew Shannon to him. “He would just come around about lunchtime,” Hagelbarger recalls. “He didn’t make a formal appointment.” Usually the two would talk about ideas, both serious and frivolous, some of which Hagelbarger would build for the two of them in the shop. Around 1954, Hagelbarger on his own built a machine with electromechanical relays that would guess whether a human player had chosen heads or tails on a coin. By taking into account a human player’s tendency to fall into various patterns of guessing, the machine could beat a player about 53 percent of the time, a success rate that Bob Lucky, a Bell Labs executive, later noted, “by chance alone would happen with probability less than one in 10 billion” over the course of about 10,000 trials.16 Shannon was fascinated by his friend’s machine, so he built his own—a simplified version with a smaller memory but greater calculating speed. “After considerable discussion concerning which of these two machines would win over the other we decided to put the matter to an experimental test,” Shannon recalled. The men built a third “umpire machine” to pass information between the two competing machines and keep score. Shannon recalled, “The three machines were plugged together and allowed to run for a few hours to the accompaniment of small side bets and large cheering.”

  For Shannon, it was an ideal experiment, with an almost optimal range of psychological and technological combinations. Here were machines working with (and against) machines; here were men working with (and against) machines. And here were men working with (and against) each other. Even with the perspective of a half century, it is hard to say whether, or how much, the contest meant to the evolution of computing. But it certainly meant something to Shannon. His machine won, by an “outguessing” ratio of about 55 to 45. The victory thrilled him.

  . . .

  NOT ALL MACHINES HAD TO BE ELECTRONIC or built with complicated relay circuits to earn Shannon’s devotion. By his own admission, he apprecia
ted movement the way an aesthete appreciates beauty.

  One year, Betty gave him a unicycle as a gift. Shannon quickly began riding; then he began building his own unicycles, challenging himself to see how small he could make one that could still be ridden. One evening after dinner at home in Morristown, Claude began spontaneously to juggle three balls, and his efforts soon won him some encouragement from the young kids in the apartment complex. There was no reason, as far as Shannon could see, why he shouldn’t pursue his two new interests, unicycling and juggling, at Bell Labs, too. Nor was there any reason not to pursue them simultaneously. When he was in the office, Shannon would take a break from work to ride his unicycle up and down the long hallways, usually at night when the building wasn’t so busy. He would nod to passersby, unless he was juggling as he rode. Then he would be lost in concentration. When he got a pogo stick, he would go up and down the hall on that, too.

  Here, then, was a picture of Claude Shannon, circa 1955: a man—slender, agile, handsome, abstracted—who rarely showed up on time for work; who often played chess or fiddled with amusing machines all day; who frequently went down the halls juggling or pogoing; and who didn’t seem to care, really, what anyone thought of him or of his pursuits. He did what was interesting. He was categorized, still, as a scientist. But it seemed obvious that he had the temperament and sensibility of an artist.