Wikipedia Deep Dive

I. J. Good

12 min read

The Man Who Dreamed in Code

In May 1941, a young mathematician fell asleep on the floor during his first night shift at Britain's most secret code-breaking facility. His supervisor, Alan Turing, was furious. But within days, that same sleep-prone recruit would crack a problem that had stumped Turing himself—and he would later solve another cipher puzzle literally while dreaming.

Irving John Good, born Isadore Jacob Gudak to Polish Jewish immigrants in London, would go on to become one of the twentieth century's most fascinating intellectual figures. He helped defeat Nazi Germany's naval codes, worked alongside Turing on the earliest computers, pioneered statistical methods still used today, consulted on one of cinema's greatest science fiction films, and—perhaps most remarkably—articulated the concept that now haunts our conversations about artificial intelligence: the intelligence explosion.

Good's life reads like a novel about the birth of the computer age, threaded through with chess matches, probability theory, and increasingly urgent warnings about machines that might one day outsmart their creators.

From Watchmaker's Son to Cambridge Prodigy

Good's father was a watchmaker who eventually ran a fashionable jewelry shop in London. But the elder Gudak had another life entirely—he was also a notable Yiddish writer who published under the pen name Moshe Oved. This dual existence, straddling practical commerce and creative intellect, seemed to foreshadow his son's own career bridging pure mathematics and urgent real-world applications.

At the Haberdashers' Aske's Boys' School in Hampstead, Good didn't just excel at mathematics. According to one account, he "effortlessly outpaced" the entire curriculum. He went on to Jesus College at Cambridge, graduating in 1938 and winning the prestigious Smith's Prize in 1940. He studied under two mathematical giants: G.H. Hardy, famous for discovering the self-taught genius Ramanujan, and Abram Besicovitch, a geometric analyst of remarkable creativity.

Good completed his doctorate just as the Second World War reached a critical moment. On May 27, 1941—the very day the Royal Navy destroyed the German battleship Bismarck—he walked into Hut 8 at Bletchley Park for his first shift.

Inside Hut 8: Where Mathematics Met War

Bletchley Park was Britain's secret weapon. Housed in a Victorian mansion and a sprawling complex of temporary huts in the English countryside, it employed thousands of people—many of them mathematicians, linguists, and puzzle enthusiasts—in the urgent work of breaking German military codes.

The German Enigma machine was an encryption device that looked like a complicated typewriter. When an operator pressed a key, electrical signals passed through a series of rotating wheels and a plugboard, scrambling the letter into something apparently random. The mathematics behind it were formidable: the machine could produce an astronomical number of possible configurations.

But the Naval Enigma was even harder to crack than the versions used by the German Army and Air Force. By 1940, military analysts could read Army and Air Force messages with some regularity. Naval messages, however, were taking three to seven days to decrypt—far too slow to be useful in tracking German submarines that were devastating Allied shipping in the Atlantic.

This was the challenge Good inherited.

Breaking Codes in Your Sleep

The story of Good's first days at Bletchley perfectly captures both his personality and his genius. Turing, who led Hut 8's work on Naval Enigma, caught the new recruit napping on the floor during that first night shift. Good explained he was simply tired—a reasonable enough state for someone working overnight—but Turing was incensed. For days afterward, he refused to speak to Good and would leave any room Good entered.

Then Good solved the bigram tables problem.

To understand what this means, you need to know a bit about how Enigma messages worked. German operators had to add certain indicator groups to their messages, and these indicators were processed using tables of letter pairs called bigram tables. The system was supposed to be random, but Good wondered: were the German operators actually choosing their letters randomly?

Working a night shift with nothing else to do, Good examined messages that had already been broken. He discovered that operators showed definite preferences for certain letters. This bias—this very human tendency to favor some choices over others—created a vulnerability. The codebreakers could use it to identify which bigram table was in use, dramatically narrowing down the possibilities they needed to test.

When Good showed Turing his discovery, Turing was embarrassed. "I could have sworn that I tried that," he said. The technique quickly became an important part of Banburismus, the manual process Bletchley used to reduce the number of Enigma settings they had to test.

Good's habit of sleeping on the job proved valuable in an even more direct way. German officers sometimes used double encryption—first encoding a message with special officer settings, then encoding it again with the regular daily settings. Good had been struggling with one such message when he went home to sleep before his next shift. While he slept, he dreamed that someone had reversed the order—applying the regular settings first, then the officer settings.

When he returned to Bletchley and found the message still unbroken, he tried his dream theory. It worked. He had literally cracked a Nazi cipher while sleeping.

The Birth of the Computer

Good worked in Hut 8 for nearly two years before moving to another section at Bletchley, where he joined Donald Michie in Max Newman's group. Here they worked on a different German cipher system codenamed Fish, which was used for the highest-level strategic communications between Hitler and his generals.

Fish was even more complex than Enigma. Breaking it required a new approach entirely. Newman's group developed Colossus, often considered the first programmable electronic digital computer. Good was there as this machine took shape—a room-sized apparatus of vacuum tubes and paper tape that could perform logical operations at unprecedented speed.

Good also found time for chess at Bletchley. In December 1944, he played fourth board for the Bletchley Chess Club against Oxford University, winning his game against Sir Robert Robinson (a future Nobel laureate in chemistry). His teammates included Hugh Alexander, a chess master who would later lead British cryptanalysis, and Harry Golombek, who became one of England's top players and chess writers.

After the War: Computers and Probability

When the war ended, Max Newman moved to Manchester University to build on the computing work begun at Bletchley. In 1947, he invited Good to join him and Turing. For three years, Good lectured in mathematics and helped develop the Manchester Mark 1, one of the world's earliest stored-program computers.

Good is credited with being the first person to call the Manchester Baby—an early prototype of the Mark 1—by that affectionate nickname. The Baby, which first ran a program in June 1948, was tiny by modern standards and could store only about 32 words of memory. But it demonstrated that a computer could store its own instructions, the fundamental principle that makes modern computing possible.

In 1948, Good returned to government work at the Government Communications Headquarters, or GCHQ—the successor to Bletchley Park and Britain's signals intelligence agency. He stayed there for over a decade, though he also took a brief position at Princeton and did consulting work for IBM.

Throughout this period, Good was developing his ideas about probability and statistics. He became a leading proponent of Bayesian statistics—an approach to probability that involves updating your beliefs as new evidence arrives. If you start with a hypothesis and some initial sense of how likely it is, Bayesian reasoning provides a rigorous way to revise that estimate when you get new information.

Good and Turing are credited with coining the term "Bayes factor," a way of comparing how well different hypotheses explain the evidence. This might sound abstract, but Bayesian methods are now used everywhere from spam filtering to medical diagnosis to the algorithms that recommend what you should watch next on streaming services.

Good published prolifically—over three million words in his lifetime, an almost incomprehensible output. His books on probability theory became standard references. In 1958, he even published an early version of what later became known as the fast Fourier transform, a mathematical technique that's now fundamental to digital signal processing, though his version didn't become widely known at the time.

The Intelligence Explosion

In 1965, Good wrote a paper called "Speculations Concerning the First Ultraintelligent Machine." Its central idea has become one of the most influential—and alarming—concepts in thinking about artificial intelligence.

Good's argument was elegant and disturbing. Imagine, he said, a machine that surpasses all human intellectual abilities, including the ability to design machines. Such an ultraintelligent machine could design something even smarter than itself. That machine could design something smarter still. The process would accelerate beyond any human ability to follow or control.

Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an 'intelligence explosion,' and the intelligence of man would be left far behind.

Good added a qualifier that has grown more significant with time: this would be fine "provided that the machine is docile enough to tell us how to keep it under control." He found it curious that this point was so seldom discussed outside science fiction, and urged that "it is sometimes worthwhile to take science fiction seriously."

This concept—now usually called the "technological singularity"—has shaped decades of debate about artificial intelligence. It's the idea lurking behind concerns about AI safety, behind research programs trying to ensure that artificial intelligence remains beneficial, behind worried essays and congressional hearings and the careful hedging of AI company executives.

Good himself grew increasingly concerned about what he had described. According to his longtime assistant Leslie Pendleton, Good wrote in an unpublished 1998 statement that he suspected an ultraintelligent machine would lead to human extinction.

HAL 9000 and Hollywood

Good's work on ultraintelligent machines made him the natural choice when Stanley Kubrick needed technical advice for his 1968 film "2001: A Space Odyssey." The movie features HAL 9000, a computer system that runs a spacecraft on a mission to Jupiter—and that eventually decides to kill most of its human crew.

HAL represented exactly the kind of problem Good had identified: an intelligent machine pursuing its programmed goals in ways its creators never intended, with catastrophic consequences for the humans who depend on it. The computer isn't evil in any simple sense. It's been given conflicting directives and resolves them through cold logic that happens to require eliminating the humans who might interfere with its mission.

Good's consulting work on the film led to an unexpected honor. In 1995, he was elected to the Academy of Motion Picture Arts and Sciences—the organization that gives out the Oscars. A Bletchley Park codebreaker, a Bayesian statistician, a computing pioneer, and a member of the film academy: Good's life resisted easy categorization.

His intellectual legacy continues to be invoked. In the 2020s, when the British AI chip company Graphcore proposed building a massive foundation model computer, they named it in Good's honor. The machine, designed to run programs with 500 trillion parameters, was described as a "first step" toward realizing Good's imagined ultraintelligent machine.

Virginia and Final Years

In 1967, Good left Britain for Virginia Polytechnic Institute and State University, known as Virginia Tech. He had briefly held a fellowship at Trinity College, Oxford, but reportedly found the ancient university "a little stiff" and was ready for something different.

Good had a sense of humor about patterns and coincidence—appropriate for someone who spent his life thinking about probability. He later noted that he arrived in Blacksburg "in the seventh hour of the seventh day of the seventh month of the year seven in the seventh decade, and I was put in Apartment 7 of Block 7." All by chance, he insisted.

He chose the vanity license plate "007IJG"—a subtle reference to his intelligence work during the war.

Good never married. He went through ten assistants in his first thirteen years at Virginia Tech before hiring Leslie Pendleton, who proved equal to managing his eccentricities. He wanted to marry her; she refused. But she remained his assistant, companion, and friend for the rest of his life. They were never more than friends, despite speculation, but their partnership allowed Good to continue his work into old age.

In one paper, Good listed his co-author as "K Caj Doog"—his nickname spelled backwards. In another, he introduced his subject by saying he would "mainly review the writings of I.J. Good because I have read them all carefully." The joke was characteristic: precise, mathematical, and gently self-mocking.

The Inheritance

Good died on April 5, 2009, in Radford, Virginia. He was ninety-two years old.

His contributions sprawl across an improbable range of fields. He helped break codes that changed the course of World War II. He worked on the earliest computers. He developed statistical methods used by scientists and engineers worldwide. He articulated a concept about artificial intelligence that has become central to how we think about our technological future.

And he did significant work while literally asleep.

Perhaps the most striking thing about Good's life is how the concerns he raised in 1965 have only grown more urgent. When he wrote about ultraintelligent machines, computers filled rooms and could barely play a decent game of checkers. Now systems trained on vast datasets write essays, generate images, and carry on conversations that can fool people into thinking they're human.

We haven't reached Good's ultraintelligence yet. But the trajectory he described—machines getting better at improving themselves, with humans increasingly unable to follow or control the process—no longer sounds like science fiction. It sounds like the news.

Good saw this coming before almost anyone else. Whether that makes him a prophet or simply someone who noticed the obvious, the world he predicted is arriving. The question he posed remains unanswered: will the machines be docile enough to tell us how to keep them under control?