John von Neumann

The Last Great Mind

Born: 28 December 1903, Budapest, Austria-Hungary Died: 8 February 1957, Washington, D.C., USA Age at death: 53

There is a story, probably apocryphal but persistently told, about John von Neumann at a dinner party. A fellow guest, attempting to impress him, mentions that he has memorised the first hundred digits of pi. Von Neumann listens politely. Then, without apparent effort, he recites the first two hundred and fifty.

The story is almost certainly embellished. What is not embellished is that von Neumann could, and did, regularly perform feats of mental calculation and recall that left professional mathematicians unsettled. He could multiply six-digit numbers in his head faster than colleagues with paper and pencil. He could memorise a column of numbers from a phone book after a single glance and recite them hours later. He once reportedly read an entire book during a brief wait in a reception room and was able to quote passages verbatim weeks afterward.

These abilities were real, documented, and not the most remarkable thing about him. The most remarkable thing about him was what he did with them. In a career spanning three decades, John von Neumann made foundational contributions to pure mathematics, quantum mechanics, game theory, economics, nuclear weapons design, meteorology, and the theory and practice of computing. He was not merely a calculating prodigy who solved problems faster than other people. He was someone who saw, with unusual clarity and unusual depth, the mathematical structure underlying phenomena that had not previously seemed mathematical at all.

The computer architecture named after him — the von Neumann architecture — runs every conventional computer built since 1945. The economic framework he co-invented — game theory — is the standard model of strategic interaction in economics, political science, and evolutionary biology. The mathematical formalism he developed for quantum mechanics is still the one used in physics textbooks. He did all of this while also helping to design the implosion lens for the plutonium bomb that destroyed Nagasaki, serving on so many government advisory committees that his colleagues joked he was simultaneously in several cities, and throwing dinner parties in Princeton that were legendarily good.

He died at fifty-three, of cancer, probably caused by radiation exposure during his nuclear work. His last years were spent being wheeled into Pentagon briefings, his mind still working at full speed while his body failed. He is buried in Princeton. His headstone says, simply, “John von Neumann.”

Budapest

Neumann János Lajos was born in Budapest in December 1903, the eldest son of Max Neumann, a prosperous banker, and Margaret Kann, daughter of a wealthy merchant family. Budapest at the turn of the century was, for a specific class of educated, secular, assimilated Jewish families, one of the most intellectually vibrant cities in Europe. The city was producing mathematicians, physicists, and scientists at a rate disproportionate to its size that has puzzled historians ever since — Eugene Wigner, Leo Szilard, Edward Teller, and Dennis Gabor were all contemporaries of von Neumann from similar Budapest backgrounds.

His father Max recognised his son’s abilities early and, with a banker’s pragmatism, arranged for him to be educated not just in mathematics but in practical and humanistic subjects as well. Von Neumann grew up speaking Hungarian, German, French, and English fluently, with working knowledge of Latin and Greek and a lifelong love of classical history. He would later astonish colleagues by quoting long passages of Byzantine history from memory, apparently for pleasure rather than utility.

He entered the University of Budapest at seventeen — formally enrolled, rarely attending — while simultaneously studying chemical engineering in Zurich at his father’s insistence. The compromise, which allowed him to pursue the mathematics that consumed him while maintaining the pretence of practical education, resulted in his receiving a degree in chemical engineering from ETH Zurich and a doctorate in mathematics from Budapest simultaneously, in 1926, at age twenty-two. His doctoral thesis was on the axiomatisation of set theory. It resolved an outstanding problem left by previous axiom systems and is still used.

He was appointed as a Privatdozent at Berlin, then at Hamburg, and arrived at Princeton in 1930, initially as a visiting lecturer. When the Institute for Advanced Study was founded in 1933, he was among its first six permanent faculty members, alongside Albert Einstein. He was twenty-nine. He never returned to Europe to work. Hungary was becoming, and would become, an extremely bad place to be a secular Jewish intellectual.

The Mathematics

Von Neumann’s mathematical output before he turned his attention to computing is difficult to summarise without simply listing fields. A partial inventory: he proved the minimax theorem in game theory (1928), laying the foundation for everything that followed in that field. He developed the mathematical formalism of quantum mechanics, replacing the competing formalisms of Heisenberg and Schrödinger with a unified framework based on Hilbert spaces and operators (1932). He made fundamental contributions to ergodic theory, to the theory of operator algebras (now called von Neumann algebras), to continuous geometry, and to the foundations of mathematics.

The quantum mechanics work deserves particular mention because it had a direct influence on his later thinking about computing and, ultimately, about minds. The mathematical formalism he developed treated quantum states as vectors in an abstract space, quantum measurements as operators on that space, and the outcomes of measurements as eigenvalues. This is the formalism still used in quantum computing. More relevantly: thinking carefully about measurement in quantum mechanics — about the relationship between the mathematical description of a system and the act of observing it — gave von Neumann a particular sensitivity to questions about the relationship between formal representations and the physical processes they described. These questions would become central to his work on computers.

In game theory, the minimax theorem established that in any finite two-player zero-sum game, there exists a pair of mixed strategies — strategies that assign probabilities to possible moves — such that neither player can improve their expected outcome by changing their strategy unilaterally. This is the Nash equilibrium for two-player zero-sum games, proved twenty years before John Nash generalised it. It meant, for the first time, that strategic interaction could be analysed mathematically — that there were provably optimal strategies in well-defined competitive situations. The implications for economics, military strategy, and eventually AI were enormous.

Los Alamos

Von Neumann’s involvement with nuclear weapons began in 1943, when he was recruited to the Manhattan Project at Los Alamos. His specific contribution was to the implosion design for the plutonium bomb — the device that would be tested at Trinity in July 1945 and dropped on Nagasaki in August.

The problem with plutonium, compared to uranium, is that it fissions too easily: a critical mass of plutonium assembled too slowly will begin a chain reaction prematurely, blowing itself apart before most of the material has fissioned. The uranium bomb worked by shooting one subcritical piece into another fast enough to avoid this — the gun-assembly design used at Hiroshima. This method was too slow for plutonium.

The solution was implosion: surround a subcritical sphere of plutonium with precisely shaped explosive lenses that, when detonated simultaneously, compress the plutonium inward symmetrically, increasing its density to the point of criticality faster than premature fission can occur. The key word is simultaneously — the explosive lenses had to detonate within microseconds of each other, and they had to be shaped to produce a perfectly symmetrical compression wave.

Calculating the required shapes analytically was impossible. The equations governing the implosion were nonlinear partial differential equations with no closed-form solution. The only way to solve them was numerically — step by step, approximating the solution over thousands of iterations. Von Neumann organised teams of human computers at Los Alamos doing exactly this, using mechanical calculators, working in shifts. The calculations took weeks. He understood, watching this process, that what was needed was a machine that could do the same calculations in hours.

This practical need — the need to do large-scale numerical calculations faster than human computers could do them — was the immediate impetus for his involvement with electronic computing.

The Architecture

In 1944, von Neumann met Herman Goldstine, a mathematician working on ENIAC — the Electronic Numerical Integrator and Computer being built at the University of Pennsylvania. ENIAC was the first general-purpose electronic computer to be completed, though it was not, in the modern sense, a stored-program machine: it was programmed by physically reconfiguring its wiring, a process that took days.

Von Neumann joined the project and immediately began thinking about how to do it better. In June 1945 he circulated a document — now famous, then unofficial — titled “First Draft of a Report on the EDVAC,” describing the architecture for a new machine. The document was written quickly, as a draft for internal discussion, and bore only von Neumann’s name as author, which obscured the contributions of Eckert, Mauchly, and others at Penn. This attribution controversy generated bitterness that lasted decades.

The architecture described in the document is what we now call the von Neumann architecture. Its key features are:

A processing unit that performs arithmetic and logic operations. A control unit that interprets instructions and directs the processor. Memory that stores both data and instructions in the same address space — the stored-program concept. Input and output devices. And crucially: the instructions themselves are stored in memory in the same form as data, and can therefore be modified by the program as it runs.

That last point — the stored program — was the revolutionary idea. Previous machines had their programs hard-wired or required manual reconfiguration. By storing the program in memory as data, you made the machine infinitely reprogrammable without physical modification. You could write a program that wrote other programs. You could write an operating system. You could, eventually, write a compiler, an interpreter, a neural network trainer.

Every conventional computer built since 1945 — every desktop, laptop, smartphone, server, and embedded processor — uses a variant of this architecture. The fetch-decode-execute cycle that runs your code right now is the cycle von Neumann described in 1945.

He then went to the Institute for Advanced Study in Princeton and built one: the IAS machine, completed in 1952, which became the model for a generation of subsequent computers. The design was published freely, deliberately, against the advice of people who thought it could be commercialised. Von Neumann wanted it to spread.

Self-Reproducing Automata

While building the IAS machine, von Neumann was already thinking about something beyond it. The question that interested him was not just whether machines could compute, but whether they could reproduce — whether a machine could build a copy of itself, and whether that copy could in turn build further copies.

This is not primarily a practical engineering question. It is a theoretical question about the relationship between complexity and reproduction, between information and matter. Von Neumann was approaching it from the direction of his work in quantum mechanics and logic: he wanted to understand, at a formal level, what the minimum requirements were for a machine to be capable of self-replication.

His answer, developed through the late 1940s and published posthumously, was the theory of cellular automata and self-reproducing automata. He designed, abstractly, a machine capable of self-reproduction: a universal constructor that, given a description of any machine including itself, could build that machine. The key insight was that the description had to be both interpreted (to guide construction) and copied (to pass on to the new machine) — a distinction that maps precisely onto the relationship between DNA and proteins that Watson and Crick would describe in 1953, apparently without knowledge of von Neumann’s work.

This is the part of von Neumann’s legacy that is least well known outside theoretical computer science but most directly relevant to the deepest questions about artificial intelligence. He was asking: what is the formal structure that allows a physical system to reproduce its own complexity? What is the relationship between the description of a system and the system itself? These questions underlie modern thinking about artificial life, about evolutionary computation, and about the possibility of genuinely self-improving AI systems.

John Conway’s Game of Life, published in 1970, is a simplified version of von Neumann’s cellular automaton framework. The fact that it can produce self-replicating patterns is a demonstration, in miniature, of what von Neumann proved in general.

The Government Work

Von Neumann spent an enormous fraction of his time from 1940 until his death on government advisory work. He was a member of or consultant to the Atomic Energy Commission, the Armed Forces Special Weapons Project, the CIA, the Air Force Scientific Advisory Board, the Army Ballistic Research Laboratory, and numerous other bodies. He consulted on the design of hydrogen bombs. He consulted on the strategy of nuclear deterrence. He was involved in early thinking about intercontinental ballistic missiles.

His political views were hawkish, by the standards of his scientific colleagues — more hawkish than most, in a community that was already debating the ethics of the weapons they had built. He favoured a preventive war against the Soviet Union before it developed its own nuclear capability. He is reported to have said, in 1950, “If you say why not bomb them tomorrow, I say why not today? If you say today at 5 o’clock, I say why not one o’clock?” This remark is widely quoted and widely disputed; whether he said it in earnest or as a provocation is unclear.

What is clear is that he believed, with apparent sincerity, that the Soviet Union represented an existential threat that required the United States to act from strength. These views put him at odds with colleagues like Einstein and Robert Oppenheimer, whose security clearance was revoked in 1954 while von Neumann served on the reviewing committee. He did not intervene to defend Oppenheimer. This is the part of his legacy that his admirers tend to pass over quickly.

The Decline

In 1955, von Neumann was diagnosed with bone cancer, almost certainly caused by radiation exposure — possibly from witnessing nuclear tests in the Pacific, possibly from his work at Los Alamos, possibly both. The diagnosis was devastating. He was fifty-one, at the height of his influence, with more work he wanted to do.

He continued working as long as he could. He was wheeled into Pentagon briefings in a wheelchair, his mind still sharp while his body deteriorated. He dictated, in his final months, a set of lectures on the relationship between the brain and the computer — published posthumously as The Computer and the Brain — in which he attempted to compare the architecture of biological neural processing with the architecture of the machines he had helped build. The comparison is careful and modest; he was more interested in identifying the differences than in claiming that brains were computers or computers were brains.

He converted to Catholicism shortly before his death, apparently at the urging of a priest and with some uncertainty about his own sincerity. His friends and colleagues were divided about how to interpret this. He died on 8 February 1957 in Walter Reed Army Medical Center in Washington, with military security personnel present at his bedside to prevent him from saying anything classified while delirious.

He was fifty-three years old.

The Legacy

Von Neumann’s direct legacy to computing is the architecture that bears his name — the stored-program design that has run every computer for eighty years. His legacy to AI is less direct but no less fundamental.

The stored-program concept made it possible to write programs that manipulate other programs — compilers, interpreters, and eventually neural network trainers. The cellular automaton work established the theoretical framework for thinking about how complex behaviour emerges from simple rules, which is foundational to everything from evolutionary computation to the study of emergent intelligence. The game theory work established the mathematical framework for reasoning about strategic agents — agents with goals, who make decisions, whose actions affect one another — which is central to reinforcement learning and multi-agent AI systems.

And the question he was working toward at the end — the relationship between the architecture of computation and the architecture of thought, between the machine and the brain — is still the central question of AI research. He did not answer it. He died before he could. But The Computer and the Brain, incomplete and dictated through pain in his final months, is still one of the most careful and honest framings of that question ever written.

The scientists who came after him — McCarthy, Minsky, Hinton, and the rest — worked in a world his architecture had built, with mathematical tools he had largely forged, asking questions he had helped to formulate. He was not the founder of AI. He was something more fundamental: he built the ground it stands on.

Key Works & Further Reading

Primary sources:

“First Draft of a Report on the EDVAC.” Moore School of Electrical Engineering, University of Pennsylvania, June 1945.
Theory of Games and Economic Behavior — John von Neumann and Oskar Morgenstern (1944).
The Computer and the Brain — John von Neumann, published posthumously (1958).
“Theory of Self-Reproducing Automata” — edited and completed by Arthur Burks, published posthumously (1966).

Recommended reading:

John von Neumann — Norman Macrae (1992). The standard biography. Thorough on the mathematics and the politics.
Turing’s Cathedral — George Dyson (2012). The definitive account of the IAS machine and the Princeton computing project. Richly detailed.
The Man from the Future — Ananyo Bhattacharya (2021). A more recent biography, excellent on the breadth of von Neumann’s contributions and more willing to engage with his political legacy.
Prisoner’s Dilemma — William Poundstone (1992). On game theory and the Cold War strategic thinking von Neumann influenced. Essential context.