(This post was originally published on the NKS Forum.)
Today (December 28, 2003) would have been John von Neumann’s 100th birthday—if he had not died at age 54 in 1957, and that gives me an excuse today to write a little about him. (I would have liked to spend longer on this, but December 28 only lasts so long, and I have many other things to do.)
I’ve been interested in von Neumann for many years—not least because his work touched on some of my most favorite topics. He is mentioned in 12 separate places in my book—second in number only to Alan Turing, who appears 19 times.
I always feel that one can appreciate peoples’ work better if one understands the people themselves better. And from talking to many people who knew him, I think I’ve gradually built up a decent picture of John von Neumann as a man.
He would have been fun to meet. He knew a lot, was very quick, always impressed people, and was lively, social and funny.
One video clip of him has survived. In 1955 he was on a television show called “Youth Wants to Know”, which today seems painfully hokey. Surrounded by teenage kids, he is introduced as a commissioner of the Atomic Energy Commission—which in those days was a big deal. He is asked about an exhibit of equipment. He says very seriously that it’s mostly radiation detectors. But then a twinkle comes into his eye, and he points to another item, and says deadpan “except this, which is a carrying case”. And that’s the end of the only video record of John von Neumann that exists.
Some scientists (such as myself) spend most of their lives pursuing their own grand programs, ultimately in a fairly isolated way. John von Neumann was instead someone who always liked to interact with the latest popular issues—and the people around them—and then contribute to them in his own characteristic way.
He worked hard, often on many projects at once, and always seemed to have fun. In retrospect, he chose most of his topics remarkably well. He studied each of them with a definite practical mathematical style. And partly by being the first person to try applying serious mathematical methods in various areas, he was able to make important and unique contributions.
But I’ve been told that he was never completely happy with his acheivements because he thought he missed some great discoveries. And indeed he was close to a remarkable number of important mathematics-related discoveries of the twentieth century: Godel’s theorem, Bell’s inequalities, information theory, Turing machines, computer languages—as well as my own more recent favorite core NKS discovery of complexity from simple rules.
But somehow he never quite made the conceptual shifts that were needed for any of these discoveries.
There were, I think, two basic reasons for this. First, he was so good at getting new results by the mathematical methods he knew that he was always going off to get more results, and never had a reason to pause and see whether some different conceptual framework should be considered. And second, he was not particularly one to buck the system: he liked the social milieu of science and always seemed to take both intellectual and other authority seriously.
By all reports, von Neumann was something of a prodigy, publishing his first paper (on zeros of polynomials) at the age of 19. By his early twenties, he was established as a promising young professional mathematician—working mainly in the then-popular fields of set theory and foundations of math. (One achievement was alternate axioms for set theory—see the NKS book, page 1155.)
Like many good mathematicians in Germany at the time, he worked on David Hilbert’s program for formalizing mathematics, and for example wrote papers aimed at finding a proof of consistency for the axioms of arithmetic. But he did not guess the deeper point that Kurt Godel discovered in 1931: that actually such a proof is fundamentally impossible. I’ve been told that von Neumann was always disappointed that he had missed Godel’s theorem. He certainly knew all the methods needed to establish it (and understood it remarkably quickly once he heard it from Godel). But somehow he did not have the brashness to disbelieve Hilbert, and go looking for a counterexample to Hilbert’s ideas.
In the mid-1920s formalization was all the rage in mathematics, and quantum mechanics was all the rage in physics. And in 1927 von Neumann set out to bring these together—by axiomatizing quantum mechanics. A fair bit of the formalism that von Neumann built has become the standard framework for any more mathematical exposition of quantum mechanics. But I must say that I have always thought that it gave too much of an air of mathematical definiteness to ideas (particularly about quantum measurement) that are in reality depend on all sorts of physical details. And indeed some of von Neumann’s specific axioms turned out to be too restrictive for ordinary quantum mechanics—obscuring for a number of years the phenomenon of entanglement, and later of criteria such as Bell’s inequalities.
But von Neumann’s work on quantum mechanics had a variety of fertile mathematical spinoffs, and particularly what are now called von Neumann algebras have recently become popular in mathematics and mathematical physics.
Interestingly, von Neumann’s approach to quantum mechanics was at first very much aligned with traditional calculus-based mathematics—investigating properties of Hilbert spaces, continuous operators, etc. But gradually it became more focused on discrete concepts, particularly early versions of “quantum logic”. In a sense von Neumann’s quantum logic ideas were an early attempt at defining a computational model of physics. But he did not pursue this, and did not go in the directions that have for example led to modern ideas of quantum computing.
By the 1930s von Neumann was publishing several papers a year, on a variety of popular topics in mainstream mathematics, often in collaboration with contemporaries of significant later reputation (Wigner, Koopman, Jordan, Veblen, Birkhoff, Kuratowski, Halmos, Chandrasekhar, etc.). Von Neumann’s work was unquestionably good and innovative, though very much in the flow of development of the mathematics of its time.
Despite von Neumann’s early interest in logic and the foundations of math, he (like most of the math community) moved away from this by the mid-1930s. In Cambridge and then in Princeton he encountered the young Alan Turing—even offering him a job as an assistant in 1938. But he apparently paid little attention to Turing’s classic 1936 paper on Turing machines and the concept of universal computation, writing in a recommendation letter on June 1, 1937 that “[Turing] has done good work on … theory of almost periodic functions and theory of continuous groups”.
As it did for many scientists, von Neumann’s work on the Manhattan Project appears to have broadened his horizons, and seems to have spurred his efforts to apply his mathematical prowess to problems of all sorts—not just in traditional mathematics. His pure mathematical colleagues seem to have viewed such activities as a peculiar and somewhat suspect hobby, but one that could generally be tolerated in view of his respectable mathematical credentials.
At the Institute for Advanced Study in Princeton, where von Neumann worked, there was strain, however, when he started a project to build an actual computer there. Indeed, even when I worked at the Institute in the early 1980s, there were still pained memories of the project. The pure mathematicians at the Institute had never been keen on it, and the story was that when von Neumann died, they had been happy to accept Thomas Watson of IBM’s offer to send a truck to take away all of von Neumann’s equipment. (Amusingly, the 6-inch on-off switch for the computer was left behind, bolted to the wall of the building, and has recently become a prized possession of a computer industry acquaintance of mine.)
I had some small interaction with von Neumann’s heritage at the Institute in 1982 when the then-director (Harry Woolf) was recruiting me. (Harry’s original concept was to get me to start a School of Computation at the Institute, to go along with the existing School of Natural Sciences and School of Mathematics. But for various reasons, this was not what happened.) I was concerned about intellectual property issues, having just had some difficulty with them at Caltech. Harry’s response—that he attributed to the chairman of their board of trustees was: “Look, von Neumann developed the computer here, but we insisted on giving it away; after that, why should we worry about any intellectual property rights?”. (The practical result was a letter disclaiming any rights to any intellectual property that I produced at the Institute.)
Among several of von Neumann’s interests outside of mainstream pure mathematics was his attempt to develop a mathematical theory of biology and life (see the NKS book, page 876). In the mid-1940s there had begun to be—particularly from wartime work on electronic control systems—quite a bit of discussion about analogies between “natural and artificial automata”, and “cybernetics”. And von Neumann decided to apply his mathematical methods to this. I’ve been told he was particularly impressed by the work of McCullough and Pitts on formal models of the analogy between brains and electronics (see the NKS book, page 1099). (There were undoubtedly other influences too: John McCarthy told me that around 1948 he visited von Neumann, and told him about applying information theory ideas to thinking about the brain as an automaton; von Neumann’s main response at the time was just “write it up!”.)
Von Neumann was in many ways a traditional mathematician, who (like Turing) believed he needed to turn to partial differential equations in describing natural systems. I’ve been told that at Los Alamos von Neumann was very taken with electrically stimulated jellyfish, which he appears to have viewed as doing some kind of continuous analog of the information processing of an electronic circuit. In any case, by about 1947, he had conceived the idea of using partial differential equations to model a kind of factory that could reproduce itself, like a living organism.
Von Neumann always seems to have been very taken with children, and I am told that it was in playing with an erector set owned by the son of his game theory collaborator Oskar Morgenstern that von Neumann realized that his self-reproducing factory could actually be built out of discrete robotic-like parts. (There was already something of a tradition of building computers out of Meccano—and indeed for example some of Hartree’s early articles on analog computers appeared in Meccano Magazine.)
An electrical engineer named Julian Bigelow, who worked on von Neumann’s IAS computer project, pointed out that 3D parts were not necesary, and that 2D would work just as well. (When I was at the Institute in the early 1980s Bigelow was still there, though unfortunately viewed as a slightly peculiar relic of von Neumann’s project.)
Stan Ulam told me that he had independently thought about making mathematical models of biology, but in any case, around 1951 he appears to have suggested to von Neumann that one should be able to use a simplified, essentially combinatorial model—based on something like the infinite matrices that Ulam had encountered in the so-called Polish Book of math problems to which he had contributed.
The result of all this was a model that was formally a two-dimensional cellular automaton. Systems equivalent to two-dimensional cellular automata were arising in several other contexts around the same time (see the NKS book, page 876). von Neumann seems to have viewed his version as a convenient framework in which to construct a mathematical system that could emulate engineered computer systems—especially the EDVAC on which von Neumann worked.
In the period 1952-3 von Neumann sketched an outline of a proof that it was possible for a formal system to support self reproduction. Whenever he needed a different kind of component (wire, oscillator, logic element, etc.) he just added it as a new state of his cellular automaton, with new rules. He ended up with a 29-state system, and a 200,000 cell configuration that could reproduce itself. (von Neumann himself did not complete the construction. This was done in the early 1960s by a former assistant of von Neumann’s named Arthur Burks, who had left the IAS computer project to concentrate on his interests in philosophy, though who maintains even today an interest in cellular automata.)
From the point of view of NKS, von Neumann’s system now seems almost grotesquely complicated. But von Neumann’s intuition told him that one could not expect a simpler system to show something as sophisticated and biological as self reproduction. What he said was that he thought that below a certain level of complexity, systems would always be “degenerative”, and always generate what amounts to behavior simpler than their rules. But then, from seeing the example of biology, and of systems like Turing machines, he believed that above some level, there should be an “explosive” increase in complexity, with systems able to generate other systems more complex than themselves. But he said that he thought the threshold for this would be systems with millions of parts.
Twenty-five years ago I might not have disagreed too strongly with that. And certainly for me it took several years of computer experimentation to understand that in fact it takes only very simple rules to produce even the most complex behavior. So I do not think it surprising—or unimpressive—that von Neumann failed to realize that simple rules were enough.
Of course, as one often realizes in retrospect, he did have some other clues. He knew about the idea of generating pseudorandom numbers from simple rules, even suggesting the “middle square method” (see NKS page 975.) He had the beginnings of the idea of doing computer experiments in areas like number theory. He analysed the first 2000 digits of pi and e, computed on the ENIAC, finding that they seemed random—though making no comment on it (see the NKS book, page 912). (He also looked at ContinuedFraction[2^(1/3)]; see the NKS book, page 914.)
I have asked many people who knew him why von Neumann never considered simpler rules. Marvin Minsky told me that he actually asked von Neumann about this directly, but that von Neumann had been somewhat confused by the question. It would have been much more Ulam’s style than von Neumann’s to have come up with simpler rules, and Ulam indeed did try making a one-dimensional analog of 2D cellular automata, but came up not with 1D cellular automata, but with a curious number-theoretical system (see the NKS book, page 908).
In the last ten years of his life, von Neumann got involved in an impressive array of issues. Some of his colleagues seem to have felt that he spent too little time on each one, but still his contributions were usually substantial—sometimes directly in terms of content, and usually at least in terms of lending his credibility to emerging areas.
He made mistakes, of course. He thought that each logical step in computation would necessarily dissipate a certain amount of heat, whereas in fact reversible computation is in principle possible. He thought that the unreliability of components would be a major issue in building large computer systems; he apparently did not have an idea like error-correcting codes. He is reputed to have said that no computer program would ever be more than a few thousand lines long. He was probably thinking about proofs of theorems—but did not think about subroutines, the analog of lemmas.
Von Neumann was a great believer in the efficacy of mathematical methods and models, perhaps implemented by computers. In 1950 he was optimistic that accurate numerical weather forecasting would soon be possible (see the NKS book page 1132). In addition, he believed that with methods like game theory it should be possible to understand much of economics and other forms of human behavior (see the NKS book page 1135).
Von Neumann was always quite a believer in using the latest methods and tools (I’m sure he would have been a big Mathematica user today). He typically worked directly with one or two collaborators, sometimes peers, sometimes assistants, though he maintained contact with a large network of scientists. (A typical communication was a letter he wrote to Alan Turing in 1949, in which he asks “What are the problems on which you are working now, and what is your program for the immediate future?”.) In his later years he often operated as a distinguished consultant, brought in by the government, or other large organizations. His work was then often presented as a report, that was accorded particular weight because of his distinguished consultant status. (It was also often a good and clear piece of work.) He was often viewed a little ambivalently as an outsider in the fields he entered—positively because he brought his distinction to the field, negatively because he was not in the clique of experts in the field.
Particularly in the early 1950s, von Neumann became deeply involved in military consulting, and indeed I wonder how much of the intellectual style of cold war U.S. military strategic thinking actually originated with him. He seems to have been quite flattered that he called upon to do this consulting, and he certainly treated the government with considerably more respect than many other scientists of his day. Except sometimes in his exuberence to demonstrate his mathematical and calculational prowess, he seems to have always been quite mature and diplomatic. The transcript of his testimony at the Oppenheimer security hearing certainly for example bears this out.
Nevertheless, von Neumann’s military consulting involvements left some factions quite negative about him. It’s sometimes said, for example, that von Neumann might have been the model for the sinister Dr. Strangelove character in Stanley Kubrick’s movie of that name (and indeed von Neumann was in a wheelchair for the last year of his life). And vague negative feelings about von Neumann surface for example in a typical statement I heard recently from a science historian of the period—that “somehow I don’t like von Neumann, though I can’t remember exactly why”.
I recently met von Neumann’s only child—his daughter Marina, who herself has had a distinguished career, mostly at General Motors. She reinforced my impression that until his unpleasant final illness, John von Neumann was a happy and energetic man, working long hours on mathematical topics, and always having fun. She told me that when he died, he left a box that he directed should be opened fifty years after his death. What does it contain? His last sober predictions of a future we have now seen? Or a joke—like a funny party hat of the type he liked to wear? It’ll be most interesting in 2007 to find out.
Below is an addition from February 8, 2007.
John von Neumann’s box
At the end of my post about the 100th anniversary of John von Neumann’s birth, I mentioned that his daughter had told me about a box to be opened on the 50th anniversary of his death. That anniversary is today.
And being reminded of this last week, I sent mail to his daughter to ask what had become of the box.
Disappointingly, she responded that: “The Big Box opening turned out to be a Big Bust”. Apparently she and her children and their grandchildren had all assembled … only to discover that it was all a big mistake: the box was actually not John von Neumann’s at all!
Perhaps it for the best. Von Neumann’s daughter sent me a piece she wrote about him, pointing out the unpredictability (irreducibility?) of technological and other change—and expressing concern that our species might wipe itself out by the year 1980.
Fortunately that of course hasn’t happened. And it was recently suggested to me that perhaps The Box might contain some Cold-War-inspired plan for a Dr.-Strangelove-like Doomsday Machine. So—”von Neumann machine” self-replicators notwithstanding—it’s perhaps just as well that in the end there was no box.