the end of the story of von Neumann and the machines. Like Turing, von Neumann was fascinated by the idea of artificial intelligence, although he had a different perspective on the rise of the robot. But unlike Turing, he lived long enough (just) to begin to flesh out those ideas.
There were two parts to von Neumann's later work. He was interested in the way that a complex system like the brain can operate effectively even though it is made up of fallible individual components, neurons. In the early computers (and many even today), if one component, such as a vacuum tube, failed the whole thing would grind to a halt. Yet in the humanbrain it is possible for the âhardwareâ to suffer massive injuries and continue to function adequately, if not quite in the same way as before. And he was also interested in the problem of reproduction. Jumping off from Turing's idea of a computer that could mimic the behavior of any other computer, 16 he suggested, first, that there ought to be machines that could make copies of themselves and, secondly, that there could be a kind of universal replicating machine that could make copies of itself and also of any other machine. Both kinds of mimic, or copying machines, come under the general heading âautomata.â.
Von Neumann's interests in working out how workable devices can be made from parts prone to malfunction, and in how complex a system would have to be in order to reproduce itself, both began to grow in 1947. This was partly because he was moving on from the development of computers like the one then being built at the IAS and other offspring of EDVAC, but also because he became involved in the pressing problem for the US air force in the early 1950s of how to develop missiles controlled by âautomataâ that would function perfectly, if only during the brief flight time of the rocket.
Von Neumann came up with two theoretical solutions to the problem of building near-infallible computing machines out of fallible, but reasonably accurate, components. The first is to set up each component in triplicate, with a means to compare automatically the outputs of the three subunits. If all three results, or any two results, agree, the computation proceeds to the next step, but if none of the subunits agrees with any other, the computation stops. This âmajority votingâ system works pretty well if the chance of any individual subunit making a mistake is small enough. It is even better ifthe number of âvotersâ for each step in the calculation is increased to five, seven, or even more. But this has to be done for every step of the computation (not just every âneuronâ), vastly (indeed, exponentially) increasing the amount of material required. The second technique involves replacing single lines for input and output by bundles containing large numbers of linesâso-called multiplexing. The data bit (say, 1) from the bundle would only be accepted if a certain proportion of the lines agreed that it was correct. This involves complications which I will not go into here; 17 the important point is that although neither technique is practicable, von Neumann proved that it is possible to build reliable machines, even brains, from unreliable components.
As early as 1948, von Neumann was lecturing on the problem of reproduction to a small group at Princeton. 18 The biological aspects of the puzzle were very much in the air at the time, with several teams of researchers looking for the mechanism by which genetic material is copied and passed from one generation to the next; it would not be until 1952 that the structure of DNA was determined. And it is worth remembering that von Neumann trained as a chemical engineer, so he understood the subtleties of complex chemical interactions. So it is no surprise that von Neumann says that the copying mechanism performs âthe fundamental act of reproduction, the duplication of the genetic material.â The
There were two parts to von Neumann's later work. He was interested in the way that a complex system like the brain can operate effectively even though it is made up of fallible individual components, neurons. In the early computers (and many even today), if one component, such as a vacuum tube, failed the whole thing would grind to a halt. Yet in the humanbrain it is possible for the âhardwareâ to suffer massive injuries and continue to function adequately, if not quite in the same way as before. And he was also interested in the problem of reproduction. Jumping off from Turing's idea of a computer that could mimic the behavior of any other computer, 16 he suggested, first, that there ought to be machines that could make copies of themselves and, secondly, that there could be a kind of universal replicating machine that could make copies of itself and also of any other machine. Both kinds of mimic, or copying machines, come under the general heading âautomata.â.
Von Neumann's interests in working out how workable devices can be made from parts prone to malfunction, and in how complex a system would have to be in order to reproduce itself, both began to grow in 1947. This was partly because he was moving on from the development of computers like the one then being built at the IAS and other offspring of EDVAC, but also because he became involved in the pressing problem for the US air force in the early 1950s of how to develop missiles controlled by âautomataâ that would function perfectly, if only during the brief flight time of the rocket.
Von Neumann came up with two theoretical solutions to the problem of building near-infallible computing machines out of fallible, but reasonably accurate, components. The first is to set up each component in triplicate, with a means to compare automatically the outputs of the three subunits. If all three results, or any two results, agree, the computation proceeds to the next step, but if none of the subunits agrees with any other, the computation stops. This âmajority votingâ system works pretty well if the chance of any individual subunit making a mistake is small enough. It is even better ifthe number of âvotersâ for each step in the calculation is increased to five, seven, or even more. But this has to be done for every step of the computation (not just every âneuronâ), vastly (indeed, exponentially) increasing the amount of material required. The second technique involves replacing single lines for input and output by bundles containing large numbers of linesâso-called multiplexing. The data bit (say, 1) from the bundle would only be accepted if a certain proportion of the lines agreed that it was correct. This involves complications which I will not go into here; 17 the important point is that although neither technique is practicable, von Neumann proved that it is possible to build reliable machines, even brains, from unreliable components.
As early as 1948, von Neumann was lecturing on the problem of reproduction to a small group at Princeton. 18 The biological aspects of the puzzle were very much in the air at the time, with several teams of researchers looking for the mechanism by which genetic material is copied and passed from one generation to the next; it would not be until 1952 that the structure of DNA was determined. And it is worth remembering that von Neumann trained as a chemical engineer, so he understood the subtleties of complex chemical interactions. So it is no surprise that von Neumann says that the copying mechanism performs âthe fundamental act of reproduction, the duplication of the genetic material.â The
Similar Books
