Exclusive: John von Neumann’s letters to Norbert Wiener on computational machinery (1945)

Continued from part 1, Letter 2

… of which we can already foresee the character, the caliber, and the duration? And are the latter two not excessive and impractical?

I feel that the answer to these questions can already be given, and that it need not be unfavorable. Specifically:

I am talking of molecular weights 10^7 to 10^8. The major proteins have molecular weights 10^4 to 10^5. (the lowest one known actually appears to be only about 7,000) and the determination of their exact structure may not be hopeless. This is controversial, but some good authorities like Langmuir and Dorothy Wrinch consider it promising. Langmuir assert that a 2-4 year efforts with strong financial backing should break the back of the problem. His idea of an attack is: Very high precision K-ray analysis, Fourier transformation with very massive fast computing, in combination with various chemical substitution techniques to vary the X-ray pattern.

I realize that this is in itself a big order, and that it is still by a factor 10^3 off our goal but it would probably be more than half the difficulty.

In addition there is no telling what really advanced electron-microscopic techniques will do. In fact, I suspect that the main possibilities may well lie in that direction. The best (magnetic) electron-microscope resolutions at present are a little better than 10 Å = 1 mμ*. With 4 x 10^6 atoms in a volume of 3.7 x 10^-17 cm^3 = 4 x 10^3 mμ^3, the average atomic volume is 10^-2 mμ^3, and hence the average atomic distance about 1/5 mμ. Hence the 1 mμ resolution is inadequate – but not very far from what might be adequate. A resolution that is improved by a factor of 10- 20 might do. It is dubious whether electron lenses can be improved to this extent. On the other hand, the proton microscope need not be more than 2-4 years in the future, and it would certainly overcome these difficulties.

Besides, all these developments might be pushed and accelerated. Of course, everybody knows what a 1 to 1/2 Å resolution would mean: One could “look” at an H atom, and with a little more, say 1/5 Å, one could “see” the Schroedinger-charge-cloud of the orbital electrons! But the physiological implications are even more extraordinary, and they should receive a great deal of emphasis in the immediate future.

At any rate, I think that we could do these things

Study the main types of evidence: Physiology of viruses and bacteriophages, and all that is known about the gene-enzyme relationship. (Genes are probably much like viruses and phages, except that all the evidence concerning them is indirect, and that we can neither isolate them nor multiply them at will.)

Try to learn a reasonable amount about the present state of knowledge and opinion concerning protein structure.

Study the methods of organic-chemical structure determination by X-ray analysis and Fourier-analysis with their necessary complement of chemical manipulations, study the principles and methods of electron-microscopy, both in the (direction of electron optics and in the direction of object-manipulation).

Try to get oriented as to the possibilities of proton-microscopy.

Finally: Compile for our common use two lists: (1) relevant publications with the main emphasis for the immediate future, considering our lack of education on books and survey articles. (2) Persons from whom we might learn most about the state of affairs and the outlook in these fields.

I did think a good deal about self-reproductive mechanisms. I can formulate the problem rigourously, in about the style in which Turing did it for his mechanisms.

I can show that they exist in this system of concepts. I think that I understand some of the main principles that are involved. I want to fill in the details and to write up these considerations in the course of the next two months. I hope to learn various things in the course of this literary exercise, in particular, the number of components required for self-reproduction. My (rather uninformed) guess is in the high ten thousands or in the hundred thousands, but this is most unsafe. Besides, I am thinking in terms of components based on several rather arbitrary choices. At any rate, it will be necessary to produce a complete write-up before much discussing is possible.

Certain traits of the gene-enzyme relationship, of the behavior of some mutants, as well as some other phenomena, seem to emphasize some variants of self-reproductivity, which one would be led to investigate on purely combinatorial grounds as well. E.g.: Self-reproductivity may be symbolized by the schema A→A. What about schemata like A → B→C→ A, or A → B → C→C, or A→ B-C→D→E→C, etc.?

This is as far as my ideas go at this moment. I hope you will not misinterpret the anti-neurological tirade at the beginning of this letter. Of course I am greatly interested in that approach and I have the greatest respect for the important results that have been obtained in that field, and in our border area with it. I certainly hope that these efforts will continue. I wanted to point out, however, that I felt that the decisive “break” was more likely to come in another theater. I was trying to formulate and to systematize my motives for believing this, and the simplest literary mode to do this is the controversial one. I hope therefore that I have not given you a false impression of the spirit in which I am starting a “controversy”. I am most anxious to have your reaction to these suggestions. I feel an intense need that we discuss the subject extensively with each other. Hoping that this letter has not been unbearable just by its sheer length, and hoping to hear from you and to see you again soon, I am, with the best regards,

Yours, as ever

John von Neumann

Original in Von Neumann Papers, Library of Congress, language of original: English

*He seems to use this notation to refer to a milli-micron, i.e. 1/1000 of a micron. Quirky by modern standards but then again this was 1946.