A Tale of Two Neumanns


I have been looking into the history of the first digital stored-program computer. This is either attributed to John von Neumann under the umbrella of the "von Neumann architecture" and/or to Alan Turing in pop culture, but this is not the whole story.

(Max Neumann. Credit Wikipedia, CC-BY-SA.)

Max Newman, whose surname at birth was Neumann (unrelated to John von Neumann), was extremely involved, if not directly responsible, for actually getting the first stored-program computer built. It was called the "Manchester Baby".

Going further into the archives, it appears as though Turing attended a lecture given by Newman at Cambridge in 1935 (Stanford Encyclopedia of Philosophy, "The Modern History of Computing"). According to that source, Turing was subsequently inspired to create his famous paper on Turing Machines when he heard Newman describe one of Hilbert's problems under a mechanistic interpretation.

Newman collaborated with Turing and John von Neumann at Manchester, but it was Newman who founded the Computing Machine Laboratory and recruited F.C. Williams and Tom Kilburn, who did the electrical engineering and construction aspects of the project.

My interest in this history is that it has parallels to formal artificial intelligence. To make an analogy: traditional machine learning is to analog computing as formal AI is to the stored-program computer. History may yet repeat in the sense that we will gain the ability to modularize, reprogram, and re-purpose models and AI solutions.

All analogies eventually fail somewhere, but the general sketch is correct. Analog computers and traditional machine learning share in common a monolithic, indecomposable form that must be rebuilt (retrained) for each purpose. By contrast, formal artificial intelligence is like the stored-program computer; digital, flexible, and reusable.

Think of all the benefits that we have gained by having computers that can be reused just by writing new programs. Likewise, it is hard to predict the full impact of having an AI paradigm shift where this kind of reuse is both efficient and practical.