A Century Before AI there was Walter Shewhart

Welcome back to WISDOM WEDNESDAY. This week we’ll start with the beginning of statistical control and then fast forward to the current electronics environment surrounding AI chips. Wisdom Wednesday is sponsored by Micron Corporation that manufactures precision electronics at a Micron level.

Walter A. Shewhart

To use current linguistics in an older context “Walter Shewhart is the man.” That is to say he was the man who brought “quality control” into the American lexicon under the auspices of statistical methods. Walter A. Shewhart was born in 1891 and began his work on quality control at Western Electric Company in 1918. Western Electric was the manufacturing arm of Bell Systems which included American Telephone and Telegraph Company (AT&T, founded in 1885) and then in 1925 he wound up at Bell Telephone Laboratories. As is often the case with pioneers his work was not well appreciated at first by Bell Systems and American industry in general. It was, however, accepted by the agricultural sector of the economy as witnessed by the invitation extended to Shewhart from the federal organization known as the Graduate School of the Department of Agriculture. Accepting the invitation he delivered a series of four lectures on statistical method from the viewpoint of quality control. That effort stemmed from his seminal work “Economic Control of Quality of Manufactured Product” as published by D. Van Nostrand Company in 1931.

Shewhart studied subjects such as the deviations in physical characteristics of manufactured telephone poles and the significance of impurities in the insulating capacity of materials. Shewhart analyzed and measured the items of his studies then synthesized their meaning and reported his results and conclusions. W. Edwards Deming expanded upon Shewhart’s work and introduced it to the world. In his foreword to Shewhart’s collection of lectures to the Department of Agriculture (“Statistical Method from the Viewpoint of Quality Control”, Dover Publications,1939) Deming states the following: “There is no true value of anything. There is, instead, a figure that is produced by application of a master or ideal method of counting or of measurement.” Deming knew that because Shewhart showed it to be true.

Jumping Backward and Forward

Jumping forward a century the influence of statistical methods in quality control is evident in the tools and methods that are Six Sigma.  The words “six sigma” indicate the range of specification limits within a normal statistical distribution that extend six standard deviations (sigma) away from the process average. That, in a nutshell, is statistical methods applied to quality control. Thank you, Walter.

As best as has been surmised, quality control has always been a part of human existence. As an example, consider the following vignette of how quality has ensured the continuation of “products” over millennia. There is an ivory figurine that was found in the Stadel Cave of Germany which is dated to 32,000 years ago (as far as the current ideal method of measurement allows). It is a lion-person with a mostly human body and a leonine head. It is constructed well enough to be absolutely recognizable as a lion-person by anyone familiar with the two concepts. And a handsome lion-person at that. Zoom forward to the present day and a very similar depiction of a lion-person adorns many of the vehicles made by Peugeot SA. The Peugeot Lion trademark was incepted if not created in the village of Valentigney in eastern France just 200 miles from the Stadel Cave. [This comparison is made on pages 23 and 29 of “Sapiens” by Yuval Noah Harari, HarperCollins Publishers, 2015]. One might think twice before making the case that quality does not have an effect on the longevity of concepts and “products.”

Phones, Drones and GPUs

In this current set of globally connected societies there is no other category of devices that make manifest the necessity of quality than do our smart phones. This can be seen and heard from our normal position facing the outside of these devices. How could one possibly watch a YouTube video using a QR code from an Amazon package delivered next-day from a manufacturer half-way around the world without quality. But you can go even deeper; though I’m not suggesting you do this. If you were to take your phone apart, you would encounter a myriad of electronic components in addition to the amazing screen, camera, microphone, and speaker. None of these parts and components would be possible without an exceptional focus on the economic control of the quality of those manufactured products. Thanks again Walter.

Drones are also built and guided with quality and precision that allows for a host of highly useful and productive activities and not just waging war. Crop fields, home values, remote transport lines, storage tanks and many more items can be monitored because of the quality that is built into every drone. Not to mention all the R&D that went into drone development that was supported by statistical control of variables using high-quality equipment.

Now to Nvidia and the AI “boom.” A typical iPhone contains between one thousand and two thousand electronic components. An Nvidia A100 GPU contains at least an order of magnitude or more components. And that is not even counting the number of transistors on board any of the integrated circuits. Even small deviations in the functioning properties of precise components, even something as mundane as the resistance of a resistor, could potentially ruin the functionality of the entire A100 assembly. We could list many of the steps involved in the production of any one of the hundreds of different types of components involved in these circuits (though we wouldn’t know how to do many of them). All of them must be under tight control in order to manufacture quality product with minimum errors. In fact, Six Sigma would specify that no more than 3 or 4 errors be made for each million attempts at properly completing a step in the production process. Oh Walter, what have you done?

That’s it for this week. We look forward to sharing knowledge with you again next week. And you can always share information with us at MICRON CORPORATION by contacting info@microncorp.com.