: New: Book Report: Humble Pi

This book talks about math errors and the consequences that follow. There are errors of engineering, software errors (dear to my heart), and plain old computation errors. Some of these get pretty interesting. E.g., until I read this book, I thought the designers of the Millennium Bridge must be pretty darned incompetent. When pedestrians walked across the newly-constructed Millennium Bridge, it started to shake itself apart. Engineers have known about the "breakstep bridge" problem for a long time; there was no excuse for this to happen to a modern bridge. Except except, as I learned from this book, the Millennium Bridge was shaking itself apart in a new, exciting way. It wasn't bouncing up and down as people stomped on it. Rather, it waggled from side to side as walkers shifted.

I found out about medical calculators, used for lives-are-at-stake calculations like drug dosages. If I haul out my phone calculator and type in 2  3 · 4 , it ignores the second dot and shows 2.34. But it's strange that I hit the · key twice—one of those was probably an accident. Maybe I meant to enter 2.34, maybe I meant 23.4. These fancy-pants medical calculators are more careful: they show a warning about the too-many-dots problem. (The book also discusses a problem that can arise from using too many fancy-pants complex error checkers: if the system gets too complex, that's just more opportunities for errors as layers of a system interfere with each other.)

I found out that the Spurious Correlations website is pretty funny and reminds us that plenty of unrelated things correlate with crime rates, cancer rates, whatever rates; and you should take any pundit's discovery of the true cause of whatever with a big grain of salt. I found out about a project Tommy Flowers worked on after his WWII codebreaking work: ERNIE, a random number generator that generated entropy from neon tubes. I learned… uhm, there's a lot of cool stuff in this book. Recommended; check it out.

Tags: book programming trust the machine