Emeritus Professor of Computer Science, Princeton
The Discrete Charm of the Machine
Hillsboro 621.382 STEIGLITZ 2019
A well written book by Princeton computer science professor emeritus Ken Steiglitz.
Recommended; Many technologists know most of what Steiglitz writes about, but his goal is teaching the non-technical how to think about digital and analog information. We often need to do the same, and his approach is useful.
He discusses Turing machines, P != NP, Bell's inequality, and the brain. He is frank about what we don't know yet, and what we may never know.
Unlike many computer science writers, Steiglitz seems realistic about the prospects for AI, near and longer term. For now, we should be thrilled if we can match the power of a bumblebee brain, 10ppm of a human brain, but vastly more complex than existing computers. This is an NP (non-polynomial) problem; the analysis grows exponentially (or worse!) with scale of the system.
Steiglitz is somewhat more optimistic than I am about continued silicon scaling, but not nearly as optimistic as most AI zealots (and companies touting breakthroughs to pump their stock prices). But then, Steiglitz does not live downwind of Intel D1X, and can't smell the despair and desperation stemming from the damage that EUV (extreme ultraviolet) 13 nm photons (92 eV!) do to equipment and masks and transistors. A bridge too far?
I read the public library's copy of this book; if I wasn't downsizing, I would buy my own copy, to loan to others.
Steiglitz's personal website is sparse, but includes an amusing transcription of a 1915 lecture by Irving Langmuir. the origin of the term "pathological science" (a subset of "motivated reasoning"; in both phrases the adjective describes how the noun is invalidated).
People make mistakes, and are stubborn about correcting them if the "results" are spectacular, or conform to existing prejudices. I see a lot of that in politics, mass media, and discussions on mailing lists.
I rank Richard Feynman above Albert Einstein in my pantheon of great physicists because Feynman paid as much attention to error as he did to "TRUTH!". For some, science is all about finding The TRUTH; for experimental scientists (and the most trustworthy theoreticians), science is about the empirical and logical identification and disproof of error. Better to say nothing than promulgate falsehoods.
Regards resistance to error propagation, Langmuir was somewhere in the top 10% (and a much better thinker than I shall ever be).
Langmuir's first example is about the (false) "Davis-Barnes" effect, an artifact of the highly subjective way that Davis and assistant Barnes measured the deflection of alpha particles in an experiment, and deduced some wacky nonsense about atomic physics. What they actually measured was fatigue and visual hallucinations. Innocent but pernicious errors that could have crippled physics.
Motivated reasoning - omitting, rearranging or even inverting empirical evidence to protect prior assumptions - goes beyond innocent error. It is an easy accusation, but the facts (ALL THE FACTS) must be considered when attributing error. If you lack the mental preparation to consider all the facts, you are likely to produce errors, not detect and eliminate them.
If the data are classified or behind a paywall, you will come to different conclusions than others who enjoy better access. The Pentagon Papers are a classic example; the United States was in the wrong war for the wrong reasons, and may never heal from the damage that caused to national strength and prestige. If only we had known ...
Much of the climate "debate" rests on unequal access to data. The "big picture" requires rockets and satellites, infrared spectroscopy, and multivariable vector calculus embedded in vast computer programs; inaccessable to almost all. The best discussions are in the Journal of Geophysical Research - Atmospheres, which was paywalled until very recently. Now that JGR-A is open access, I expect the debate to improve, especially as intransigent ideologues die of old age.
With luck, Steiglitz will live to 100yo, and find time to look into those climate computer programs and improve their validity, traceability, and transparency. Open boxes are more trustworthy than black boxes.
Blind men and elephants; arrogant blind men who've never touched an elephant may have strong opinions about them, often in inverse proportion to the validity (and scientific integrity) of those opinions.
Langmuir clung tenaciously to a very few of his own errors, such as cloud seeding. Everyone has blind spots, everyone makes errors. Apology teaches humility, humility teaches prudence, prudence requires re-testing of prior beliefs.
If you've never been wrong, you've never been right.
The larger "message" of the Steiglitz book is the technical measurement of error (noise). Perhaps someday, data science can teach us new techniques to detect and eliminate errors in personal and national strategies and beliefs.