Separate Signals from Noise

draft notes for a one-page quick-start "Think Better" module:

"Don't over-interpret the noise!"

• Signals convey information
• Noise is random from natural sources
• Interference is deliberate by adversaries

... in a Nutshell

Observations are imperfect — they come with errors. Clocks may be fast or slow; people lie when asked questions in a poll; sources get confused; cosmic rays flip bits in computer memories; enemies use jammers or camouflage; data compression fuzzes up images; translators introduce ambiguity or overlook nuance. Information theory is the math to describe, in ideal circumstances, signals and noise.

In real-world situations, however, often the math doesn't apply. Sources of noise come and go unpredictably, new phenomena emerge, and deceitful opponents change their strategies to maximize confusion. Instead of blind trust in arcane formulæ, wisest is to be aware and consider:

• How big is the signal compared to the noise? If a signal is weak perhaps filtering, averaging, or other forms of post-processing can help sharpen it — but it may still need to be distrusted, particularly if it doesn't fit other sources. Rather than strenuous work to amplify faint data, often a better approach is to improve collection: get a bigger antenna, take a larger sample, get closer, or ask better questions.
• What characteristics does the noise have? Is it seasonal or directional? Is it concentrated at certain frequencies or in particular modes? Does it ever recede, and can moments of clarity be recognized and taken advantage of?
• Is some of the noise artificial interference from a malicious adversary? If signals are concealed on purpose, then new opportunities appear. Can deliberate noise be analyzed, reverse-engineered, and then subtracted? Might enemies be recruited to turn off the jamming? Could a source of interference be attacked directly?

Remember

Don't over-react! Short-term fluctuations may well be random noise, not indicative of long-term change.

(cf Science and Pseudoscience (2001-10-06), Expert Political Judgment (2010-05-13), Signal and Noise (2012-12-25), Pearl Harbor (2017-10-20), ...) - ^z - 2020-02-01