Lecture 8

readings for 2018-03-05

As promised, this week we go somewhat back in time, from this week’s “bratty” mathfights of 1955-1956 in the academy to the actual behind-the-fence birth of computational statistics at bletchley park during WWII

We closed last week with the section ‘Statistics goes to war’ (Sec. 3.7 of “The Empire of Chance” (1990)), so we’ll open this week with “Bayes goes to War”, Ch 4 of “The Theory That Would Not Die” (2011), a popular book written by the science writer Sharon Bertsch McGrayne. (please also read the 2-page dessert, ch 5)

McGrayne opens with

• death of Bayes
• u-boats
• Poles
• note: behind the fence knowledge, leaking to industry not academia
• side note: role of Monte Carlo

A more technical secondary source is commentary “Statistics at Bletchley Park”, written by Sandy Zabell, a statistics professor at Northwestern. This chapter appears as part of the Wiley edition (2015) of the “General Report on Tunny” (written in 1945 but only declassified in 2000) written by the Bletchley Park researchers I.J. Good, D. Michie, and G. Timms.

Of this, please focus on

• the beginning:

• pp lxxv to xci
• (read through “The Bayesian Approach” but not “The vulnerability of the SZ 40 prototype revisited”)
• (that’s pp 75-91)

• the end:

• pp xcvii-ci
• (read “Aftermath” through the end)
• (that’s pp 97-101)

Zabell opens with

• Bayes before WWII (topics McGrayne covered in ch 1,2)
• digression on Bayes
  • hypotheses require priors
  • sequential updating natural in Bayesian framework
• turing goes to Bletchley
• side note: role of Monte Carlo

The third excerpt is “Breaking Codes and Finding Trajectories: Women at the Dawn of the Digital Age”, Ch 1 of “Recoding Gender: Women’s Changing Participation in Computing” (2012) By Janet Abbate, a professor at Virgina Tech.

Of this, please focus on

• pp. 14-16,
• pp. 21-22,
• bottom of 26-29, and
• pp. 33-35.

Abbate brings this across the atlantic and after the war. observe:

• role of physical labor
• role of engineering
• role of military funding leading to new corporate success: tech startups
• note: behind the fence knowledge, leaking to industry not academia
• side note: role of Monte Carlo

discussion 2017

Inside fence: inside information, shared freely within fence

• Crypto community including Proto-NSA, technological community including Bell Labs, Bletchley Park

  • Includes those outside gov in US, but not those outside gov in UK
  • Not trying to publish things & not listing affiliations

• Outside fence: public domain, Bayesian in general

• Frequentists (old school) vs. Bayesians (new school)

  • Freedman: "Some Issues in the Foundation of Statistics"- objectivists (Frequentists) vs. subjectivists (Bayesians)
    • Discusses model validation
    • Basic problems in applying statistical models to social phenomena
  • Remember Wallach's 3 uses of data: statistics used for prediction, explanation, exploration

After War:

• UK barely had computers, wasn't until after war that general purpose computing was born
• UK economy destroyed at end of war (empire lost, huge period of retrenchment, hard to argue for truly staggering costs involved in producing electronic computers, huge resources went to atomic bomb because they know from US example that it works)
• Explosion of operations research post-war
• Incredibly large-scale simulation using mechanical device
  • Monte Carlo now
  • Bayesian: score is not flat like golf course, but wave of possibilities (sees what looks like German)

Turing:

• Turing trying to build general purpose computers same time as Americans

• Americans could say that computers are important for bomb

  • British had 6 years empty on resume
  • Turing using Bayesian methods without having knowledge of Bayesian thinking or branding

• Post-WWII, everyone thought physicists won war

  • Couldn't hide atom bomb they invented
  • Got all credibility & funding
  • Physics of bomb are not hard
    • Actually building a device is difficult --> chemical engineers at Los Alamos similarly classified
    • Took 50 years for world to figure out how data scientists won the war
  • Role of Bell Labs as computational, statistical analysis of personal data
    • For sake of private enterprise
    • How much of Bell Labs overlaps with what become the NSA? How much of the NSA overlaps with Bletchley Park
  • Bell Labs as conspirator with NSA

math

Bayes

• Statisticians' fallacy:
  • Island example: if you test positive for sick, 99% chance you're sick --> wrong
  • Yet by using Bayesian probabilities, only 50% chance you're sick because given prior probability
• Using Bayes = using probability in Wiggins' mind
  • In a position where you need to make concrete decisions about which hypotheses you want to test --> because huge amount of labor required

Fisher: whole knowledge is in false positive rate (alpha)

• False positive rate is 1%
• One case in which Fisher's approach works: false positive rate is small, false negative rate is small, and even odds
• Frequentist: uses significance test as grounds for him concluding
• Bayes - subjectivist
  • Ban weight of evidence
  • Probabilities being about beliefs

Computers

• All-purpose for any numerical calculation- calculator
  • E.g. ENIAC
• All-purpose:
  • Symbolic manipulation
  • Turing machine 1936
• Data storage- computer as information processor
• Before war computer meant: person, often a woman
• Severe resource constrains on doing very high performance computing
  • Geological, temperature, security considerations in where to put data center
• UX - user facing GUI
• Bayes: data storage & numerical calculation

NOTA BENE: FISHER+NEYMAN+PEARSON FIGHT WAS 1955 AFTER ALL OF THIS!!

• speaks to how fully OUT OF IT academic stats was from revolution
  • bayes
  • comp
  • data