Biological boot loaders

Elon Musk a few months back summarized Nick Bostrom’s book Superintelligence: Paths, Dangers, Strategies:

While I find this issue indeed existentially important, I was a bit disappointed by the book because Bostrom turned out to be himself a biological boot loader – this time for Yudkowsky’s and Hanson’s memeplexes.

While these are by far not the worst memeplexes to be infected by, still in this important problem space one would like as many independent high-caliber search paths as possible.

A Christmas classification problem

Talking about classification problems with blood-soaked consequences, the Christmas period offers an archetypal classification problem:

The three wise following the star announcing the birth of the new king, stop on their way at their colleague King Herod. Not minding the old king’s feelings, they autistically tell him the “good” news, that a child is born, which is going to take his kingdom.

King Herod, a true politician, keeps his poker face, sends of the three autist on their way to find this new king baby and asks them to stop by on their way back to let him know, where he can find his heir apparent so he too can go to murder to death greet the little fellow.

Anyhow, long story short, kings find the baby, but are warned by an angel to take a different route avoiding further confrontations with Herod.

Herod is left in the dark, but being a laudably proactive type of ruler decides to run a quick search-and-destroy op for the little boy:

Then Herod, when he saw that he was mocked of the wise men, was exceeding wroth, and sent forth, and slew all the children that were in Bethlehem, and in all the coasts thereof, from two years old and under, according to the time which he had diligently inquired of the wise men.

Matthew 2:1-23 (King James Version)

or in more readable format:

kill * where (age<2 & gender='M' & region='Betlehem')

Matthew 2:1-23 (SQL Version)

See, Herod needs to find the baby Jesus, a most important classification problem if there was one, but has only three features to work with: approximate age, gender and rough region. The problem is particularly hard, because there is large class imbalance (new_king=TRUE only for a single tuple, the little Jesus).
He therefore executes a crude classification algorithm (a simple decision tree to be specific) which has a high sensitivity, but abysmal specificity.

Tintoretto: The Massacre of Innocents. And you thought you had a bad Christmas, because you got socks again.

Thus the first Christmas was soaked in babies’ blood and Game of Thrones is a Disney production compared to the Bible.
And that’s this year’s uplifting Christmas story for you!

Replication problems: pre-socratic edition

It seems replication problems are nothing new in science:

Fortunately twitter also suggests a light at the end of the tunnel:

You might also recall an alternative solution – Kahneman’s daisy-chain replication proposal.

CIA torture and the Type I vs. Type II error

The trade-off between Type I (false positive) and Type II (false negative) errors is one of the fundamental laws of nature. High sensitivity comes at the steep price of a low sensitivity and vice versa. Where along the trade-off curve you position yourself depends on your utility function. While it might seem as only an arcane statistical/machine learning topic, it is in fact ever-present in our lives. A typical example is the security vs. freedom or security vs. privacy debates.

A recent, unusually pure form of the trade-off was seen in the highly controversial Meet the Press interview with former US Vice President Dick Cheney (full transcript) [1].

In this discussion, Cheney is defending the use of enhanced interrogation tactics torture in reaction to the recent Senate’s report. At a critical point of the interview, he states that he believes that national safety required a high-sensitivity (and hence tragically low specificity) policies:


I mean, let me go to Gul Rahman. He was chained to the wall of his cell, doused with water, froze to death in C.I.A. custody. And it turned out it was a case of mistaken identity.


–right. But the problem I had is with the folks that we did release that end up back on the battlefield. Of the 600 and some people who were released out of Guantanamo, 30% roughly ended up back on the battlefield. Today we’re very concerned about ISIS. Terrible new terrorist organization.

Not only is the trade-off here demonstrated in a crystal-clear form, but it is also a situation of highest stakes – the choice of the trade-off point is set by the potentially most important utility function a society has – its national, democratic and civic values as interpreted by its policy-makers.

Update 12/01/2015: Thirtyfiveeight have an excellent article on the costs of Type I errors in cancer treatment.

[1] The interview is very interesting (if I am allowed to say this given the tragic nature of it) from many aspects. Cheney seems to see himself and a hard decision-maker, willing to do unpopular decision for the greater good of his nation. The terrifying part is, that it seems that doubt never seems to play any part in his epistemology.