John Lilly, Ketamine, and a Book

My previous post about John  Lilly talked about he used ketamine. I was listening to a podcast (“More or Less” a BBC podcast about statistics in the news, recommened) and they were talking about drugs in England which are classified as A, B, C with A being the most dangerous. They asked people about what they think the classification was for  three drugs: cannibis (class B), magic mushrooms (class A), and ketamine (class C). They mentioned that ketamine (class C) was “a dance drug popular with ballroom dancers”. This was news to me so I did a search and found this and this. I wonder if this had something to do with Edyta. Of course, it didn’t but who needs a reason to look at pictures or Edyta.

The host of the “More or Less” podcast is Tim Harford and he wrote an excellent popular book on economics called The Undercover Economist which I thought was excellent. If you want to understand what David Ricardo said, read this book.

John Lilly

In 1982-1983 I was on sabbatical in Los Angeles at Xerox (El Segundo) working on the Xerox Star with my friend from graduate school at the University of Washington Gael Curry. I lived in West LA next to my sister and spent a lot of time researching things at the UCLA Medical Library (who knew there was a Journal of Psychedelic Drugs?) and other UCLA libraries. In my researches I, of course, read about John Lilly. He did a lot of research on human consciousness,  psychedelic drugs, isolation tanks, etc. In one of his books he had been taking what he called “Vitamin K” (ketamine) and also looking into the effects of isolation tanks. Then he thought, why not try them together? He was my kind of guy.

He also had a theory about coincidences that is best described with a quote from his home page: “There exists a Cosmic Control Center (C.C.C.) with a Galactic substation called Galactic Coincidence Control (G.C.C.). Within which is the Solar System Control Unit (S.S.C.U.), within which is the Earth Coincidence Control Office (E.C.C.O.). The assignments of responsibilities from the top to the bottom of this system of control is by a set of regulations, which translated by E.C.C.O. for humans is somewhat as follows.” Some might call this a bit wacky but my coincidence related to him makes me wonder.

Lilly was most well-known in the sixties and early seventies and in 1982 when this happened you did not hear much about him. It was just a couple of weeks after I read Lilly’s book and I had been discussing it with everyone. In fact, my sister and I had a running joke about the ECCO and attributed all coincidences to it. I was walking around UCLA in the evening, after dark. and I happened to go around the back of a large building. The building had a sub-basement for utilities and things and one entrance to it was at the end of a long sloping tunnel maybe 75 feet long. As you walked by you could only see the end of a tunnel for a few seconds unless you stopped to look. There was a light by the door at the end on the tunnel and on the door someone has written “Who is John Lilly?” Of course, he was still alive. I assumed it was a variation of the graffiti “Who is John Galt?” that the Any Rand fans wrote around. I suppose Alan Greenspan (who was closely associated with Ayn Rand) might have tagged a few places with it in his youth.

If I had been looking the other way as I was walking, even for a few seconds, I would have missed. Who at UCLA in 1982 would write that? And how soon would it be cleaned up? I think we have another example of the ECCO, or possibly synchronicity.

Blink by Malcolm Gladwell

I read Malcolm Gladwell’s Blink a few years ago and my impression was that he had observed an interesting phenomenon but he had not discovered the underlying reason for the “Blink effect.” In this document I want to give my analysis of the Blink effect.

Summary of Blink: There are many cases where an instant decision, made within seconds, is nearly as good, and sometime better, than decisions made more deliberately and rationally with a lot of thought and analysis. By understanding how these “snap judgments” are made and when to trust them and when not to, you can improve your decision-making.

The first story in the book is about the kouros (a type of statue) that the J. Paul Getty museum was thinking about buying. They performed several tests and analyses and had determined that it seemed genuine. Then they brought in several art experts to look at it and they all felt uneasy about it and thought there was a good chance it was a fake. It was later shown to be a fake.

The book discusses several cases where a quick (within two seconds), intuitive decision is more reliable and then more lengthy decision process. Some examples are:

• the Getty kouros
• Deciding whether a couple will stay together by examining their interactions for just a few minutes
• Deciding whether you will like a professor by matching him or her teach for just a few minutes

Gladwell calls this process “thin slicing”, looking at tiny parts of the world for the information to make a decision. He notes that these decisions processes take place behind a “locked door”, that is, in the unconscious part of the brain, and that even the people making the decisions do not understand how they are making the decisions.

Gladwell talks about how sometimes too much information can hurt a decision process rather than help it. We are overwhelmed with the data and tend to cherry-pick it to support our initial decision. A long decision process can be too predictable and not good for finding unusual solutions.

Sometimes these quick decisions are not good, when they are based on prejudices and preconceptions. Also we are often prey to salesmen who can manipulate us to make bad decisions. Another case where first impressions can be faulty is in things that you have to live with a long time and you become expert with. Beginner behavior is unlike expert behavior.

Gladwell’s conclusions are that fast, intuitive decision making can often be better and we should not underestimate its power or denigrate it as not logical. It is important to understand which situations it works in and where you need to consider both types of decision making. Gladwell does not really say why this snap decision might be better but just observes the fact and gives many examples of it.

My explanation for the Blink effect: Most decisions are an adversary relationship where someone else is trying to influence your decision. If the other person knows how your decision process works they can take steps to subvert it and cause you to make the decision in a way that is wrong for you but good for them. This is often called “gaming the system”. Instant, intuitive decisions are made unconsciously and even the person making the decision does not understand the decision process. So it is hard for someone else to subvert the decision process. On the other hand, a careful, rational decision process is often public and someone else can take steps to subvert it. The result is that the quick decision is better since it is not manipulated.

Gladwell’s kouros example: Gladwell opens Blink with the example of the Getty museums kouros (a statue). In brief, they had a statue and they were trying to decide if it was genuine. They performs many of the standard tests and the statue appeared to be genuine. But one art expect took a look at it and decided instantly that it was a fake. Eventually it was shown to be a very clever fake.

How does my analysis apply here? The rational, lengthy procedures for detecting fake statues are well known in the art world. Someone producing a fake will know exactly what tests will be performed. So the faker knows where to spend time in constructing the fake, in those places where it will be tested. If the forger is very skilled then he will produce a very good fake that passes the standard tests.

An art expert has probably seen thousands of statues and hundreds of fakes. Over the years the expect develops a feeling for which are genuine and which are fake. This feeling is intuitive and even the expert does not really know how the decision is made. In the Getty example, the expect could not say exactly what made him think the statue was a fake. There probably is an unconscious decision process going on that looks for certain clues. But since the decision process is not generally know and the clues that it uses are not known, even to the expert, the faker does not know how to make sure his fake passes the intuitiion test. In other words, you cannot game the system if you do not know how the system works.

Testing knowledge: Tests provide a particularly clear example of how one would game a system. Let’s take a very simplified example to illustrate the idea. Suppose we have a class where you are supposed to learn 10,000 facts. The final exam consists of 100 questions, each asking about one of the 10,000 facts. The 100 questions are chosen randomly from the 10,000 facts. If you get 90 correct then we infer that you know 90% of the material, that is, you know about 9,000 of the 10,000 facts. Pretty much all tests work roughly like this. If someone can determine ahead of time which 100 questions will be asked, for example by stealing a copy of the test before it is given, then that person can just memorize those 100 facts and get 100%. We then assume that they know all of the facts when they really know only 1% of the facts.

The test relies on several related assumptions: that the questions are randomly chosen, that the test-taker does not know in advance which questions will be asked, that the knowledge of the test taker is uniformly distributed among the facts, etc. The cheater games the system by invalidating one or more of the assumptions that make the test work, in our example, the assumption that the test-takes does not know the questions in advance.

Decision making processes: A decision making process is a procedure that involves a series of data points. The decision maker then goes through an algorithm that uses the values of these data points and come out with a decision, maybe a number or maybe a yes or no decision.

In the testing knowledge example, the data points are the answers to the 100 questions, in particular, whether they are right or wrong. The decision procedure produces a numerical score between 0 and 100 which estimates the percentage of the 10,000 facts the tester knows.

In the example of the kouros it consists of the results of several scientific tests of the composition of the statue, the extent of the weathering of the stone, etc. The algorithm is that it must meet some minimum standard on all the tests to be considered genuine.

Sampling: Almost all decision making processes involve sampling. They do not test everything but just take samples that are considered representative of the whole. If the samples pass a test, it is assumed everything else like it would also pass the test.

This is where the decision making process is vulnerable to gaming. If someone knows what samples will be taken then he can make sure it passes the test at those places but not others.

The Blink effect: So the reason for the Blink effect is that quick, intuitive decision processes are hard to game and so remain valid even in the face of an opponent who is trying to subvert the validity of the decision. It is not the fact that they are quick that is important, it is the fact that the decision procedure in unknown. The data points sampled are not known and the algorithm for making the decision from the data points is not known.

So we can get the same effect with a longer, more rational decision process if we take care to prevent the system from being gamed. Of course, saying that does not mean that this is possible, in many cases it might not be.

Blink effect failures: But the quick decision processes are not immune from being gamed, as Gladwell points out in his book. The art fake example is hard to game because it is uncommon and exists in a small world of art expects. And, even though some of this art is valuable, it is not worth the effort to figure out how to game it.

Trust: Where is it worth it to learn how to game quick, intuitive decisions? All the examples come down to one area: trust and selling things. A salesman wants to convince you to buy his product. He wants you to trust him and the representations he makes about the product. It is notoriously hard to know who to trust, who to believe. It is very profitable to be able to lie convincingly and so humans have studied the problem extensively over many thousands of years. Good salesmen have many tricks to get you to trust them. Basically they have figured out some parts of the internal, intuitive decision process that people use to decide who to trust.

People are very bad at knowing who to trust. They are fooled all the time. The problem, in the terms we have been using here, is that people have learned how to game the decision making process that people use to decide who to trust. People use a sampling process to decide trust because it is not possible to look inside someone’s brain and really know what they are thinking. Other people have learned what those sample points are and use them to fool people.

The only real defense is to make the cost of fooling the decision process higher than the value of defeating it. This is the approach of all security, complete security is not possible but it is possible to make it very expensive to defeat a security system. In the case of trust, the tried and true method is to trust people you have known a long time. If you have known someone 10 years and they have been trustworthy then it is usually safe to assume that they will remain trustworthy. This method has two problems. First, it is not foolproof. Conditions can change and someone can be trustworthy for 10 years and they betray you. Second, it is very expensive. It only allows you to trust someone you have known for a long time. How do you buy something on a trip?

Generalization: It is my belief that one of the major problems that we face is people trying to game our systems. Of course, we have had this problem with trust for thousands of years. People are always trying to convince us of things and trying to subvert our decision making processes in order to convince us of something contrary to our own interests.

Rudolf Reconsidered

I always considered Rudolf the Red Nosed Reindeer just another Christmas song but I heard it recently and got to thinking about it. The story goes like this: Rudolf looks funny (red nose) and so the other reindeer reject him. Then, on a foggy Christmas Eve, Santa chooses Rudolf to “guide” his sleigh, Rudolf becomes famous and all the reindeer now love him. Not be a Scrooge here but this is wrong on so many levels.

First, this is a bad moral message to send: reject someone and only change your mind if they become famous and then “love” them presumably to let some of their fame rub off on you. This is shallow and mean.

Second, clearly Rudolf, that is his red nose, is being used basically as a headlight for the sleigh. The headlights do not “guide” the car/sleigh. Santa stills guides the sleigh but now he can see to guide.

Third, suppose the Christmas Eve was not foggy but just dark. Wouldn’t Santa need some kind of light every Christmas Eve? Why was this one special?

Fourth, a shiny light is not what you need on a foggy night. Bright lights reflect in the fog and make it harder to see. You need special fog lights that, for one thing, point down.

Fifth, the weather varies around the world. It is probably foggy in some places every Christmas Eve.

Sixth, red is a dark color and not as good for illumination as other colors. This is why school signs and some fire engines and now in chartreuse.

Seventh, how much fun can “reindeer games” be anyway? Who needs it?

Gilead and Nick Hornby

A few years ago I was reading Nick Hornby’s collections of Stuff I’ve Been Reading columns he wrote for The Believer. You can see the beginning here but you have to buy the magazine to get most of it. You can get the collections at your library. They are great reading if you are a Nick Hornby fan, as I am. In one column he went on and on about how good Gilead by Marilynne Robinson was. I didn’t read it then because, despite his rave, I didn’t think I would like it, especially since I don’t tend to enjoy much of any kind of fiction these days. Recently we picked it for our book group and it was GREAT! I listened to the audio book. The prose was luminous and the book enthralled me, as Nick Hornby said it would. I enjoyed it more than any fiction book in years.

Gilead is the story of John Ames, a preacher in Gilead Iowa, in the form of a long letter to his young son. It talks about his life, the lives of his father and grandfather, his preacher friend Boughton, and the people around them all. The thing I liked best was the sympathetic portrayal of Christianity. Being a confirmed (it was a beautiful ceremony) atheist I don’t have a favorable impression of very many religions, certainly not of Christianity. But Ames was so, well, Christian, that you had to respect him. For him, religion was charity and forgiveness and good will. In one section someone brought up predestination, a sticky subject, and his opinion was, well, yes, we do believe in that but it is a mystery and we really don’t think about it much and don’t let it get in the way of our main purpose which is to help people and preach forgiveness and other virtues.

I liked it so much I got Marilynne Robinson’s first book Housekeeping and I am partly through it. The language is just as good but I am not as captivated by the story. More on it when I finish.

The Tipping Point by Malcom Gladwell

Gladwell does not have any mathematics in his book. They say that each equation halves the sales of a book. But some of us think better mathematically. It makes things more clear. So I will discuss the book with a few equations and models.

Exponential growth: Basically the book is about exponential growth processes but complicated ones. He uses some ideas from simple exponential growth as examples to motivate his ideas so let’s start there.

Let’s start with the equation a^b, a to the b power, where a > 1 (otherwise it will get smaller instead of growing) and b > 1 (if b=1 it won’t grow and b < 1 gives negative exponential growth). This gives rise to the familiar graph of exponential growth. (See Wikepedia.)

In mathematics we would allow non-integer exponents (values for b) but here we want to restrict to discrete, generational growth so let’s change the equation to aⁿ and require that n = 1, 2, 3, 4, 5, … where each value of n is a generation of growth. This gives rise to the familiar doubling style of exponential growth. This is often used as an example to impress people about how exponential growth works with stories of grains of wheat on a chessboard or folding a piece of paper. (Again see the Wikipedia entry.) As we know, it starts slowly but gathers speed and after a while grows very quickly.

Networks/Graphs: This aⁿ model is the start of what Gladwell is talking about but we need a more complicated model to capture the ideas he is presenting. Almost all of his important examples are networks (that is, graphs) of people where a link between two people represents a social connection. So we have a large graph where people are the nodes and a link between two people means they know each other, come into contact with each other, or interact with each other in some way. The number of links to a node can vary widely, some people have many links and some only a few. The links are not directional so we can’t say that they go into or out of the node.

Gladwell most often talks about “infectious” processes on these graphs, that is, where some “infection” (maybe an actual infection or maybe an idea or bit of information) starts in one or more nodes and then, each generation, it moves along the links out from that node to other nodes. But the infection doesn’t always move along the link. He talks about various factors (stickiness, mavens, etc.) that affect how likely the infection is to move along the link.

So the process is one of the infection spreading through the graph in steps, each step is a generation. This is a complex form of the doubling. In simple doubling, each node has three links, one the infection comes in on and two it goes out on, and the infection always goes out on a link. If you start with one node in the graph infected then each generation the number of infected nodes will double and you get the equation aⁿ.

Networks node parameters: So for each node there are two parameters: the number of links it has and the probability that an infection will move along a link. For example, if that probability was 0.1 then the infection would move along 10% of the links. Gladwell does not talk about cases where the probability of an infection moving along a link varies between the links. This is a level of complexity in the model that he does not need.

So we could rewrite the equation aⁿ as (probInfect*numLinks)ⁿ where “probInfect” is the probability an infection will move along a link and “numLinks” is the number of links out of the node. Note that 0.0 ≤ probInfect ≤ 1.0. The (probInfect*numLinks) factor is different for each node and so we can’t use the equation to predict the number of infected nodes after n generations. We are just using the equation form as an analogy to show how it grows out the simple exponential growth model. We could easily write a computer program that would simulate the graph and provide growth statistics.

We need to add one more factor to the equation: (probSticks*probInfect*numLinks)ⁿ where “probSticks” is a probability (note that 0.0 ≤ probSticks ≤ 1.0) related to the “stickiness” of the infection. It means that the infection does not always move along the link. This stickiness does not vary from node to node.

Gladwell and the network model: Now we have the notation to describe some of the concepts that Gladwell uses in his book.

1. A “Connector” is a node with a high “numLinks” value.
2. A “Maven” is a node with a high (close to 1.0) “probInfect” value.
3. A “Salesman” is a node with a high (close to 1.0) “probInfect” value.
4. An “Innovator” is a node where the infection is likely to arise spontaneously.
5. The “Power of Context” affects the value of “probSticks”.

Theoretical work on networks: Erdos and Renyi did work in “random graphs” some years ago. There has been a lot of work on non-random networks in the past 10 years.

Due to the Stanley Milgram study and the subsequent play we hear a lot about “six degrees of separation”. Here are some other estimates:

1. The average separation of nodes on the web is 19 clicks.
2. The average separation of chemicals in a cell is 3 chemical reactions.
3. The average separation of scientists in different fields is 4 to 6 coauthorships.
4. The average separation of neurons in the brain of C. elegans (a tiny worm) is 14 synapses.

Power laws: Most networks conform to power laws meaning that the number of nodes with
n links is proportional to n^-k. For example, in the web the number of links, n, incoming to a node is proportional to n^-2.1, that is, roughly a square. So there are about 1/4 as many nodes with 10 incoming links as there are nodes with 5 incoming links, and about 1/16 as many nodes with 20 incoming links, etc. The k for outgoing links is about 2.5. Think about road maps or airline route maps. These follow power laws in terms of the number of links each node has.

Scale-free networks: Networks like this are called scale-free because they look the same at all levels of details. (We might note that this is also true of fractals.)

The rich get richer:
As scale-free networks grow new nodes prefer to link to existing nodes that already have a lot of links. If you are putting in a new airline route, it is more likely to go from Hobbs to Albuquerque than Hobbs to Portales. You are more likely to become friends with a friendly person who already has a lot of friends than a loner. If nodes are added using this “the rich get richer” model then you get scale-free networks where the number of links to a node follows a power law.

Robustness, failures and attacks: Since well-connected nodes are rare in scale-free networks, they are robust in the face of random failures. But they are vulnerable to directed attacks (just attack the big nodes).

Introduction to Charlie’s Book Notes

The purpose of this blog is to record my thoughts about the books I read. This is mainly to help me to remember the book but also so other people can read my thoughts if they are interested. I always search for comments about the books we read in my book groups.

I am going to start with a number of existing notes on books I have read, just to get them recorded here.