Putting Blackbox Models in Charge

Something I read in the Economist this week struck me, (April 20, 2013 edition - pg. 30 "That swooning feeling").  This seems worth quoting completely.  It is speaking about various theories about the repeated apparent dip in the economy each spring.  "One theory is that models interpreted the economy's plunge in late and 2008 and early 2009 as partly seasonal, and responded by nudging up subsequent winter figures and nudging down summer data to compensate."  Really!  Are the models that the Bureau of Labor Statistics this autonomous and opaque that no one who is in charge of using them can even figure out whether such a thing is occurring?  Their testing solution for this evidently works like this - "But the federal Bureau of Labor Statistics has found that the pattern persists even if the job numbers are seasonally adjusted without those recession months."  Could they really have so little comprehension of the model they are using, that they can't figure out whether such an effect is likely, and the only way of testing is to remove the data and run it again?  The stock market swings up and down, hiring and lay-off decisions are made, investments made or abandon on this number, and evidently no one has any idea how it actually comes into existence.  It would be better to publish the raw employment surveys and let people construct their own analysis than to use a black box of this sort.

Inflation and Wealth Concentration vs. Asset Bubbles

The Fed has been pumping money out into the economy for some years now, leading to a continual gnawing fear in some circles that inflation is about to skyrocket.  I would say that this is overlooking a transformation which has occurred over the course of the last couple of decades.  First, I would make some hypotheses about money.  I would say that money wants to multiply, and that money wants to concentrate.  These are a bit anthropomorphic, but money is concretized human desire, so a bit of anthropomophic speculation is perfectly reasonable.  Next, money dispersed over a broad population is what sustains demand, and therefore makes it possible for prices to rise - inflation in other words.  When money is concentrated beyond a certain degree, inflation will no longer be possible, instead asset bubbles will predominate.  Money will always attempt to concentrate and multiply, so left to its natural tendencies, it will flow to a smaller and small portion of the population.  This small proportion of the population cannot sustain demand of a broad consumer market, so inflation cannot increase.  Formerly, it was necessary to hire people and pay them an increasing wage in order to make more money, but automation and financial instruments have decoupled this connection to a great degree.  Paying money to labor is dispersion, and the return is most likely logarithmic, while investing in obscure financial instruments is concentration, and the return is exponential.  It is easy to see what the trend will be.  Now, the question would be, how to put some numbers on this and make some sort of convincing case?

Is there a place for the mind in physics?

I read a blog post yesterday on npr reviewing a book on the mind and consciousness.  NPR - Blogs, Is There a Place for the Mind in Physics? The review was by a physicist of a book by Thomas Nagel that proposed that the mind had an inherent existence.  The reviewer, Adam Frank begins by asking the reader to imagine a blue monkey.  Is this blue monkey real, or is it just some sort of phenomenon arising from activity of neurons in the brain?  If it is real, where is it real?  I remember another idea which occurred to some months ago while I was unlocking the shed to free my bicycle.  It was locked with a combination lock which I was carefully turning back and forth till it unlatched.  It occurred to me that in the physical world, there seemed to be two types of locks, but in the software world, there was apparently only one kind.  The first kind of lock in the physical world is a ward lock.  A carefully shaped object is used to alter the shape of pins inside so that one is able to turn it.  In the case of the second lock, one has some sort of secret knowledge that can be used to adjust the mechanism into some configuration that will allow it to open - a combination lock in other words.  Inside a software program, there is apparently only one sort of lock possible, that being the secret knowledge sort.  Any sort of method that I thought of to imitate the ward lock worked out generating a random number, that if one were to capture, could be used to open the file.

Thinking further, the combination lock is a mechanical representation of the numbers of the combination.  In a sense any sort of computer or calculating device - the brain as well - is something much like the combination lock.  It is a mechanical representation of the information in the combination.  If one were to actually build Babbage's Analytic Engine, one could construct a program to calculate some value.  One could calculate the same value using a modern computer, using essentially the same method as in the program on the Analytic Engine.  One could almost certainly generate a protein sequence that would fold in a particular manner to calculate the same value, using the same method.  Thus it appears that both the method and the value do not exist in any particular mechanism.  They seem to be both everywhere and nowhere.

Turing Machines and the Brain

I was listening to NPR this morning.  A surgeon was describing his impression of the operation he was performing on a girl's brain.  He commented that this was the most amazing part of the operation... we can see where Maribel thinks and feels - this is Maribel.  My first thought was, no Maribel is a program running on this hardware.  Part of Maribel, how she identifies herself as Maribel, would be the consistent input of sensations from her body, so Maribel isn't located here in the brain exactly at all.  Then I started thinking about what sort of computer the brain was.  Every sort of computer we have built ourselves is a Turing machine in effect.  The Instruction Pointer trundles up and down a strip (code segment) reading instructions, calculating, and writing out results.  Meanwhile the brain is not built anything like this.  Neurons individually are sum and difference calculators.  They take inputs from many other neurons, sum the excitatory inputs, sum the inhibitory inputs, subtract one from the other, and then fire or not depending on whether the result is greater than some threshold.  The nets of neurons form feedback loops which generate quasi-periodic waves of excitation with some of the information being encoded by the amplitude of the waves across the surface of the cortex.  Meanwhile, there is yet another level of calculation going on.  The glial cells support the individual synapses and share information regionally, this determines which synapses are strengthened and which are removed.  The pruning and augmentation of synapses could be considered part of the long term calculation which the brain is performing as well, and the glial cells mediate this, so there is another level of cells involved in the calculations the brain is making apart from the neurons.  I wondered if this structure could be modeled by a Turing machine.  Supposedly, everything that is calculable can be reproduced by a Universal Turing Machine, which is basically what a computer is at the machine language level, but can the brain's calculating architecture by modeled by this structure?  If the brain is not representable as a Turing machine, then some of its operation would not representable as calculation, at least according to the Church-Turing theorem, and to this degree, it could not be captured by a computer program.

Language, Music, and Cheesecake

I heard a bit on NPR this morning describing music as "cheesecake," a pure accident, something we had the sensory and neurological equipment to appreciate, but had not evolved to seek or create. They then proceeded to a number of examples, for instance, monkeys could not be trained to follow a beat in a year of trying, but parrots could learn to dance on their own. This ability was evidently a characteristic of species that are vocal mimics. Now, you might ask what species of primates would be better described as vocal mimics than humans? This is the way that language is learned by infants. Parents make sounds that are easy for infants to produce, and imitate sounds that infants make, and infants try to imitate the sounds which they hear their parents making - vocal mimicry beyond doubt. Parrots seem to be using vocal mimicry as a means to bond together a large group of individuals in a complex environment. Mimicry allows individuals to interact and create bonds as individuals, and to hear where they are in the flock even if the environment does not allow them to see one another. Black birds gather in large flocks and sing in fall and winter near where I live. Each bird sings a simple repetitive part, but all the parts are locked together in rhythm. If some human ancestor did this, we would call the result music, but not so for  black  birds. One can imagine a human ancestral species behaving like this, the individuals spread out but still connected in a group. The individuals calling out relatively continually and imitating one another to show that they are members of the same group. The sounds they create can communicate many things, possible threats and presence of resources, but also emotions and relations between members of the group. Because the ability to bond and act as a group is a great survival advantage, evolution selects for enhanced neural and vocal equipment to make this form of communication more effective. Thus, by this model, music is older than language, and is what language evolves from. The main question for this model is why we move from being bipedal parrots with some specific threat calls to having specific calls for every object we see around us.

Better modeling for Homo Economus

Using a single motivation as the basis for all human behavior, as classical economic theory does, would necessarily be a little limiting. One can see the necessity of doing this in the 19th century, as it made the mathematics tractable, but in the present it obviously is too limited. A computer simulation can obviously model a much more complex structure of human motivations. I am visualizing a pool of several thousand agents that would model a market. They would make decisions based on two scores, one of which would determine tolerance for risk, the other would determine the degree to which they would grasp for reward. Both scores are calculated by having all the agents in the simulation exchange tokens with each turn. The first calculates public mood, or what everyone "thinks" about the situation being modeled. All the agents have an internal token reflecting what they "think" about the situation. They trade tokens with all other agents accessible to them in the situation being modeled. At the end of the turn, the tokens will be averaged and combined with the value of the internal token. Introvert agents will take 90% internal token, and 10% external average, extrovert agents will take 10% internal token and 90% external token. The value will then be loaded into the internal token for the next round. A pool of agents who take a 50/50 split could be set up as well with the usual normal distribution 16% / 68% / 16%, even though perfect balance personalities are probably extremely rare. The pools of agents could then be adjusted until the simulation behaves something like real life.  Also, rather than greed being the fundamental human motivation, I would propose jealousy. In order to add this to the simulation, a second internal token will be added to each agent. This token will record how far above or below the average the agent finds itself to be. Each turn all agents that are directly connected will exchange tokens of their perceived economic state. Introverted agents will exchange with a smaller pool than extroverted agents. All agents will receive information from the total pool as a somewhat randomized average to reflect the influence of the media. The pair of tokens will be used to determine whether the agent will buy or sell when presented with opportunities in the model. The first determines response to risk, the second, motivation to acquire. I would expect a pool of several thousand of these agents to behave much more like a group of humans in a market than a simulation based of classical economics would, and yet be easily simulated.

Interpersonal Communications Class and Other Silly Ideas

Listening to my steady drip of primary coverage over NPR, I heard that Newt Gingrich always made sure to inform his wife that he loved her in front of his then mistress. (The things you learn on NPR). Now, if this isn't an example of the favorite bullet point in my miserable interpersonal communications text book, I don't know what would be. I quote, "Meanings are in people." People have been arguing whether there was any sort of inherent meaning in language for since at least 500 BC, but the fine worthies who, loosely speaking, wrote this textbook, have solved all these arguments by decree... no explanation or justification required.

Now, we may understand the truth of "Meanings are in people" in regard to interpersonal communications by employing the "Transactional Model" of human communication. (A term completely undefined in the text, though they do have a very pretty diagram about which they say little to nothing of substance). Gingrich is sending a "message" out to his two listeners by selecting from a collection of arbitrary sounds the sylables "I love you," which in the transaction will be held to have some sort of meaning, (which of course is in people). Gingrich is encoding in the arbitrary sounds coming out of his mouth two distinct messages, one for each listener, which he hopes they will interpret in a manner agreeable to him. For his wife on the phone, his intended message is, "Everything's fine, nothing to worry about, and I certainly won't be screwing anyone else's brains out here in a few minutes." For the woman with him he is encoding the message, "I certainly won't leave my wife for you. That would cause scandal, and all manner of political inconvenience." In this particular transaction all parties receive the messages, interpret them, and go away happy. Particularly Newt, who gets to screw his brains out while avoiding political scandal. Later, his wife will develop an issue with the arbitrary collection of sylables, "I love you," which Newt employed to encode his message. This is because her decoding of the message was flawed... due to an insufficiency of information. But plainly the problem is with her decoding of the message - as the "Transactional Model of Interpersonal Communications" plainly demonstrates.

The book, (which I bought for $82 at the college bookstore as I suffer from brain damage as well as social insufficiency), was unfortunately much less clear on the meaning of the "Transactional Model of Interpersonal Communications," than the example above employing our all round lucky fellow Newt. By way of explanation the esteemed worthies who authored the book offer a diagram which consists of several ovals inside of other ovals with some wavy lines running from top to bottom labeled "Noise." I left those out of my excellent descrption above, you'll have to figure those out for yourself. I was finding this superlative text to be trifle annoying by the end of the first chapter, when I paused to reflect on what I'd just read, and found that I couldn't remember a single thing. I probably should have dropped the course, and taken the book back to the bookstore at that point, but my folly is boundless. Several chapters later, I experienced a sudden insight. I said to myself, "These people just love headings, which they follow with some glop which doesn't really establish anything the heading says... and they are always explaining major terms with a diagram, no actual explanation anywhere... I get it! The whole book is actually just a bunch of PowerPoint (tm) presentations they bundled up and shipped off to the publisher almost unrevised, without actually sitting down and writing anything." This Zen moment inspired in me a new reverence for these worthies and their excellent abilities to extract doubtless considerable financial rewards for almost no work on their part. I must continue to encode messages, and strictly follow the "Transactional Model of Interpersonal Communications" in the hope that, one day, I too will be found worthy.