SciFi Politics: Part 2

In “Hitchhiker’s Guide to the Galaxy” the greatest torture device is a device that makes you realize how small you are in relation to the universe. The joke is both that we are insignificant, we ‘know’ we are insignificant, and even then we still would need some special device to fully appreciate just how insignificant we are. I think the same thing about the complexity of reality. No one can truly grasp it all in their head, and it’s hard to measure in a useful way, so when explaining why it’s not obvious that the minimum wage is a good thing, I sometimes stammer “Well… you know… it’s hard to know that because reality is complex.”

Image result for hitchhiker's guide to the galaxy

To simplify that complexity, our most basic model tends to be a story. A full historical account of an event is fundamentally taking what is tiny sample of the total set of information related to the event, and turning it into a story that maps to the structure of how humans understand reality.

After all, what even IS a story? I mean, scientifically, you know? Andrew Gelman offers his theory, which mainly focuses on the wrong way to use stories “Storytelling has been championed by a wide range of scholars who would like to escape the confines of rigor.” The historical uncertainty comes from the fact that the sample is typically biased and an incredibly small set of total information and that the historical story is a set of functions mapping a high-dimensional set of primary sources to a lower dimensional account of causality that humans appreciate. The baby died because the hyena ate it. The stock market increased because GDP went up.

There is typically the ‘obvious’ cause: Slavery caused the civil war. The assassination of Franz Ferdinand caused WWI. The self-immolation in Tunisia caused the Arab Spring. The gulf of Tonkin incident caused the Vietnam war. Then there might be tombs of literature, often in disagreement, attaching these singular causes to their roots. Then some clever but unverifiable theories attaching them to more general trends that reach even farther back in time. Decades perhaps. That’s the story of both chaos theory as well as the limits of human understanding. We daisy-chain causal stories.

There is this interesting book called Human Smoke, which is sort of about WWII, that takes a different approach. The author spent years going through old newspapers leading up to and during WWII, and took out snippets that are dissonant with the common historical narrative. Each page has a snippet from a newspaper, or a short excerpt, with no argument or author written story. The book was reviewed by the New York times, but I first heard about it on Unqualified Reservations, whose author’s goal is to collect all historical documents and books over the past few centuries that clash with the modern progressive movement.

As a method of offering information to a reader, it is almost more honest with its bias.

I think historical narratives are weird. We often are trying to understand more than is possible. A way to think about it is that while all humans have different cognitive abilities, there is a serious upper bound. A single human can only know so much information, and have so many analytical tools to decompose that information. So our knowledge of historical narratives has to fit inside our brain, and at some level of complexity or information can’t process any further. We try to overcome these issues by specializing, but this can bring in massive bias through shared societal priors.

It took the mind of a genius to write War and Peace. The main characters in the book never existed, but the story covered a historical narrative. It was based on history, but it was a story. Tolstoy imagined it in his head. The computer he calls a brain took fragments of history, his observations of humans, and a beautiful ability to build characters, to create a story. The difference between War and Peace, and Das Kapital or Chomsky is tiny. The humans who created the works had an uncanny ability to pull together disparate information in their head.

Our brains have an incredibly ability to find patterns where none exist. We can take feelings and use them to generate fake landscapes that are correlated with those feelings. Or we play with words whose sounds and feelings feel meaningful. Novels and plays let the artists capture and represent a distilled aspect of human existence. The reasoning we use to procedural generate these patterns are part of the same brain algorithms we use to find patterns in the physical world, or in human interaction. We need to assume these pattern matching algorithms are hopelessly broken. They are biased classifiers, which desperately need a scientific structure.

Understanding history without over-fitting your model is incredibly challenging. In 1931 a History professor at Cambridge, Herbert Butterfield, called this concept of understanding the past through the lens of the future as “Whig History.” His criticism was more pointed, and focused on the Whig practice of searching for some modern social accomplishment and tracing it back through time to paint a history of noble struggle, and using it as evidence of the indisputable validity of modern social progressive policies.

 It is part and parcel of the whig interpretation of history that it studies the past with reference to the present; and though there may be a sense in which this is unobjectionable if its implications are carefully considered, and there may be a sense in which it is inescapable, it has often been an obstruction to historical understanding because it has been taken to mean the study of the past with direct and perpetual reference to the present. Through this system of immediate reference to the present day, historical personages can easily and irresistibly be classed into the men who furthered progress and the men who tried to hinder it; so that a handy rule of thumb exists by which the historian can select and reject, and can make his points of emphasis

When the need arises to sort and disentangle from the present one fact or feature that is required to be traced back into history, the historian is faced with more unravelling than a mind can do, and finds the network of interactions so intricate, that it is impossible to point to any one thing in the sixteenth century as the cause of any one thing in the twentieth. It is as much as the historian can do to trace with some probability the sequence of events from one generation to another, without seeking to draw the incalculably complex diagram of causes and effects for ever interlacing down to the third and fourth generations

if Protestants and Catholics of the sixteenth century could return to look at the twentieth century, they would equally deplore this strange mad modern world, and much as they fought one another there is little doubt that they would be united in opposition to us; and Luther would confess that he had been wrong and wicked if it was by his doing that this liberty, this anarchy had been let loose, while his enemies would be quick to say that this decline of religion was bound to be the result of a schism such as his. The issue between Protestants and Catholics in the sixteenth century was an issue of their world and not of our world, and we are being definitely unhistorical, we are forgetting that Protestantism and Catholicism have both had a long history since 1517, if we argue from a rash analogy that the one was fighting for something like our modern world while the other was trying to prevent its coming. Our most secular historian, and the ones who are most grateful for that “process of secularisation”, that “break-up of medievalism”, of which so much has been traced to the Reformation, are inclined to write sometimes as though Protestantism in itself was somehow constituted to assist that process. It is easy to forget how much Luther was in rebellion against the secularisation of Church and society, how much the Reformation shares the psychology of religious revivals, and to what an extent Luther’s rebellion against the Papacy helped to provoke that very fanaticism of the Counter-reformation against which we love to see the Protestant virtues shine. And it is not easy to keep in mind how much the Protestantism that we think of today and the Catholicism of these later times have themselves been affected in turn, though in different ways, by the secularisation that has taken place in society and by the dissolution of medieval ideals.

How is historiography like forecasting? Can we be better historians by spending our time forecasting the future? Can forecasters be better historians than historians themselves, because they understand the relationship between current information and future predictions?

In time-series forecasting we refer to it as the information set available at time t. In this case it’s relatively intuitive, although the concept is still frequently ignored in practice and academia. When fitting a model to predict the future we always start with a set of data. Fitting a model over the whole time-period overfits the data, as the parameters are being tuned without respect to time — allowing the future to inform the past with information not available at the time. It’s cheating. If you fit a regression on a series of data that occurred over time, you are letting the parameters take on a value that assumes it is scientifically acceptable for them to have known what was going on the whole time.

The short solution is to use pseudo-prediction–out of sample validation–where you leave some of your data out. The challenge is that we choose our models, and we re-run backtests when we want, and as we run models we learn more about the problem space. Learning more is important, since we can then re-specify our model. As it turns out though, there is no clear method to ensure we are learning more and re-specifying our model to be more accurate, or we are cheating by using information from past models and results to overfit our model and then placing that model within an academic narrative that is also overfitted nonsense. If the signal is strong enough this method will still identify it, but it might also identify misleading signals it mistakes for real.

The problem is our brains are incapable of excluding information subset on time. A computer can do it trivially. For example, economic forecasting models were unable to predict the 2008 recession. Being able to predict the 2008 recession would have implied being able to predict asset backed security default rates. Now there is lots of research trying to find how this could have been predicted. The question is implicitly asking if there was a signal, conditional on information available in 2007, that could have forecast the recession (which because of the dynamic nature of markets, would have caused a slightly less severe recession earlier).

A cooler scifi way to solve this problem would be to gather all the smartest economists in a room a day before the crisis began, and give them access to all the information they had available at the time, and say “Tomorrow the economy will experience a unique event, what do you predict this event is?”. Since this is scifi, we can imagine we are speeding up their brains and give them 1,000 years each to build their model. We can then evaluate: Did they predict the right event? We can play with other dimensions as well, we can give them incrementally more information, augment their brains to be incrementally smarter, and experiment with asking them to predict it a day earlier, a year earlier, and so forth.

Could they have done it? It’s okay if you think a few guys already predicted it and no one else believed them, we can accommodate your scifi experiment too! We could freeze time and find the moment these guys made their prediction, and place them in a room, individually, iteratively, with every other meaningful investor and/or economist to allow them to make their case.

 

Instead we are trying to find a way to approximate this ideal scifi experiment. I think it’s worth doing, it’s fun. I’m not sure why, but humans enjoy it. Plus, we will only get better as we try. Now though we are dishonest about how successful we are, and as a result try to learn more from grand narratives of the past than is possible, while often ignoring the smaller cleaner lessons of human behavior when they contradict the grand historical narratives.

 

Leave a Reply

Your email address will not be published. Required fields are marked *