Thursday, August 27, 2015

Never Trust a Historian to Predict the Future

George Santayana, the Spanish-born and American-educated naturalist philosopher, had a aphorism that has been quoted to death: “Those who do not remember the past are condemned to repeat it.” This quote is a criticism of people who fail to use history to understand the present and unavoidably fall into the same pitfalls as others have before them. It is based on the observation that people have not changed over the last few centuries, as regards their veracity, ambition, greed, and tactics, so not learning about the history of recent events sets up someone to be victimized as others were. P.T. Barnum had a saying with a similar core: “There’s a sucker born every minute.” Santayana was alluding to political leaders and other important social figures, while Barnum was talking about ordinary people. However, the point is the same. Gullible people abound at all levels.

Santayana was quite right in that ignoring history can, in some instances, lead to a pitfall. But the quote is overused and overemphasized. Emphasizing historical knowledge as a way to predict the future is not a good idea. The future is not the past.

Historians, or pseudo-historians, rather anyone who studies history, reads books about history, writes about history, thinks about history, and dreams about history is going to predict the future using what they know and love: history. Ask a historian what will happen in the future and the only option they have is to cleverly think through the past, the part they know, and find analogies and pick the closest one. More clever ones may combine two or more. Historians are endemically incapable of predicting novelty or something which has not happened before. They have learned how to understand the past, and they are bound to use that understanding to analyze the present and figure out which past event is going to recur, albeit with different faces, different times, different locales, and so on, but basically the same framework and sequence of events.

Many historians are quite eloquent and are able to express themselves very competently and are able to make arguments as to why their particular historical analogy is appropriate to the present and how the future will therefore unfold. Historians are no different than anyone else predicting the future in that they lace their commentary with caveats, and give themselves slack especially in questions of timing and duration, but they suffer from a much greater and deeper failing. History is their only card.

If something has features in it which make it different from anything in the past, then historians have the option of belittling it, downplaying it, ignoring it, or pretending that this novelty is just like a past novelty so the history of how the former novelty played out should be used to predict how the new novelty will unfold. They cannot see something completely new, or with completely new aspects. They look one way, backwards, and while their guessing may be right on in some instances, novelty is their kryptonite. They melt.

There is another type of historians who are not often so eloquent, but more numerate. These are the trend followers. Their idea is that whatever happens in the future is a project of what is going on now. They don’t use so much of the past as a historian, but they use a slice of the past to compute or otherwise deduce what trends are going on, and then project them. In numerate situations, straight-line projections are often used. Noise in the data is filtered out. This type of prediction is very useful if nothing is happening that would change a trend, but unfortunately it almost always is. We live in a changing world, and one that is tightly interwoven, so one thing changing makes another thing change. This is especially true in situations where people are making decisions, as a decision can be made in a split second, and then the trend line snaps.

Often, trend predictors are mathematically sophisticated, so they invent a model which captures how they believe a number of variables are interacting, and take some data to get correlations or other relationships between them. Sometimes the model is emphasized and sometimes the data. Always the number of features incorporated is limited, and often the data is not robust enough to establish the details of the relationships between variables, so it is hypothesized, and the hypothetical nature sometimes is underemphasized. The flaws in such an approach in a situation where external, unincorporated variables are undergoing significant change should be obvious. Modeling is a tool for exploring consequences, not a predictive method.

So, if the past cannot be used to predict the future when things are new, and the present cannot be used to predict the future when change is occurring, what is left? In our area of interest, alien civilizations, trying to find some past Earth situation to analogize is prima facie inappropriate. Projecting our rate of changes is going to fall far short. Nothing is left to predict what alien civilization will look like and what activities they might take, except three things.

One is normative thinking. This is looking at a situation and asking what physical, chemical, mathematical and other laws must govern it, and then what implications there are from these conditions and limitations. When talking about aliens, sometimes someone will say that, for example, the laws of physics will be overturned by alien scientists and so they will be able to do anything. The history of physics doesn’t indicate this, but we can simply go forward with the assumption that magical events, ones which do not follow the science we know, are not going to be considered.

The second one is to find uncoupling by time and length scales. Things which happen in seconds don’t interact well with things that happen in years. Averaging happens in the quick to long direction, and constancy happens in the long to quick direction. There are a number of time scales and length scales which are important to thinking about star travel and evolution and other key phenomena, and if uncoupling dominates some relationship, they can be thought of independently.

The third thing is parameterization. If we can’t figure out some variable, we can break up the range of possibilities into a few classes and look at each one independently. Usually results are not sensitive to the exact value of a variable, so a few classes are enough to generate insights about the situation where the parameter plays a role. Obviously this cannot be done with many variables simultaneously, so the insights discovered by exploring one parameter have to be used to decide on a most likely value. Then this tentative case can be explored more deeply.

I don’t want to imply that history is not a fascinating subject, it is. I don’t want to imply that modeling is not a challenging and important skill, it is. I do want to imply that these tools may not be very useful in understanding the answer to the blog’s original question: “Where are all the aliens?”

No comments:

Post a Comment