I: History As Ingression

An addendum to the last post:

Some other readings that illustrate different approaches rooted in processes with long-range dependence in both literal (statistical) and metaphorical time series: 

David Krakauer, "The Stargazer and the Flesh-Eater," 2011.

Jack McDonald, "The Ingression Engine," 2013.

Samuel Arbesman, "Stop Hyping Big Data and Start Paying Attention to Long Data," 2012. 

Peter Turchin, "An Imperfect Time Machine," 2013

David Christian, "A Single Historical Continuum," 2011

All of these have in common the idea that more continuous data as well as more advanced methods (for example, carbon-dating as Christian cites it) allow for a distinctly different approach to history. The most explicitly computational aspects of this are captured by McDonald (discussing surveillance with big data elements): 

The amount of information available to a potential ‘snooper’ is independent of the timing of the act of surveillance. A kid born today, where intelligence agencies hoover up this kind of info lives forever to the right of the “Metadata collection begins” point on the graph. Governments aren’t necessarily surveilling everyone, but they’re building the datasets required to ingress into anyone’s history, back to the earliest point of metadata collection, whenever they are interested. How does one control this? The common option is tied to the use of “surveillance” as a concept: stop the government from collecting any data. That’s quite unlikely, I think. What interests me is that we have no control over the temporal limits of metadata collection (how far back records go), nor do we have any control (realistically) over deletion of metadata. We only have trust. I trust Google (perhaps stupidly) to delete data when I ask them to, but who trusts an intelligence agency to do the same?

History perceived here is simulation, but of a strikingly different kind than Taleb's Monte Carlo idea. Rather, a history on the lines of what McDonald is actually akin to old fashioned animation techniques. A child with a flipbook thumbs through a large set of static pictures at many different points. Other examples include time-lapse photography, some of which have made interesting YouTube videos. 

We are still in the realm of "simulation," but less in the sense of a generic computer simulation and more of Francis Bacon's original definition of simulation as display of an imitation and dissimulation as concealment of the truth, which Baudrillard alludes to here: 

To dissimulate is to feign not to have what one has. To simulate is to feign to have what one hasn't. One implies a presence, the other an absence. But the matter is more complicated, since to simulate is not simply to feign: "Someone who feigns an illness can simply go to bed and pretend he is ill. Someone who simulates an illness produces in himself some of the symptoms" (Littre). Thus, feigning or dissimulating leaves the reality principle intact: the difference is always clear, it is only masked; whereas simulation threatens the difference between "true" and "false", between "real" and "imaginary".

Of course, even in a time-lapse video you are only focusing on one narrow aspect of a system. That way it has narrative and perhaps explanatory power. One critique of Braudel's style of history I find persuasive is Charles Tilly's note that it lacks focus, direction, or the ability to really come to any useful conclusion. 

II: Computational Ethics

Baudrillard's argument is actually fairly old: from philosophers to quack psychologists who burned comic books, intellectuals have been suspicious of the idea of artificial or synthetic things. Much of the post-modernist critique of simulation (particularly when it comes to war) is hyperbolic and fearmongering in the extreme. All good modelers understand that the simulation is not, and can never be reality. Moreover, there is an actual scientific explanation for both good and bad effects of artificial objects throughout human history. 

There is, however, an enduring problem with the validation of models. To get a bit of this, we need to think more narrowly about what a model is. 

Miller and Page nicely formalize this by arguing that the model is an equivalence class of reality, with a real world system S and the model as s, with s=E(S). Model system s's state at t+1 is given by f(s). Real world system, in turn, is given by F(S). Of course, while S has a bijective mapping to its future state at the next time step, S only partially maps to the model system s. 

In all models, we're looking to track some property of interest. Those things should be the elements of S that actually do map to s, as s is a subset of S. We want our subset, which only contains said elements of interest, to "coincide with reality," as Page points out. There are, of course, vast difficulties with responsibly abstracting, particularly with models that utilize relatively new methods.

It was only after starting a long journey as a modeler (via my PhD program this semester) that I really finally "got" the overarching point of Taleb's oeuvre. Responsibly abstracting is an ethical and moral component of simulation of any kind that is not typically addressed---both out of ignorance (Taleb's targets did not understand the famous philosophical critique of induction) or greed (incentives to "simulate"--again, in Bacon and Baudrillard's use of the word simulation to mean faking or artifacting). 

Where I find the overriding, centuries-long "simulation is bad" critique lacking is that modeling as a whole is a basic cognitive function that we all engage in: 

The first question that arises frequently—sometimes innocently and sometimes not—is simply, "Why model?" Imagining a rhetorical (non-innocent) inquisitor, my favorite retort is, "You are a modeler." Anyone who ventures a projection, or imagines how a social dynamic—an epidemic, war, or migration—would unfold is running some model. ....But typically, it is an implicit model in which the assumptions are hidden, their internal consistency is untested, their logical consequences are unknown, and their relation to data is unknown. But, when you close your eyes and imagine an epidemic spreading, or any other social dynamic, you are running some model or other. It is just an implicit model that you haven't written down. .....

The choice, then, is not whether to build models; it's whether to build explicit ones. In explicit models, assumptions are laid out in detail, so we can study exactly what they entail. On these assumptions, this sort of thing happens. When you alter the assumptions that is what happens. By writing explicit models, you let others replicate your results.

What makes an explicit model or simulation better is also the possibility of sensitivity analysis: 

Another advantage of explicit models is the feasibility of sensitivity analysis. One can sweep a huge range of parameters over a vast range of possible scenarios to identify the most salient uncertainties, regions of robustness, and important thresholds. I don't see how to do that with an implicit mental model. It is important to note that in the policy sphere (if not in particle physics) models do not obviate the need for judgment. However, by revealing tradeoffs, uncertainties, and sensitivities, models can discipline the dialogue about options and make unavoidable judgments more considered.

Ironically, many critics of simulation themselves dissimulate (again quoting Bacon) by hiding the assumptions of their conceptual, implicit models. This perhaps leads into why Taleb writes with such urgency and also explains the bitterness with which he attacks his targets: he has seen from his time in the financial sector that to hide ones limits and assumptions is, in fact, a competitive advantage.