Black Leopard Red Wolf

By Marlon James

Made it about 20 pages, I couldn’t really follow the narrative structure, so I bailed.

Nothing is True and Everything is Possible

By Peter Pomerantsev

Part memoir/part essay on modern Russia under Putin by journalist/writer/documentary filmmaker Peter Pomerantsev.

Pomerantsev, raised in England by Soviet exiles/expats, moves to Moscow near the beginning of the Putin era to work for TNT (the Russian TV network, not the US one) on various documentary and reality TV shows.

The title actually captures the ethos of modern Russia (from Pomerantsev’s perspective rather well). There is a tremendous amount of optimism on the part of the Russian people, but the Putin regime has created an information environment in which nothing can ever be trusted.

This is particularly unnerving given similar trends in the United States. It’s a model of authoritarianism that doesn’t try and control all information like totalitarian states of the past. Instead, the whole concept of truth itself is undermined so that nobody can trust anything. The result is a sort of political inertia which keeps the current power structure in place.

The Book of Why: The New Science of Cause and Effect

By Judea Pearl and Dana Mackenzie

The one is a trade book written by Judea Pearl (along with ghostwriter Dana Mackenzie), UCLA professor of Computer Science and Turing award winner about his work on the mathematical foundations of causality.

What do we mean by causality? Just this initial question is a bit of a fiendish one to answer rigorously. Pearl goes back, as most do, to Hume who defined causality in two different ways even though it’s not exactly clear he considered them separate definitions. The first is just correlational. If B always happens after A then it suggests (but can never prove!) that A in fact causes B. The second is related to counterfactual reasoning. To say that A causes B is to say that if A had not happened then B would not have happened.

Pearl gives an overview of the long and tortured history of mainstream statistics trying its best to avoid talking about causality at all. The discipline only wants to talk about correlations and goes through some epic contortions to try and frame questions we would like to ask of the world in ways that do not rely on causality as a valid concept.

Pearl’s approach is in direct contradiction to the traditional approach of statistics. He want to create a mathematical framework that allows us to ascend what he calls the “ladder of causation, ” which has three rungs:

1. On the bottom rung we have correlations in observed data.
2. On the next rung up we have the causal effects of interventions. That is, not just the conditional probability P(A given B) but the conditional probability P(A given do(B)) which represents the conditional probability of A given that B is done as an intentional intervention as opposed to just something happening prior to A.
3. On the highest rung then we have counterfactual reasoning. That is, what would have happened to A if B had NOT happened.

The academic work that Pearl has done over the years has produced a few useful results that allow us (given certain preconditions discussed below) to “climb” the ladder of causation so they we can make rigorous causal claims based only on observational data or at least understand rigorously when we can only make causal claims based on randomized controlled experiments.

Specifically, Pearl’s work provides a couple of important tools:

1. The “backdoor criterion” which, given a well defined causal model, can give us a rigorous way of identifying confounding variables.
2. The so called do-Calculus which is a set of mathematical tools for reformulating conditional probabilities with do operators such P(A given do(B)) as conditional probabilities without do operators. That is, restate conditional probabilities that reflect the causal effect of a specific intervention as conditional probabilities derived entirely from observed data.
3. A rigorous grounding for mediation analysis. Essentially, a way to understand how causal mechanisms flow through both direct and indirect (or mediated) mechanisms. Using these tools we can characterize the size of direct and indirect effects and ultimately answer in a rigorous way counterfactual questions.

Of course this all sounds too good to be true. He is claiming to have solved deep philosophical issues that have vexed science and mathematics since humans have started thinking about these things. And you would be right to think so.

Where this falls somewhat flat for me is that it is all predicated on having a correct causal model of the system you intent to analyze. So there is an element of circularity to this whole thing. We are still ultimately left with a fundamental model uncertainty. That is, given a causal model we can make do a lot of wonderful things with the mathematical tools Pearl and his collaborators have developed over the years, but how do we know what the causal model is to begin with?

Of course, that is a bit harsh. Part of the pitch Pearl is making is that we usually have a causal model in our mind when thinking about scientific questions. We should be explicit about it and create an actual causal model. Then using that model we can derive hypotheses which can be validated against data and experiments in a rigorous way. So we can falsify causal models but we can’t derive them so to speak.