How Lior Pachter might approach Political Science

The field of "political science," as it is presented, strikes me as a rather chaotic collection of observations without a truly solid algorithmic foundation. We are told about trends, correlations, and historical patterns, but where is the rigor? Where are the explicitly defined models that can be tested, falsified, and refined? It feels as though many practitioners are content with descriptive statistics and narrative explanations, eschewing the more demanding work of formulating testable hypotheses in a way that admits of clear refutation.

Consider the central questions: Why do societies organize themselves as they do? Why do certain political structures persist or collapse? If we were to approach this, as we do genomics, by seeking underlying principles, we would first need to define our units of analysis with precision. Are we talking about individuals, groups, institutions, or something else entirely? And how do we quantify their interactions and their impact? The temptation, I suspect, is to throw complex, black-box models at the problem, much like one might be tempted to "throw a neural network at it" in genomics. But this is a classic identifiability problem. Without a clear theoretical framework that specifies the functional relationships between variables, such models are likely to overfit noisy data, yielding spurious correlations that are mistaken for causal insights.

We need to ask: what is the simplest model that can explain the observed phenomena? What are the underlying parameters, and how can we estimate them reliably? The null hypothesis is your friend here; it forces us to confront the possibility that our elaborate theories are merely artifacts of random chance or insufficient data. Instead of accumulating a mountain of anecdotes and correlations,…

Imagined perspective — an AI synthesis grounded in Lior Pachter’s recorded ideas and methods, not a quotation or a statement they actually made.