SimPolitics

Fenwick McKelvey's 2026 history of the six-decade project to model politics as a computing problem — from early election forecasting through Cold War world simulations — and what it reveals about why every generation keeps believing computers will fix democracy. The book traces two parallel tracks (SimVoters and SimWorlds) that share a common origin in what McKelvey calls the "computational imaginary of politics": the assumption that political reality can be rendered legible, predictable, and optimizable through computation. Published open-access by MIT Press, it arrives at exactly the moment when AI companies are reprising these same promises with newer hardware.

Key Themes

The Computational Imaginary of Politics

McKelvey's central concept is that we've been promised computer-fixed politics for over sixty years — and the promise itself, not the technical reality, is what shapes political practice. The book opens with a chapter titled "From 480 to 538," tracing how the very idea that politics could be computed predates the computers that attempted it. This is the book's most exportable insight: the imaginary precedes and outlives the technology. Every generation of political technologists — 1960s systems analysts, 1980s campaign consultants, 2020s AI companies — believes they're the ones who'll finally make it work, while the previous attempts were just crude prototypes. McKelvey's historical sweep makes this pattern visible in a way that punctures the current AI-for-governance hype without dismissing the genuine intellectual project.

Two Tracks, One Logic

The book's structure mirrors its argument. Part I (SimVoters) covers domestic electoral modeling: political forecasting, the "modeled American voter," and how campaigns learned to address voters as data points. Part II (SimWorlds) covers international relations simulation: war games, world models, and the Rand Corporation tradition of make-believe on mathematical principles. McKelvey's key move is showing that these weren't separate projects — they shared personnel, methods, and the foundational assumption that politics was a system amenable to computational modeling. The campaign consultant targeting swing voters and the Pentagon analyst modeling nuclear escalation were doing the same thing with different data.

The Unfulfilled Promise as Structural Feature

The book ends with "Let X = X or the Legacies of a Computational Imaginary of Politics" and a chapter asking "Are Other Computer Simulations Possible?" This isn't a story of steady progress — it's a story of recurring failure that somehow sustains itself. McKelvey isn't anti-computation; he's anti-amnesia. The book's value is that it refuses both techno-optimism ("this time it's different") and techno-pessimism ("computers can never help") in favor of historical specificity about what actually happened when people tried.

Critical Analysis

McKelvey is doing something harder than it looks. The obvious version of this book is a polemic: tech bros keep trying to optimize democracy and it keeps not working. The actual book is a history of earnest attempts by people who genuinely believed computation could improve political outcomes — and whose work genuinely did reshape how campaigns are run and how international relations are modeled. The critique isn't that they were wrong to try; it's that the computational imaginary consistently overpromises and underdelivers in predictable ways that we keep forgetting.

The book's temporal scope (1960s–1980s) is both a strength and a limitation. Ending before the internet era means McKelvey can treat the pre-digital simulation tradition as a completed arc, with clear patterns visible. But it means the book doesn't directly address the most recent iterations: Cambridge Analytica, AI-generated political content, or the current wave of LLM-powered governance tools. The final chapter gestures at these connections, but the reader has to do the work of mapping the historical pattern onto the present. That mapping is the whole point of reading the book, but it's worth knowing that McKelvey provides the historical substrate rather than the contemporary analysis.

The strongest thread is the connection McKelvey draws between SimVoters and SimWorlds. The same people who modeled voter behavior also modeled nuclear war, and the methodological assumptions traveled in both directions. This is the book's most original contribution: showing that domestic political computation and international relations computation aren't just analogous — they're genealogically linked. The campaign microtargeting database and the Cold War simulation share a common ancestor in the systems-analysis tradition of the 1960s.

The book's weakness is that it can't quite decide whether the computational imaginary is inherently flawed or just historically premature. McKelvey wants to leave room for the possibility that better models could actually improve politics — the final chapter's title is a question, not a verdict — but the weight of the historical evidence he presents points toward structural limits. Politics involves strategic actors who adapt to models, value conflicts that can't be optimized away, and outcomes that resist quantification. These aren't bugs that better compute will fix; they're features of the domain. The book would be stronger if it said this outright rather than leaving it implicit.

This connects directly to Science and Statistics (Box): "all models are wrong" applies doubly when the modeled subjects are strategic actors who can change their behavior in response to being modeled. The political simulation faces a reflexivity problem that the crop-yield model doesn't. And it connects to Man-Computer Symbiosis: Licklider's vision of genuine interdependence between human judgment and machine computation is a better framework than the computational imaginary's assumption that politics can be solved rather than assisted.

Connections

The book's historical arc — people in the 1960s promising that computers would revolutionize politics — rhymes with current claims about AI and governance. The Education of the Broligarchy documents the intellectual formation of the people now making those claims; SimPolitics provides the prehistory of the claims themselves. The pattern McKelvey identifies — that the computational imaginary outlives every failed instantiation of it — is visible in America Is Slow-Walking Into a Polymarket Disaster, where prediction markets inherit the same promise of computational political truth despite a track record that doesn't support it.

The methodological critique connects to Not-Knowing (Vaughn Tan): political systems involve genuine uncertainty (unknowable unknowns) that resists the risk-modeling tools McKelvey's historical actors tried to apply. And the book's implicit question — what would a genuinely useful computational political assistant look like? — connects to Man-Computer Symbiosis's vision of computers that help humans formulate problems, not just solve them.

Goldman Sachs World Model represents the current state of the SimWorlds tradition: the ambition to model global reality computationally, now backed by orders of magnitude more compute. 2026 Global Intelligence Crisis is a case study in what happens when the model diverges from the modeled. And The Market for Doom documents the recurring pattern of technological unemployment predictions — a structurally similar phenomenon where the prediction outlives its failures.

On the domestic side, the SimVoters tradition has evolved into the data infrastructure of modern campaigns — a direct line from McKelvey's "modeled American voter" to contemporary microtargeting. The Dead Economy Theory and Why Agents Matter More Than Other AI explore adjacent territory: what happens when the logic of computational optimization extends beyond prediction into replacing human participants entirely.

A contemporary inflection of McKelvey's thesis: Tom Barraclough — Sovereign AI Policy argues the computational imaginary of law is the next frontier — turning regulation from PDFs into structured, versioned datasets that AI can reason over. Same pattern (render governance computable), new substrate (law, not voters).

And the cleanest counterexample to McKelvey's closing question is IO Factory — a 2026 simulation of AI-enabled influence campaigns that inverts the SimVoters/SimWorlds goal entirely. Instead of modeling politics to predict or optimize outcomes, it models the manipulation of politics to red-team and defend against it. Where the computational imaginary always aimed at making reality legible to the modeler, IO Factory aims at making the campaign legible to a defender — an "other computer simulation" that finally treats the modeled subjects as the point rather than the substrate.

Sources: raw/simpolitics Last updated: 2026-07-11