Political simulation during the Cold War

Political simulation was invented in a Santa Monica office in the 1950s. It worked. It could never scale. The economics have now changed.

In the mid-1950s a group of RAND Corporation analysts began doing something that looked, from the outside, like an eccentric use of US Air Force money. They played games. Herbert Goldhamer and Hans Speier ran exercises in which teams of analysts took the roles of governments. One team played Washington, another played Moscow, and a control team played reality itself, ruling on which moves were plausible. The question was whether acting out an adversary's position produced insight that reading about it did not.

It did. Participants kept reporting the same discovery: playing the other side exposed assumptions nobody knew they held, and generated moves that no planning paper had listed. Goldhamer and Speier wrote up the method in 1959 and were careful, in a way that still reads well today, about its limits. Games produced hypotheses and disciplined thinking. They did not produce predictions you could take to the bank.

The approach spread through the strategic community. Thomas Schelling ran crisis games for the Pentagon in the early 1960s, including exercises on Berlin at the height of the standoff, and found that a crisis unfolds differently when experienced officials must actually choose, move by move, rather than follow their own doctrine on paper. Herman Kahn turned scenario thinking into a discipline, insisting the unthinkable be worked through in specifics. Olaf Helmer and Norman Dalkey, also at RAND, built the Delphi method so that expert judgement could be aggregated into forecasts without the loudest voice in the room winning by default.

The constraint that never lifted

Everything about the method was expensive. A serious political game consumed weeks of preparation, the calendars of senior people and a room full of scarce expertise, and it produced exactly one run-through. There were no counterfactuals. You could not rerun Tuesday's game with a different opening move to see what changed, and the results depended heavily on who happened to be in the room.

So simulation stayed rationed. It was reserved for existential questions, nuclear crises and superpower confrontation, with the occasional corporate outpost. Shell's scenario planners under Pierre Wack famously readied the company for the 1973 oil shock, and in doing so proved mainly how artisanal the craft remained. For everyone else, the method that worked was simply unaffordable.

What changed

Two things. First, evidence about public actors now exists at depth and is retrievable: decades of votes, speeches, interviews, rulings, filings and behaviour under pressure. Second, modern language models make it possible to hold that evidence together as a persistent, interrogable model of a named actor rather than a filing cabinet about them.

Put those together with discipline and you can do what Santa Monica could not. Build the players once, from evidence, and keep them current. Then run the game as many times as the question deserves. A negotiation can be rehearsed forty times before the real meeting. A policy announcement can be tested against every plausible ministerial response rather than the two the room thought of. A crisis can be played to the end, then played again with one variable moved.

This is the lineage Nostrada sits in. Our twins are the players. The scenario engine is the game room. The register, where every forecast is logged with a probability and scored against reality, is the scorekeeping the Cold War gamers never had the volume to attempt.

The obligation that comes with it

The inheritance imposes standards. What kept RAND's method respectable was candour about what it could show. We hold the same line. Twins are bounded by evidence, and where the record is thin they say so rather than improvising. Every twin carries a fidelity grade. Every forecast is scored.

Goldhamer and Speier would recognise the caution. The replication they could only have envied.

The questions have not changed since 1955. What will they do. How confident are we. What should we do about it. The difference is that the method built to answer them no longer requires a fortnight of everyone's time and a room in Santa Monica.