Myopic Delirium
Patch notes
All entries in Public PatchesWe are a research group studying the distance between what people intend and what people, together, produce.
Using agent-based modeling, we specify how one individual behaves, what it needs, what it can perceive, what it does about it, and then run many of them at once and observe the society that results. The durable finding of the field, going back to Thomas Schelling, is that the society is usually not a scaled-up version of anyone’s intention. Mild individual preferences produce extreme collective outcomes. Nobody chooses the result, and everybody builds it.
The work here starts from ordinary survival with agents in a landscape with water, food, heat, and predators and builds toward agents with the constraints that actually characterize human beings: memory that distorts what it stores, attention too narrow to hold everything that matters, needs that only become prioritized once more basic ones satisfied. The aim is not to show that some phenomenon can emerge, but to model that phenomenon carefully enough that its accompaniments become visible. That includes the costs, dependencies, and side effects that travel with it and that we routinely fail to notice in ourselves. This site holds the work we can safely make public.
Every law is a hypothesis about human behaviour, executed at full scale, on a live population, with no control group and no clean way back.
The effects of an intervention on a generation emerge over a decade, by which point the decision, the government that made it and the conditions it addressed have all changed. Every institution is steering on evidence produced by a world that no longer exists. Our work begins below the level of the trend, at the point where a person decides what to attend to and what to disregard. Populations built this way can be run forward under conditions that have not yet arrived, and run again with the assumptions changed.
We began this work in 2024. The starting point was an observation that standardised metrics had stopped serving as proxies for how a society might improve and had begun to function as statements of what a person is worth. The early models were deliberately crude with agents with little more than the drive to persist. We added capacity gradually, not to reproduce human experience but to find out what would emerge once agency amounted to more than optimisation, whether agents would come to care, to be irrational, to confuse themselves, and whether any of it would produce commitments that outranked their most base interests. Institutions rarely fail because they misread how a population leans but because its priorities moved beneath the instruments used to measure them. Populations of several thousand agents, and tens of thousands on high-performance infrastructure, let us examine not what a policy does to an aggregate, but how it rearranges the pressures, options and priorities facing each person inside it.