Articulating complexity.

Across domains (biodiversity, climate, public health, national security) there is no shortage of beliefs about what ought to work. Yet policies routinely produce outcomes that diverge from their stated objectives, and the regularity with which this happens suggests that the failures are not incidental but structural.

The reason is that these are not technical problems. They are complex adaptive systems: systems whose components interact, learn and adjust their behaviour in response to one another. Social-ecological systems are a prominent instance, in which human agents with beliefs, strategies and competing interests are inseparable from the systems they govern. Anticipating how a policy will fare requires modelling how strategic actors will respond to it, including their capacity to deceive, negotiate, betray promises and revise their plans in light of new information.

Articulating Complexity is built on the premise that a new generation of tools can articulate this complexity, anticipate failure, and help design more resilient interventions.

Until very recently, no method existed to model the strategic, adaptive behaviour of real-world actors at scale. That has changed. The state of the art in large language models, multi-agent systems and computational capacity has moved dramatically over the past two years.

LLMs can now convincingly simulate institutional decision-making, replicate cross-cultural personality differences, sustain strategic reasoning across long interactions, and impersonate the rhetorical patterns of public figures. The window in which these capabilities can be turned into rigorous, transparent scientific instruments, rather than opaque commercial products, is narrow. This programme is an attempt to seize it.