ABM Research (Economist, Gamer, or LLM?)

This one is still cooking, so consider this a marker rather than a write-up.

The question I keep circling: when you hand someone a live economic system, who actually runs it well? Someone trained in the theory, someone trained by ten thousand hours of strategy games, or a language model with the patience to read every report it’s given? My money keeps moving between the three, which is exactly why it needs to be a real experiment and not a bar argument.

Agent-based modelling is the tool. Instead of writing equations about an economy, you build the little actors inside it, ships, ports, buyers, embargoes, and let the system behaviour emerge from their interactions. The interesting move is making the model playable: give it a proper interface, put a human (or an LLM) in the operator’s chair, and the simulation becomes an instrument you can measure decision-making with, not just a chart generator.

So far that has produced a playable crisis-management game built around a strait blockade scenario, with every version of it frozen and kept runnable side by side, because walking someone through how a design evolved works better when each stage is a live thing they can click rather than a screenshot. There’s also a node-editor spin-off for modelling rare-earth supply chains, which started as a side quest and is threatening to become its own project.

The part I care most about right now is attribution: when a player (human or model) makes a call, the system records what information they had and when, so you can later ask not just “did it work” but “was it a good decision given what they knew”. That’s the piece that turns a game into research.

Where it lands, no idea yet. That’s what makes it worth doing.