This is an event-based simulation of a hypothetical nuclear conflict between the United States and China. The purpose of the simulation is to model how advancements in AI, specifically large scale visual recognition using neural networks, might affect the nuclear balance of power.
This simulation was programmed by Torin Rudeen, based on data and models provided by Ben Chang, as part of Ben Chang's doctoral dissertation at MIT. You can read the dissertation paper based on this simulation here (pages 70-125).
To give some brief context, China's nuclear deterrence strategy is based around having a large number of mobile missile launchers, known as TELs, which are easy to move around the country and hide. Thus, anyone who wished to launch a nuclear first strike against China would be dissuaded, because it would not be feasible to accurately track the location of every single TEL which could possible launch a counterattack.
This simulation attempts to see whether this calculus could be changed by the deployment of large-scale image recognition, which may make it possible to find TELs in close to real time by analyzing satellite imagery and other data.
The simulation is coded in Python, using iPython and Pip to make it easy to reproduce and experiment with the results.
The simulation uses an event-based approach, where each event can trigger further events to happen at a future timestamp. For example, when a missile is launched, the simulation calculates how long it will take to arrive at its target, and enqueues a future event for the missile's impact. It also calculates how long the launcher will need to reload, and enqueues a future event for when the launcher will be able to fire again.
There are several versions of the simulation in different subdirectories. A good entry
point for exploring each one is lib/simulation.py, which holds the core simulation loop
and initializes all of the other entities.