Modeling the AI-science bottleneck

A simulation of how uneven AI acceleration across scientific fields can slow discovery, and how resource allocation across science can affect the outcome. Interactive companion to “Our grandchildren’s AI-science bottleneck.” Read the full article → This simulation models scientific progress as a stochastic process on a directed graph of discoveries and inventions. Each node i is a latent discovery; nodes with inbound edges to i represent previous discoveries that facilitate i (parents or prerequisites), and nodes that receive i’s outbound edges represent discoveries facilitated by it (children). Nodes are assigned scientific fields, and are preferentially connected based on field, recency, and fitness. Atop the latent network structure of discoveries, I run a Monte-Carlo simulation of the discovery process. In this process, nodes become “available” when all prerequisites have been found, and available nodes are “discovered” ba

saved by

• Davis Brown