Noah Egan

Cultural similarity and city size

Do people in larger cities become more culturally similar to each other, or less? Using web browsing data (Comscore, 280K users) and video game libraries (Steam, 368K users) across ~370 US metro areas, I measure how similarity between individuals scales with city size. In larger cities, people are more alike in the popular websites and games they use and less alike in niche ones. To explain the split, I built a birth-death Markov model of social learning and independent discovery. Variants that spread easily enough reach a larger share of people as populations grow, while the rest become less likely to be widely shared. I estimated how transmissible each of ~50K websites and games is by maximum likelihood, and the model correctly predicts which ones converge and which diverge in larger cities. Different cultural domains may look different simply because they contain different mixes of these two kinds of variants.

Superhuman AI and strategy change in chess and Go

How much do top players change their opening moves from year to year, and what happened when superhuman AI arrived? Across 50 years of tournament records (3.45M chess and 116K Go games), I represent each player's style as a set of conditional move distributions and measure how much it changes each year with Hellinger distance. I then derived a decomposition that splits this change into a part shared across the population and a part specific to each player. After AlphaGo, Go players changed together, adopting the same new moves. In chess, change has risen since the late 2000s, but it is driven mostly by players exploring on their own.

Self-assembly in fire ant pontoon bridges

With the Goldman and Randall labs at Georgia Tech, I studied how fire ants build floating bridges out of their own bodies to reach food across water. Colonies grow bridges in two ways: ants join the tip one at a time, or launch into the water, form rafts, and attach later. I built a computer vision pipeline to measure bridge growth from video and an agent-based model built from measured individual behavior. The model reproduced the experiments and correctly predicted what colonies would do in three new setups. I also derived a simple growth model that explains why bridge growth curves look the same across colonies once time is rescaled, and helped prove that a Markov chain model of bridging forms a single bridge.

DISC paper

Coexistence under higher-order interactions

At Princeton I built a spatial Lotka-Volterra model, a PDE approximation, and stochastic simulations in R, and found coexistence regimes that only appear when species have higher-order interactions.