CS 6795 · Cognitive Science · Georgia Tech
Investigating how coherent collective behavior emerges from populations of boundedly rational agents interacting through small-world networks — using empirical human decision data as a cognitive substrate.
Small increases in communicator presence produce sharp, large-scale shifts in coherence — a clear phase transition boundary in parameter space.
Democratic population archetypes achieved stable coherence at lower communication loads than polarized or hyperconnected environments.
All metrics cycled with a ~1,000-round period before converging — a collective exploration-exploitation phase preceding consensus.
Populations with at least 75% oracle agents achieved near-perfect coherence across all connectivities. As heuristic dominance increased, coherence dropped sharply.
Very dense networks can reduce coherence — simple update rules amplify noise rather than attenuate it when social saturation is reached.
Responsiveness to communication (listen ratio) had a stronger influence on tipping thresholds than network rewiring probability or topology randomness.
Figure 1
Time series of entropy, Gini, oracle alignment, and coherence across 5,000 timesteps. Oscillations with ~1,000-round period precede convergence.
Figure 2
Watts-Strogatz small-world topology (k=6, p=0.12). Local clustering combined with long-range shortcuts enables rapid belief diffusion.
Figure 3
Tipping surfaces showing critical communication ratio vs. connectivity, as a function of graph size, randomness, listen ratio, and max rounds.
Figure 5
Tipping ridge comparison across four population archetypes. Democratic populations maintain lower thresholds across most connectivity levels.
Schematic recreations. Export real figures from the simulation scripts to replace these.
This paper is shared privately due to academic publication guidelines. Reach out and I'll send you a copy directly.