Research & Publications
My research sits at the intersection of AI, social systems, and human behavior. I use computational methods — behavioral modeling, network analysis, machine learning, and online experiments — to understand and improve how algorithms shape society.
View full list on Google ScholarSelected Publications
Can adversarial attacks by large language models be attributed?
Demonstrates that attributing AI-generated content to a specific model is computationally intractable: the hypothesis space doubles every ~0.5 years, and attributing one year of U.S. AI output would require 200 years of non-stop supercomputing.
Fairness in LLM-Generated Surveys
LLMs excel at simulating macro-level social patterns but consistently show a U.S.-centric bias and significant fairness disparities across gender, education, and political identity when applied to Chilean populations.
Impact of Price Inflation on Algorithmic Collusion through Reinforcement Learning Agents
Shows that RL-based pricing agents can sustain supra-competitive prices and amplify collusive behavior under inflationary shocks — raising new concerns about autonomous pricing in economic markets.
Zero-shot Decision Tree Construction via Large Language Models
Proposes a zero-shot method for building interpretable decision trees using LLMs — no labeled training data required — enabling explainable classification in low-resource settings.
Networks Multiscale Entropy Analysis
Develops a multiscale entropy framework for analyzing the structural complexity of networks across different organizational scales, with applications to social and biological networks.
Large Language Models in Crisis Informatics for Zero and Few-Shot Classification
Evaluates LLMs for crisis event classification under zero and few-shot conditions, demonstrating competitive performance against supervised baselines for multilingual disaster response applications.
Measuring the Predictability of Recommender Systems using Structural Complexity Metrics
Introduces data-driven metrics that use SVD to determine an algorithm's maximum achievable precision, showing >0.90 correlation with top-tier algorithm performance.
Price of Anarchy in Algorithmic Matching of Romantic Partners
Borrows the Price-of-Anarchy concept from game theory to quantify how self-interest in online dating algorithms reduces social efficiency — and shows that market competition aligns agent incentives with user welfare.
Modularity of Food-Sharing Networks Minimises the Risk for Individual and Group Starvation
Shows that modular structure in food-sharing networks among hunter-gatherer societies reduces starvation risk by buffering local shortfalls — balancing within-group solidarity with inter-group independence.
- Published Auditing algorithmic bias on TwitterAI & Society
- AI & Society
- AI & Society
- Network Science
- AI & Society
- Published Assortment optimization under a multinomial logit model with position bias and social influenceMachine Learning