Felix Temel

Modelling Human Decision-Making Using Large Language Models

In my PhD research I investigate how Large Language Models (LLMs) can be integrated into Agent-Based Models (ABMs) to simulate human decision-making more realistically. Classic ABMs rely on pre-defined behavioural rules, which makes it difficult to capture factors such as personality traits, context-dependent behaviour, memory, biases, and others, especially when these factors change over time. Since LLMs generate behaviour from prompts rather than fixed rules, they can incorporate such factors into an agent's decisions and allow behaviour to emerge that was not explicitly programmed.

My goal is to develop a comprehensive framework for LLM-ABM integration, apply it to a specific model, and validate the results against real-world data. This should advance LLM-ABM modelling as a method and open up new tools across various fields. It may also help policy makers model populations more accurately, informing decisions particularly in areas such as economic and evacuation planning.