FABLES: Framework for Autonomous Behaviour-rich Language-driven Emotion-enabled Synthetic populations : modelling autonomous emotional AI-driven agents in their spatiotemporal context

The research presented in this Technical Report investigates how large language models (LLMs), through their extensive training and transcend linguistic capabilities, emerge as reservoirs of a vast array of human experiences, behaviours, and emotions. Building upon prior work of the JRC on synthetic populations (Hradec et al., 2022) it presents a complete step-by-step guide on how to use LLMs to create highly realistic modelling scenarios and complex societies of autonomous emotional Artificial Intelligence agents (AI agents). An AI agent is defined as a program that employs artificial intelligence techniques to perform tasks that typically require human-like intelligence (Ruan et al., 2023). Our technique is aligned with agent-based modelling (ABM) and facilitates quantitative evaluation. The report describes how the agents of a small subset of the existing synthetic population generated by Hradec and colleagues (2022) were instantiated using LLMs and enriched with personality traits using the ABC-EBDI, which combines the psychotherapeutic model ABC (Activation, Belief, Consequence) with the EBDI (Emotions, Belief, Desire, Intent). These intelligent agents were then equipped with short- and long-term memory, access to detailed knowledge of their environment, as well as the use of tools such as "mobile phone with a contact list" and the possibility to call friends and "public services". We found that this setting of embodied reasoning (Huang et al 2023) significantly improved the agents' problem-solving capabilities. Hence, when subjected to various scenarios, such as simulated natural disasters, the LLM-driven agents exhibited behaviours mirroring human-like reasoning and emotions, inter-agent patterns and realistic conversations, including elements that reflect critical thinking. The study shows how these LLM-driven agents can serve as believable proxies for human behaviour in simulated environments, which has vast implications for future research and policy applications, including impact assessment of different policy scenarios. The next level of implementation would cover a setting where all agents of the synthetic population have access to their complete environment, comprehensive network of contacts, functioning public services and an actual synthetic economy.