AI simulation of human irrationality

Are AI Simulations the Key to Understanding Human Irrationality?

Jordan Keane in Tech & Innovation December 2025 • 3 min read.

"New research explores how Large Language Models can mimic and model subrational human behaviors, offering insights into economics, psychology, and more."

For decades, researchers have struggled to create accurate models of human behavior, especially when it comes to irrationality. Traditional methods rely on reinforcement learning, which requires extensive data collection and complex calibration. Gathering data from human subjects is not only time-consuming but also raises ethical concerns.

Large Language Models (LLMs), like those powering advanced AI assistants, are changing the game. LLMs have demonstrated remarkable abilities in reasoning, problem-solving, and even mimicking human communication. This has led researchers to explore their potential as tools for simulating human behavior, potentially offering a new way to study subrationality.

A groundbreaking study investigates using LLMs to generate synthetic human demonstrations, which are then used to train AI agents to replicate subrational behaviors. The study explores whether LLMs can serve as implicit computational models of humans, capable of capturing quirks like myopic decision-making and risk aversion. This approach could revolutionize fields from economics to robotics, offering deeper insights into human conduct.

Why Model Subrational Behavior?

Traditional economic and AI models often assume perfect rationality – that individuals always make decisions that maximize their benefits. However, real-world behavior is far more complex. Humans are influenced by emotions, biases, and cognitive limitations that lead to seemingly irrational choices.

Understanding these subrational behaviors is crucial in various fields:

Economics: Modeling consumer behavior, understanding market trends, and predicting economic responses to policy changes.
Finance: Developing investment strategies, managing risk, and understanding investor psychology.
Robotics: Designing robots that can effectively collaborate with humans by anticipating their actions and understanding their limitations.
Public Policy: Crafting policies that account for human behavior and encourage desired outcomes.

By accurately modeling subrationality, we can gain a more realistic understanding of human decision-making and develop more effective strategies in various domains.

The Future of AI-Driven Behavioral Modeling

This research marks an exciting step toward using AI to understand the complexities of human behavior. By leveraging the power of LLMs, researchers can create more realistic and nuanced models of decision-making, paving the way for breakthroughs in economics, psychology, and beyond. While challenges remain, the potential benefits of this approach are vast, offering new insights into why we do what we do.

About this Article -

This article was crafted using a human-AI hybrid and collaborative approach. AI assisted our team with initial drafting, research insights, identifying key questions, and image generation. Our human editors guided topic selection, defined the angle, structured the content, ensured factual accuracy and relevance, refined the tone, and conducted thorough editing to deliver helpful, high-quality information.See our About page for more information.

This article is based on research published under:

DOI-LINK: https://doi.org/10.48550/arXiv.2402.08755,

Title: Llm-Driven Imitation Of Subrational Behavior : Illusion Or Reality?

Subject: cs.ai econ.gn q-fin.ec

Authors: Andrea Coletta, Kshama Dwarakanath, Penghang Liu, Svitlana Vyetrenko, Tucker Balch

Published: 13-02-2024

Everything You Need To Know

How can Large Language Models help us study human irrationality?

Large Language Models can simulate subrational behavior. By generating synthetic human demonstrations, these models can train AI agents to replicate quirks like myopic decision-making and risk aversion. This approach offers a new way to study why humans aren't always rational, providing valuable tools for researchers in various fields.

Why is it important to model subrational human behavior in fields like economics and finance?

Modeling subrational behavior is crucial because real-world human decisions are often influenced by emotions, biases, and cognitive limitations, deviating from perfect rationality. In economics, it helps in understanding consumer behavior and market trends. In finance, it assists in developing investment strategies and managing risk by accounting for investor psychology. Accurately modeling subrationality leads to a more realistic understanding of decision-making and more effective strategies.

What are the limitations of traditional methods for modeling human behavior, and how do Large Language Models overcome these?

Traditional methods, such as reinforcement learning, require extensive data collection and complex calibration. Gathering data from human subjects is time-consuming and raises ethical concerns. Large Language Models offer an alternative by using their reasoning and problem-solving abilities to generate synthetic human demonstrations, reducing the need for extensive real-world data and providing a more efficient way to simulate human behavior.

In what specific ways might robotics benefit from understanding and modeling subrational human behavior using Large Language Models?

Robotics can benefit significantly from understanding subrational human behavior because it enables the design of robots that can collaborate more effectively with humans. By anticipating human actions and understanding their limitations, robots can be designed to work alongside people in a more intuitive and helpful manner. For example, a robot that understands human risk aversion might offer a more cautious approach in a collaborative task, fostering trust and improving overall efficiency.

How could using Large Language Models to understand subrationality impact the development and implementation of public policy?

By accurately modeling subrationality, public policies can be crafted to account for how people actually behave, rather than assuming they will always act rationally. Understanding biases and cognitive limitations can help policymakers design interventions and incentives that encourage desired outcomes. For instance, knowing that individuals are prone to present bias might lead to policies that automatically enroll people in retirement savings plans, increasing participation and improving long-term financial security. Large Language Models, by simulating these behaviors, allow for policy testing and refinement before implementation.