Rethinking Prospect Theory for LLMs: Revealing the Instability of Decision-Making under Epistemic Uncertainty
Summary: arXiv:2508.08992v3 Announce Type: replace
Abstract: Prospect Theory (PT) models human decision-making behaviour under uncertainty, among which linguistic uncertainty is commonly adopted in real-world scenarios. Although recent studies have developed some frameworks to test PT parameters for Large Language Models (LLMs), few have considered the fitness of PT itself on LLMs. Moreover, whether PT is robust under linguistic uncertainty perturbations, especially epistemic markers (e.g. “likely”), remains highly under-explored.
To address these gaps, we design a three-stage workflow based on a classic behavioural economics experimental setup. We first estimate PT parameters with economics questions and evaluate PT’s fitness with performance metrics. We then derive probability mappings for epistemic markers in the same context, and inject these mappings into the prompt to investigate the stability of PT parameters.
Key Findings
Our findings suggest that modelling LLMs’ decision-making with PT is not consistently reliable across models, and applying Prospect Theory to LLMs is likely not robust to epistemic uncertainty.
Implications
The findings caution against the deployment of PT-based frameworks in real-world applications where epistemic ambiguity is prevalent. This revelation gives valuable insights into behaviour interpretation and future alignment directions for LLM decision-making.
Research Methodology
The research was conducted in three key stages:
- Estimation of PT Parameters: We employed a series of economic questions to estimate the PT parameters, which are crucial for understanding how decision-making varies under uncertainty.
- Evaluation of PT Fitness: Performance metrics were utilized to assess the fitness of Prospect Theory within the context of LLMs.
- Probability Mappings for Epistemic Markers: We derived and injected probability mappings for epistemic markers, such as “likely,” into the prompts to see how they impacted the stability of PT parameters.
Conclusions
The exploration of Prospect Theory in the context of LLMs has unveiled critical insights regarding the limitations and potential pitfalls of applying economic decision-making models to machine learning frameworks. As we move forward, it is essential to consider the nuances of linguistic uncertainty and the implications of epistemic ambiguity in AI decision-making processes. This research not only highlights the need for more robust frameworks but also paves the way for future studies aimed at enhancing the interpretability and reliability of LLMs in real-world applications.
Future Directions
In light of these findings, future research should focus on:
- Developing alternative models that can better accommodate linguistic uncertainties within LLMs.
- Investigating the effects of different forms of uncertainty on decision-making processes in AI.
- Exploring the integration of interdisciplinary approaches from psychology, economics, and AI to improve model robustness.
The ongoing investigation into the intersection of Prospect Theory and LLMs will be crucial in guiding the next generation of AI systems toward more reliable and interpretable decision-making processes.