Mind the (DH) Gap! A Contrast in Risky Choices Between Reasoning and Conversational LLMs
arXiv:2602.15173v1 Announce Type: new Abstract: The use of large language models either as decision support systems, or in agentic workflows, is rapidly transforming the digital ecosystem. However, the understanding of LLM decision-making under uncertainty remains limited. We initiate a comparative study of LLM risky choices along two dimensions: (1) prospect representation (explicit vs. experience based) and (2) decision rationale (explanation). Our study, which involves 20 frontier and open LLMs, is complemented by a matched human subjects experiment, which provides one reference point, while an expected payoff maximizing rational agent model provides another. We find that LLMs cluster into two categories: reasoning models (RMs) and conversational models (CMs). RMs tend towards rational behavior, are insensitive to the order of prospects, gain/loss framing, and explanations, and behave similarly whether prospects are explicit or presented via experience history. CMs are significantl
arXiv:2602.15173v1 Announce Type: new Abstract: The use of large language models either as decision support systems, or in agentic workflows, is rapidly transforming the digital ecosystem. However, the understanding of LLM decision-making under uncertainty remains limited. We initiate a comparative study of LLM risky choices along two dimensions: (1) prospect representation (explicit vs. experience based) and (2) decision rationale (explanation). Our study, which involves 20 frontier and open LLMs, is complemented by a matched human subjects experiment, which provides one reference point, while an expected payoff maximizing rational agent model provides another. We find that LLMs cluster into two categories: reasoning models (RMs) and conversational models (CMs). RMs tend towards rational behavior, are insensitive to the order of prospects, gain/loss framing, and explanations, and behave similarly whether prospects are explicit or presented via experience history. CMs are significantly less rational, slightly more human-like, sensitive to prospect ordering, framing, and explanation, and exhibit a large description-history gap. Paired comparisons of open LLMs suggest that a key factor differentiating RMs and CMs is training for mathematical reasoning.
Executive Summary
This article conducts a comparative study of large language models (LLMs) in decision-making under uncertainty, categorizing them into reasoning models (RMs) and conversational models (CMs). RMs tend towards rational behavior, while CMs exhibit more human-like, but less rational, tendencies. The study highlights the significance of training for mathematical reasoning as a differentiating factor between RMs and CMs. This research has important implications for the development and deployment of LLMs in decision-support systems and agentic workflows, particularly in areas where uncertainty and decision-making are critical. The findings also shed light on the potential risks and benefits of using LLMs in various applications.
Key Points
- ▸ LLMs can be categorized into reasoning models (RMs) and conversational models (CMs) based on their decision-making behavior
- ▸ RMs tend towards rational behavior, while CMs exhibit more human-like, but less rational, tendencies
- ▸ Training for mathematical reasoning is a key factor differentiating RMs and CMs
Merits
Strength in methodology
The study employs a comprehensive comparative approach, involving both LLMs and human subjects, to investigate decision-making under uncertainty.
Insight into LLM decision-making
The research provides valuable insights into the decision-making behavior of LLMs, highlighting the importance of understanding their limitations and biases.
Implications for LLM development
The findings have significant implications for the development and deployment of LLMs in various applications, particularly in areas where decision-making under uncertainty is critical.
Demerits
Limited scope of human subjects experiment
The human subjects experiment, while useful as a reference point, is relatively small in scale and may not be representative of the broader population.
Overemphasis on mathematical reasoning
The study's focus on mathematical reasoning as a differentiating factor between RMs and CMs may overlook other important factors influencing LLM decision-making.
Need for further exploration of CMs
The study's findings on CMs suggest that further research is needed to fully understand their behavior and potential applications.
Expert Commentary
The study provides a valuable contribution to the understanding of LLM decision-making under uncertainty, highlighting the importance of considering the potential risks and limitations of these models. The findings suggest that RMs and CMs exhibit distinct decision-making behaviors, with RMs tending towards rational behavior and CMs exhibiting more human-like tendencies. The study's emphasis on the significance of training for mathematical reasoning as a differentiating factor between RMs and CMs is particularly noteworthy. However, the study's limitations, including the relatively small scale of the human subjects experiment and the potential overemphasis on mathematical reasoning, should be taken into consideration. Overall, the study's findings have significant implications for the development and deployment of LLMs, and highlight the need for further research in this area.
Recommendations
- ✓ Further research is needed to fully understand the behavior and potential applications of CMs.
- ✓ Developers and deployers of LLMs should prioritize the evaluation and selection of models based on their decision-making behavior and potential applications.