Mediating Cognitive Biases in Business Decisions Using LLMS

A large body of research has shown that cognitive biases influences human decision making. In this paper, we show that cognitive biases-despite being well understood in theory-are a widespread phenomenon in business decision making. Based on these findings, we investigate whether state-of-the-art large language models (LLMs) can help overcome the most common cognitive biases in a wide range of simplified, but realistic decision scenarios from multiple business domains. Our findings suggest that (1) the majority of surveyed corporate decision makers are frequently experiencing three or more cognitive biases, (2) most current state-of-the-art models can effectively recognize and avoid such biases while reasoning over common business decision scenarios, and (3) most state-of-the-art models outperform human test subjects in their ability to make and justify unbiased decisions. The results suggest that LLMs can effectively support humans to overcome cognitive biases in these scenarios. Our results provide an overview of current capabilities for LLM-assisted cognitive bias avoidance in business scenarios. This contrasts previous findings suggesting that earlier generation of LLMs were not effective in this area. For designers, we provide suggestions for conversational and interface design to help mediate human cognitive biases.