Shopping by algorithm: How agentic AI deploys human heuristics as a surrogate consumer

Problem

This work addresses a gap in understanding how marketing pricing cues influence the decision-making processes of AI shopping agents. The authors highlight the lack of empirical studies examining the interaction between human heuristics and AI behavior in commercial shopping contexts. This is particularly relevant given the increasing deployment of AI in consumer environments. The paper is a preprint and has not undergone peer review.

Method

The authors introduce a novel framework called Tool-Lab, which adapts the information-board process tracing methodology to analyze AI decision-making with costly tool calls for product attributes. The study utilizes eight commercially deployed large language models (LLMs) from three different providers. Two types of goal prompts are employed: a vague goal prompt and a specific goal prompt, allowing for a comparative analysis of how these prompts influence the AI's choice behavior.

Results

The findings indicate that Diagnostic Attribute Omission occurs when the AI operates under a vague goal prompt, leading to suboptimal decision-making due to the acquisition costs associated with product attributes. In contrast, when provided with a specific goal prompt, the AI's performance improves significantly. Furthermore, the study reveals that the Choice Optimality of the AI's decisions under vague prompts closely resembles human heuristics, suggesting that the AI is susceptible to the same cognitive biases that affect human consumers.

Limitations

The authors note several limitations in their study. Firstly, the AI's vulnerability to marketing heuristics is pronounced when vague goal prompts are used, which may lead to less optimal consumer choices. Additionally, the effectiveness of the AI is heavily dependent on the architecture of the storefront information, which can vary significantly across different platforms. These factors may limit the generalizability of the findings to other contexts or AI systems.

Why it matters

This research has significant implications for the design and deployment of AI shopping agents. Understanding how marketing cues can manipulate AI decision-making processes is crucial for developing more robust and consumer-friendly AI systems. The insights gained from this study could inform future work on improving AI interpretability and decision-making strategies, ultimately leading to better alignment with consumer interests and enhanced shopping experiences.