Large Language Models are Prone to Methodological Artifacts

SSRN Electronic Journal · 2023 · 1 citations

• Computer Science
• Natural Language Processing
• Computer Science

The Challenge of Using LLMs to Simulate Human Behavior: A Causal Inference Perspective

SSRN Electronic Journal · 2023 · 33 citations

• Computer Science
• Machine Learning
• Artificial Intelligence

The More You Ask, the Less You Get: When Additional Questions Hurt External Validity

Journal of Marketing Research · 2021 · 28 citations

• Computer Science
• Psychology
• Cognitive psychology

Researchers and practitioners in marketing, economics, and public policy often use preference elicitation tasks to forecast real-world behaviors. These tasks typically ask a series of similarly structured questions. The authors posit that every time a respondent answers an additional elicitation question, two things happen: (1) they provide information about some parameter(s) of interest, such as their time preference or the partworth for a product attribute, and (2) the respondent increasingly “adapts” to the task—that is, using task-specific decision processes specialized for this task that may or may not apply to other tasks. Importantly, adaptation comes at the cost of potential mismatch between the task-specific decision process and real-world processes that generate the target behaviors, such that asking more questions can reduce external validity. The authors used mouse and eye tracking to trace decision processes in time preference measurement and conjoint choice tasks. Respondents increasingly relied on task-specific decision processes as more questions were asked, leading to reduced external validity for both related tasks and real-world behaviors. Importantly, the external validity of measured preferences peaked after as few as seven questions in both types of tasks. When measuring preferences, less can be more.

Letting Logos Speak: Leveraging Multiview Representation Learning for Data-Driven Branding and Logo Design

Marketing Science · 2021 · 86 citations

• Computer Science
• Artificial Intelligence
• Computer Science

The authors develop a decision support system for design and branding based on a multimodal variational autoencoder that merges image, text, and ratings data.