Machine-assisted social psychology hypothesis generation.

Area of Science:

• Social Psychology
• Artificial Intelligence
• Computational Social Science

Background:

• Synthesizing vast amounts of social psychology research is increasingly challenging for researchers.
• Overlooked connections between research topics may arise due to information overload.

Purpose of the Study:

• To explore the use of large language models (LLMs) for generating novel social psychology research hypotheses.
• To assess the quality of LLM-generated hypotheses compared to human-generated ones.

Main Methods:

• Fine-tuning GPT-3 on extensive social psychology abstracts and preprint data.
• Generating hypotheses using GPT-4 without fine-tuning.
• Expert evaluation of model- and human-generated hypotheses on key dimensions.

Main Results:

• Social psychology experts rated GPT-3 and human-generated hypotheses similarly on clarity, originality, and impact.
• Social psychology experts rated GPT-4 generated hypotheses higher than human-generated ones across multiple quality dimensions.

Conclusions:

• LLMs show promise in aiding social psychology researchers in hypothesis generation.
• GPT-4 demonstrates a strong capability in producing high-quality, original, and relevant research hypotheses.