A workflow for human-centered machine-assisted hypothesis generation: Commentary on Banker et al. (2024)

Area of Science:

• Artificial Intelligence
• Social Psychology
• Scientific Methodology

Background:

• Large language models (LLMs) show promise for revolutionizing scientific hypothesis generation.
• Previous research explored using GPT-3 and GPT-4 for generating hypotheses in social psychology.
• Existing approaches may overlook critical human and AI limitations.

Purpose of the Study:

• To propose a human-centered workflow for hypothesis generation using LLMs.
• To address limitations in current AI-driven hypothesis generation methods.
• To integrate researcher's internal and external contexts into the process.

Main Methods:

• Development of a human-centered workflow for hypothesis generation.
• Iterative engagement process between researchers and GPT-4.
• Focus on augmenting, not replacing, the researcher's role.

Main Results:

• The proposed workflow acknowledges and leverages the capabilities of both humans and LLMs.
• It incorporates the researcher's values, goals, and existing literature.
• This approach enhances the hypothesis generation process by combining human insight with AI power.

Conclusions:

• A human-centered workflow is crucial for effective LLM-assisted hypothesis generation.
• Iterative collaboration between researchers and AI optimizes the use of LLMs in science.
• This methodology respects and enhances the unique contributions of human researchers.