AgenticHypothesis: A Survey on Hypothesis Generation Using LLM Systems

Hypothesis generation is a cornerstone of scientific discovery, enabling researchers to formulate testable ideas that drive innovation and progress. Although traditional methods have proven effective, they are often constrained by cognitive biases, information overload, and time limitations. Recently, the emergence of Large Language Models (LLMs) has introduced a promising approach to automate and enhance hypothesis generation. Trained on extensive datasets, these models have demonstrated the potential to generate novel, testable hypotheses across various domains. This review critically examines state-of-the-art methodologies for LLM-based hypothesis generation, including Retrieval Augmented Generation (RAG), multi-agent frameworks, and iterative refinement techniques. Applications in biomedical research, materials science, product innovation, and interdisciplinary studies are discussed, highlighting both the versatility and the impact of these systems. Despite their promise, LLMs also present challenges such as hallucinations, data biases, and ethical concerns, which necessitate careful implementation and oversight. Future research directions include refining model architectures, integrating multimodal capabilities, and establishing robust ethical frameworks to optimize the use of LLMs in scientific research. Overall, this review provides a balanced overview of how LLMs may enhance hypothesis generation while also addressing the associated challenges. The GitHub repository containing open-source LLM-empowered hypothesis generation systems is available at https://github.com/adibgpt/AgenticHypothesis.