Abstract
Hypothesis generation and testing is a key part of experimental research. Generating hypotheses and designing experiments to test them requires deep experience in the subject matter. While large language models (LLM) can be used to generate hypotheses and can enable applying knowledge from seemingly unrelated domains to new domains, LLMs may generate non-specific hypotheses that are not actionable or useful for predictions, and that are not associated with a validation mechanism. This disclosure describes techniques that use a LLM to generate hypotheses along with corresponding test criteria to ensure hypotheses are inherently grounded in data and amenable to evaluation. The generated hypotheses are testable, specific, based on prior knowledge, and falsifiable. The techniques can be used to build hypotheses based on past experimental data. The techniques can be used for objective A/B testing and in other research/ experimental contexts.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Haji-Abolhassani, Iman and Yefremov, Dmitriy, "Verifiable Hypothesis Generation, Testing, and Ranking Using Generative AI", Technical Disclosure Commons, (May 30, 2025)
https://www.tdcommons.org/dpubs_series/8173