Autonomous verifiable reasoning system powered by LLMs

We would like to investigate an autonomous reasoning system that possesses: (a) advanced verifiable analytical capabilities, (b) and accuracy in processing diverse reasoning questions, and (c) grounding its reasoning on explainable formalisms. Our goal is to design an AI-powered reasoning agent to interpret vast heterogenous data (combining various types of reasoning tasks from math to logic) more efficiently. In doing this, we will provide an avenue through which LLM's reasoning could become less prone to hallucination and more reliable. Should the methodology prove effective, it could subsequently be adapted to a wide range of other domains as well. To achieve the above goal, our project aims to answer 2 questions: (RQ1) How can a language agent be designed to efficiently identify relevant evidence for a reasoning problem, and explore possible arguments for and against possible outcomes? (RQ2) How can verifiable symbolic formalisms be incorporated into the core processes of reasoning to enhance the precision, explainability, and reliability of outcomes?