MOSES is a research project exploring how an AI system can discover solutions to problems by searching for programs that produce the desired behavior or result. Instead of being given a complete solution, MOSES explores possible program structures, evaluates how well they solve the problem, and uses that feedback to guide its search toward better solutions.
Within Hyperon, these solutions are represented as MeTTa programs. Depending on the problem, a discovered program might represent a logical relationship, a decision-making strategy, or a sequence of actions an intelligent agent can perform.
The current implementation focuses on specific classes of problems, including Boolean program synthesis and action-based problems. Research is gradually expanding this capability toward richer domains and, ultimately, toward allowing an AI system to discover more general MeTTa programs as solutions to the problems it encounters.