MATHBAC: Mathematics of Boosting Agentic Communication

Mathematical understanding is essential to discovery
 

MATHBAC aims to enable the next generation of agentic artificial intelligence platforms that systematically improve/evolve to transform the nature of scientific discovery for defense. 

Despite their impressive accomplishments to date, and the promise they hold, AI agents and agent collectives today are tasked to use human language and an often ad hoc, trial-and-error deployment processes (i.e. an “Edisonian” approach to design) and can lead to inefficient and non-generalizable methods.  

If we view AI agents as input-output mathematical operators, their interactions are no longer just syntactic message strings, but components of a formal system of communication – the key that could unlock a true science of agentic AI. 

MATHBAC will leverage mathematics, systems theory, and information theory to create common protocols and generalizable principles to streamline how AI agents communicate with each other in problem solving and science discovery tasks. 

• Common protocols will enable AI agents to create mathematically optimal mechanisms for collaborating and sharing information; they will also provide a framework to allow agentic AI systems to evolve in an explainable way. This can be considered analogous to how organizations constantly develop new operating structures and styles of communication to effectively solve problems, particularly within multidisciplinary areas of interest.  
• Generalizable principles will focus on the content of agentic communication, extracted by disentangling complex data streams to discover novel laws or correlations applicable towards problem solving. This can be considered analogous to the development and sharing of common core knowledge between humans cooperating on a task, and the ways we adjust it towards joint success.   

By deriving and leveraging new mathematical frameworks for AI agentic systems communication, MATHBAC will enable agentic platforms to not only solve complex problems and discover new science, but also to understand the fundamentals behind the functioning, successes, and failures of the platforms themselves.