Mark Burgess Quantifies Promise Theory for Autonomous Agent Alignment

Mark Burgess's new paper, "Quantitative Promise Theory: Intentionality and Inference in Autonomous Agents" (arXiv:2606.08552), extends Promise Theory—a framework for describing cooperation between autonomous agents without centralized control—by integrating Bayesian probability, information-theoretic optimization, and Active Inference. He addresses classic pitfalls in probabilistic systems: non-local coordination (agents needing global consensus), calibration (difficult to set priors), and normalization (computationally expensive). By recasting boundary conditions—the constraints that define allowed states and decision thresholds—as promises, the framework aligns agents' local intentions with global behavior without explicit coordination.

The paper further shows how agent alignment can be defined scalably through promises, enabling swarms of agents to "congeal" into superagents that minimize their own information. This counterintuitive behavior emerges despite uncertainty that would normally maximize information. Burgess notes that Promise Theory supplements standard probability theory by avoiding the need for global normalizing constants and by localizing decisions. He also identifies research challenges, including stylistic preferences in modeling and computational scalability. The work spans AI, multiagent systems, neural and evolutionary computing, and statistical physics, offering a unified lens for designing robust, self-organizing systems in software, robotics, and biology.