Introducing ALMA: A Novel Cognitive Architecture for Adaptive AI Systems

Current AI systems treat all knowledge with uniform high-resolution processing — a simple factual lookup receives the same architectural treatment as deep analogical reasoning. Recent advances in adaptive computation (Mixture-of-Experts, Mixture-of-Depths, early-exit transformers) have begun addressing this inefficiency, but they optimize which tokens receive computation, not how knowledge itself is organized. In this one-hour talk, Rob Abbott presents ALMA (Abstraction-Level Membrane Architecture), a novel cognitive architecture that draws on Piagetian equilibration theory, Holland’s evolutionary computation, and complex adaptive systems to propose something fundamentally different: knowledge stratification governed by selectively permeable membranes, where abstraction level — not access control — determines how information flows between individual agents and collective cognitive spaces. The talk covers the formal framework, a stratified volume model for gradual concept migration, agent architectures with dual knowledge repositories, emergent metacognition through abstraction hierarchies, and architectural safeguards against knowledge base fragmentation. ALMA offers a theoretical foundation for AI systems that evolve their knowledge structures through accommodation rather than merely optimizing parameters within fixed representations.

When: February 25, 2026, 2-3PM

Where: Popp-Martin Student Union, Room 200

Registration: https://www.eventbrite.com/e/1982982285396