Ask a modern AI system something it has no business answering, and it will usually answer anyway. The confidence score comes back looking crisp. The model has no idea it just wandered off the edge of everything it was trained on. For a chatbot recommending a restaurant, that is an annoyance. For a system helping route an autonomous vehicle, triage a medical scan, or flag hostile traffic on a defended network, it is something else.
Teaching machines to recognize the limits of their own competence is the question at the center of METACOG-26: Symposium on Artificial Metacognition, running November 5-7, 2026 at the Westin Arlington in Arlington, Virginia. METACOG-26 is one of nine symposia in this year's AAAI Fall Symposium Series. The Syracuse University Institute for Artificial Intelligence (SIAI) is a sponsor of the symposium, and SIAI's director, Paulo Shakarian, serves as its chair. PCI Federal Technology and Mission Solutions and Kitware are sponsoringalongside SIAI.
Metacognition is a borrowed term. Psychologists have used it for decades to describe thinking about thinking: knowing what you know, noticing when you are confused, deciding whether to slow down before committing to an answer. People do this constantly and mostly without effort. Building it into a machine has turned out to be considerably harder than describing it, which is roughly why the topic keeps pulling researchers from machine learning, cognitive psychology, systems engineering, and cognitive architecture into the same room.
The METACOG-26 program reflects that spread. Sessions take up performance prediction and critique models, metacognitive rule learning, out-of-distribution detection, neuro-symbolic reasoning, self-adaptive systems, trust calibration between people and machines, and stress testing of models that are already fielded. The applications lean toward domains where being wrong is expensive: autonomy, robotics, and cyber operations.
Shakarian is joined on the organizing committee by Nathaniel D. Bastian of DARPA, Francesco Restuccia of Northeastern University, Christian Lebiere of Carnegie Mellon University, Arslan Basharat of Kitware, and Andrea Stocco of the University of Washington. The spread across university, industry, and government labs is deliberate. Metacognitive AI is one of the areas where the people funding the work, the people building it, and the people who study human cognition all need to be in conversation, and too often they are not. The symposium grows out of a run of earlier METACOG meetings, most recently the second Workshop on Metacognitive Prediction of AI Behavior. Syracuse work has been part of that run. At METACOG-25, SIAI member Qinru Qiu, professor of electrical engineering and computer science, presented her group's research on contrastive explanation learning for reinforcement learning. The recording gives a concrete sense of what this line of work looks like in practice.
Holding it under AAAI puts that conversation in front of the right audience. The Association for the Advancement of Artificial Intelligence is the world's leading association for artificial intelligence, and its Fall Symposium Series has a particular character. AAAI describes the series as a way to bring colleagues together in an intimate forum while still providing a significant gathering point for the AI community, which in practice means small, single-track meetings rather than a sprawling conference floor
There is also a practical reason the timing matters. Metacognitive capability has moved from a research curiosity to something program managers and safety reviewers actively ask about. Anyone deploying a model into an environment that shifts under it wants to know when the system is operating outside its competence, and wants that signal before the failure rather than in the post-mortem. Much of the work presented at METACOG-26 is aimed squarely at that gap.
Registration for the 2026 AAAI Fall Symposium Series is open. Registration is handled centrally for the series rather than symposium by symposium, so attendees planning to sit in on METACOG-26 should register through that link. Early rates hold through October 2, 2026; after that date fees rise across every category, student registrations included. AAAI members pay less than nonmembers at both tiers.
