Professor Ricky Sethi Discusses Artificial Metacognition and AI Safety on Podcast
Ricky Sethi, adjunct teaching professor in Graduate and Professional Studies, was featured in a two-part episode of AI and You, a podcast hosted by Peter Scott. In Episodes 301 and 302, which were released March 23 and March 30, 2026, Sethi discussed his research on artificial metacognition, the effort to give generative artificial intelligence systems the ability to monitor and regulate their own reasoning processes. The conversation covered how his framework, built around a five-dimensional metacognitive state vector (MSV), enables ensembles of large language models to recognize when they are uncertain, detect contradictory information, and decide when a problem requires more careful deliberation rather than a quick answer.
The discussion ranged from the neuroscience of human thinking to practical implications for AI safety, including teacher-student distillation and how structured multi-agent augmentation can serve as a check against AI hallucination and deception. Sethi explained how his multidisciplinary background in neurobiology, physics, and artificial intelligence informs the framework, which draws on cognitive psychology’s dual-process theory, organizational team science, and dialectical philosophy.
He and his collaborators have published the foundational theory at the Association for Computing Machinery (ACM) Conference on Information and Knowledge Management and have prepared a working implementation for the ACM Web Conference 2026, with further experimental validation in preparation for the Conference on Neural Information Processing Systems 2026.
In addition to his role as an adjunct teaching professor at WPI, Sethi directs research on metacognitive frameworks for generative AI systems and is a professor of computer science at Fitchburg State University. He is the author of Essential Computational Thinking: Computer Science from Scratch. A research page, demo, and full source code are available here.
Listen to the episodes:
