Neurosymbolic AI

This course explores neurosymbolic artificial intelligence (AI), an emerging approach that integrates the statistical power of modern machine learning with the rigor and transparency of symbolic reasoning. While large language models excel at generating fluent responses, they often lack reliability, consistency, and grounding in verified knowledge. These limitations make them risky for use in domains such as healthcare, emergency response, and fact-checking, where errors or omissions can carry significant consequences. Neurosymbolic AI addresses these challenges by combining data-driven learning with explicit rules, logical inference, and structured knowledge to create AI systems that are both powerful and trustworthy.

Designed for working professionals, this course equips students with the expertise to design and implement trustworthy cognitive agents. Participants will begin by exploring foundational theories of human reasoning, drawing on the work of Kahneman and Tversky on dual-process cognition and biases in judgment. Building on this foundation, the course dives into practical architectures for combining neural and symbolic components, with an emphasis on knowledge representation, logic programming, and deductive reasoning. Students will gain hands-on experience with modern frameworks, where they will learn to construct knowledge graphs, encode rules, and integrate large language models with symbolic verifiers. The course culminates in a team-based project where students develop an end-to-end neurosymbolic agent in a real-world domain, preparing them to critically assess and build next-generation AI solutions for high-consequence applications.