Why Agentic Systems Need Ontologies - Frank Coyle, UC Berkeley

AI Engineer21m 18s
Frank Coyle beside the blue and white headline “GUARDRAILS FOR AGENTS” on a black background.
0 comments · 0 votesOpen discussionClose discussion
Sign in to join the discussion

    Frank Coyle introduces ontologiesAn ontology is an explicit representation of the concepts, categories and relationships used to describe a domain. as explicit representations of entities, properties and relationships, and connects them to the symbolic and neural traditions of AI. He contrasts expert-led domain modelling with bottom-up discovery and describes reusing established vocabularies instead of inventing every concept.

    Frank Coyle illustrates how RDFS and OWL relationships can support inference and constraints, then maps those ideas onto an agent tool-use loop. The model proposes a tool invocationAI tool calling lets a model request a structured operation from software instead of only returning natural-language text., application code executes it and a validator checksDeterministic software verification checks an output with a repeatable procedure that produces the same result for the same inputs and state. whether the result fits the domain before the loop proceeds or asks for help.

    Frank Coyle recommends typed input checks alongside domain-level validation, using examples such as duplicate refunds, incorrect recipient roles and unsupported order states. The talk presents an architecture for guardrailsA guardrail is a technical or procedural control that limits unsafe, disallowed, deceptive, or otherwise unwanted AI behavior., not a demonstrated guarantee that ontologies eliminate all hallucinationsA hallucination is an AI output that presents false, unsupported, or invented information as though it were reliable. or agent failures.

    Original YouTube thumbnailWatch on YouTube

    Video topic coverage

    • Agents & workflows100
    • Coding & tools75
    • Models & learning50
    • Infrastructure & compute50
    • Research & evaluation50
    • Safety & security75
    • Business & productivity25
    • Creative & media0
    • Robotics & perception0
    • Society & policy0

    Share this video

    Frank Coyle beside the blue and white headline “GUARDRAILS FOR AGENTS” on a black background. With WWW.ARTIFICIAL-INTELLIGENCE.VIDEO, 23 July 2026 and duration 21m 18s.

    Frank Coyle argues that domain ontologies and typed validation can constrain probabilistic agents by checking tool inputs, results and permitted relationships before side effects.