Why Agentic Systems Need Ontologies — Frank Coyle, UC Berkeley

AI Engineer · Publicado el 23 jul 2026 · Duración 21:18

Frank Coyle explains ontologies as shared conceptualizations for agents, proposes neurosymbolic guardrails for probabilistic LLMs, and discusses validating tool outputs and catching business logic errors with ontologies. Lee 4 puntos de vista con sus evidencias y enlaces a las fuentes.

Frank Coyle

Leer los puntos clave Ver en YouTube

De un vistazo

Ontologies as formal shared conceptualizations
Frank Coyle defines an ontology as a formal specification of a shared conceptualization and says agents should be given that conceptualization of their domain.
Ver el momento de apoyo · 3:45
Neurosymbolic AI as LLM guardrails
Frank Coyle argues that neurosymbolic AI can help keep probabilistic LLMs on guardrails.
Ver el momento de apoyo · 4:46
Ontology-driven validation in agent loops
Frank Coyle proposes using ontologies after a tool runs, formatting the returned information for a validator that operates with domain ontologies.
Ver el momento de apoyo · 16:45
Ontologies for catching domain-specific errors
Frank Coyle says ontologies could catch errors such as a second refund on one order or a payout sent to support instead of the buyer. He describes catching such errors in English as tricky.
Ver el momento de apoyo · 19:12

Momentos clave4

Fragmentos breves y atribuidos, con el contexto necesario para verificarlos. La conversación completa permanece con su editor.

3:45

ontology design for AI agents

Ontologies as formal shared conceptualizations

Extracto original

It is a, a formal specification of a shared conceptualization, and that's what we wanna give to our agents. We wanna give them our concept-- our conceptualization of the universe, our universe, our domains.

Frank Coyle·23 jul 2026

Contexto

knowledge and, uh, graph technology really represents.

4:46

neurosymbolic AI architecture

Neurosymbolic AI as LLM guardrails

Extracto original

what I'd like to argue is that neurosymbolic AI sort of represents a way to keep the LLM on its guardrails, because LLMs are by nature probabilistic.

Frank Coyle·23 jul 2026

Contexto

And so

16:45

agent validation workflow

Ontology-driven validation in agent loops

Extracto original

if you look down there, a-after the, the tool is called, it said, "Tool runs," this is where ontologies could come in. The tool's gonna give us information. We put the information in a form that our, our, our, our validator can use.

Frank Coyle·23 jul 2026

19:12

ontology-based business logic enforcement

Ontologies for catching domain-specific errors

Extracto original

A second refund on the same order is a, is, is a problem. [laughs] Ontologies could catch it, whereas it's ver- it's very tricky to do that in, in English. A payout sent to the support desk instead of the buyer, okay?

Frank Coyle·23 jul 2026

Fuente y metodología

Estas perspectivas enlazan a sus fuentes originales. Las paráfrasis están identificadas y no son citas textuales.

Abrir transcripción o material de origen Reportar un problema