Ontologies: The Semantic Operating System for Large‑Model AI

1. Re‑examining Knowledge Engineering through Sutton’s “The Bitter Lesson”

Richard Sutton’s classic essay argues that lasting progress comes from methods that can continually leverage compute, not from manually encoding expert knowledge. The author warns against interpreting this as “knowledge is irrelevant”; instead, the problem is the high‑cost, static way knowledge is injected into systems.

2. Why Ontologies Are Needed – They Act as an AI Semantic Operating System

In a closed‑door discussion, participants likened modern ontologies to a “semantic operating system” or “semantic compiler” for AI. Enterprises suffer from a lack of a shared language across business, data and technology; ontologies supply a stable semantic base that aligns concepts, relationships, constraints, roles, events and actions, turning textual business objects into machine‑understandable units.

3. Six Core Reasons Ontologies Matter

Eliminate internal ambiguity. Different departments use divergent definitions for the same entity; a shared ontology provides a consistent semantic foundation.

Enable large models and agents to operate in business contexts. Without a common semantic layer, agents produce hallucinations and mismatches.

Static schemas cannot keep up with evolving business. Traditional data‑warehouse schemas become obsolete as processes change; semantic‑enhanced programmable graphs (SPG) can absorb dynamic changes.

Ontologies must become actionable. An “Actionable Ontology” links data, APIs, models and agents so AI can understand, locate the correct interface and execute the right action.

They become a private competitive moat. As general models improve, companies need proprietary semantic structures to avoid knowledge being fully transparent.

Missing the semantic layer now risks losing the window. Community and ecosystem momentum are required; otherwise the opportunity fades.

4. OpenKG’s Practical Insights

OpenKG demonstrates a modern knowledge infrastructure: SPG unifies data, semantics, rules and actions into an evolvable graph; Knowledge‑Augmented Generation (KAG) couples structured knowledge with LLM reasoning to improve retrieval, decomposition and logical consistency; SkillNet turns knowledge into reusable skill units; MemOS adds a memory‑operating‑system layer for long‑term learning and self‑update.

5. Two Emerging Enterprise‑Agent Deployment Paths

Current practice shows two representative routes: the “ontology‑first” path (exemplified by Palantir) builds a stable, auditable semantic skeleton before applying models, while the “model‑first” path (exemplified by Claude‑related work) centers on a base model and augments it with a Context Graph for rapid generalisation. The most competitive agents will likely fuse both approaches, using ontologies for deterministic constraints and models for creative reasoning.

6. Why Open‑Source Is the Realistic Way Forward

Because a shared semantic base is essential, open‑source collaboration is necessary. Projects such as OpenSPG, KAG‑Thinker, SkillNet and MemOS illustrate a growing stack that can be combined into an evolvable pipeline, avoiding isolated, proprietary silos. Open‑source also lowers the barrier for community‑driven improvement and ecosystem formation.

Conclusion

Large models will keep getting stronger, but enterprises need more than chatty models—they require an intelligent infrastructure that reliably understands business, executes processes, and continuously captures knowledge. Ontologies, re‑positioned as a semantic operating system, fulfill that role and must be built as code, data, standards and industry‑specific extensions.