a bit more on World model versus ontology discussion from last weeks Ontology Summit discussion

[Gary Berg-Cross]

At the last ontology summit meeting  as part of my briefing I had a small slide on world models.  This leveraged that Gary Marcus had made about their role in making modern AI systems successful. One question was whether there was a difference between an ontology and a world model and I offered a simple observation that world models were typically more inclusive and dynamic. Some follow up remarks were made by Mike Gruninger and Deborah McGuinness about the difference between T box and A box parts of ontologies that needed to be taken into account. 

Mike had more to say about handling processes with formal representation.

There has been some follow up on the forum about the difference between knowledge bases and ontologies, but I think it's worth some comments on world models.

Some recent proposed advancements in Generative AI (GenAI) architectures emphasize the integration of a "world model" within their frameworks. Broadly this concept reflects an interest in deeper contextual understanding and situational awareness, in the hope of enabling AI systems to perform tasks more effectively by incorporating knowledge of the world they operate in. There is also now often a use in simulations, but in GenAI reinforcement learning, and generative tasks.

An example of the knowledge/awareness in such models might include geographical knowledge (say via a knowledge graph), but also social dynamics, and contextual relationships between entities.

Perhaps, just for operational reasons the way we typically think of ontologies, is as more static and therefore less a part of the operational system runninga world model. But of course the structure of the world model could be highly dependent on ontologies for the formal specification of concepts and relationships within a large world "domain".

So one difference that might be argued between them is one of purpose:

A world model is  used to simulate real-world dynamics and aid in decision-making for tasks while an ontology's purpose is to
provide a structured framework for knowledge sharing and reasoning. Typically, and historically, an ontology is more static, detailing specific relationships and definitions useful for interoperability while world models have tended to be more practical about the algorithmic dynamics, integrating numerous variables and relationships.

Gary Berg-Cross 

Potomac, MD

240-426-0770

[Alex Shkotin]

Gary,

Unfortunately, you don't provide links to either Marcus's talk or your own. This means it's impossible to quickly delve deeper, with just one click. Let me answer from a general perspective.

The question for genAI is: does it have a model of the world?

The question for IT-ontology is: does it contain a model of the world?

In both cases, the number one question is: what world are we talking about?

If it's a small world, like the ISS or Tiangong, then knowledge about it can fit in both genAI and IT-ontology.

Let W be some world. For example, it could be the world of a coffee shop for a robot waiter.

What are the criteria for IT-ontology, genAI, or robot-software to have a model of the world W?

That's the question.

Antithesis: IT-ontology can contain a model of the world.

Alex

вт, 10 мар. 2026 г. в 18:43, Gary Berg-Cross <gberg...@gmail.com>:

--
All contributions to this forum are covered by an open-source license.
For information about the wiki, the license, and how to subscribe or
unsubscribe to the forum, see http://ontologforum.org/info
---
You received this message because you are subscribed to the Google Groups "ontolog-forum" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ontolog-foru...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ontolog-forum/CAMhe4f0ifp7E%3DZErpT4sRtx099imLGjjVPRK2%2B7u6wK%3DBtbsHg%40mail.gmail.com.

[Marco Neumann]

Here is a link to the slide Gary Berg-Cross mentioned.

https://www.youtube.com/watch?v=iD9LhJ1PVwA&t=3375s

I am trying to capture the context here as a general GenAI criticism
as formulated by Gary Marcus, who argues that current LLMs lack true
"world models" - as in they do not understand physical, social, or
causal reality.

But a "world model" is not the same as talking about an ontology
exclusively, while Gary Marcus may refer primarily to logical
representations in the form of an ontology or multiple ontologies. or
more general in model theory (MT) terms, with a semantic theory of
truth and intention of providing a rigorous, logical representation of
how language relates to the world.

Janet Singer is hinting at that difference in the discussion here
https://youtu.be/iD9LhJ1PVwA?t=4613

And I think for Deborah McGuinness, GenAI "world models" are primarily
a competing theme that she fears might consume ontology solutions
space as part of knowledge graph offerings, as they are not
sufficiently probability or uncertainty-aware yet, and a strength of
GenAI solutions such as LLMs.

https://youtu.be/iD9LhJ1PVwA?t=4819

Best.
Marco

> --
> All contributions to this forum are covered by an open-source license.
> For information about the wiki, the license, and how to subscribe or
> unsubscribe to the forum, see http://ontologforum.org/info
> ---
> You received this message because you are subscribed to the Google Groups "ontolog-forum" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to ontolog-foru...@googlegroups.com.
> To view this discussion visit https://groups.google.com/d/msgid/ontolog-forum/CAMhe4f0ifp7E%3DZErpT4sRtx099imLGjjVPRK2%2B7u6wK%3DBtbsHg%40mail.gmail.com.

--


---
Marco Neumann