Re: limitations of classification systems, fiction, lack of ontological commitment

Hi Milton, all,

I just wanted to briefly echo one point Milton raised that I think is 
central for this CG:

“We, I hope, strive to achieve a minimal set of structures and 
formalisms … for KR for AI.”
“I have used the concept of domains of discourse … either in 
mathematical or natural language form.”

 From my side, working on Knowledge3D (K3D), I’ve found exactly the 
same: you can’t get to usable KR for AI without some mathematically 
grounded structure, but you also can’t stop at pure math—domains of 
discourse, provenance, and context have to be explicit.

Concretely, in K3D we try to do this by:

  * treating domains of discourse as explicit galaxies / Houses, not a
    single global space;

  * representing form + meaning + rules together in K3D nodes (visual
    form, embeddings, RDF/OWL, and executable laws);

  * keeping the mathematical substrate (RPN programs, embeddings)
    separate from, but tightly linked to, natural language and ontology
    layers.

This has been very helpful in avoiding the “LLM ceiling” Milton 
describes: different domains can have different KR+math formalisms, but 
they still live in a shared, inspectable structure.

I agree with Milton that we should avoid dogmatic “one true hierarchy” 
and instead focus on practical, computable KR structures for specific 
kinds of AI systems—making the mathematical and NL pieces work together, 
rather than treating one as “above” or “outside” the other.

Best,
Daniel
Knowledge3D / AI‑RLWHF
https://github.com/danielcamposramos/Knowledge3D